High-precision forecasting method for fuel gas demand quantity of continuous annealing furnace

Abstract

The invention relates to the technical field of cold rolling annealing, and particularly discloses a high-precision forecasting method for fuel gas demand quantity of a continuous annealing furnace, which comprises the following steps: S1, acquiring strip steel production plan data and real-time operation data; s2, a dynamic correlation model is constructed, self-learning correction is achieved, the system constructs a fuel gas demand prediction framework, and the formula of the framework is S3, a self-learning mechanism is triggered, and parameter dynamic correction is achieved; s4, outputting a high-precision gas demand prediction result, and generating a gas demand curve in the next 24 hours based on the dynamically optimized model parameters; according to the method, the limitation of static empirical coefficients is broken through, high-precision self-adaptive prediction of the fuel gas demand is realized, the fuel gas demand prediction precision and the working condition adaptive capacity are remarkably improved, and technical support is provided for energy dynamic balance and safe production of iron and steel enterprises.

Description

Technical Field

[0001] The invention relates to the technical field of cold rolling annealing, and in particular to a method for high-precision forecasting of gas demand for a continuous annealing furnace. Background Art

[0002] The vertical full-radiant tube annealing process is widely used in cold rolling mills and continuous hot-dip galvanizing lines. Refined coke oven gas, natural gas, and other fuels are typically burned in the radiant tubes to indirectly heat the steel strip within the annealing furnace. The gas and air are combusted in the radiant tube burners, generating high-temperature flue gas. This high-temperature flue gas transfers heat to the tube walls through convection and radiation, raising the temperature to approximately 900°C. The hot tube walls then radiate the heat to the steel strip and the furnace. After cooling through the radiant tube's air preheater and superheated water heat exchanger, the flue gas is forcibly discharged into the chimney by the exhaust fan.

[0003] Gas consumption is mainly affected by factors such as gas calorific value, strip heating temperature, strip steel type, strip thickness, strip width, strip thermal conductivity, and strip running speed. It is also affected by multiple factors such as flue gas emission temperature and furnace insulation capacity.

[0004] Based on the known strip production plan (including basic information such as strip steel grade, specification, weight, target operating speed, etc.), gas consumption in the future can be predicted. This can not only provide a decision-making basis for the scheduling and adjustment of gas supply for steel joint enterprises, but also reasonably allocate gas supply to each region in advance, reduce resource waste, and ensure the safe and stable operation of gas storage tanks; it can also guide the production site to optimize control parameters to avoid risks such as environmental protection violations and strip overheating and warping caused by large fluctuations in gas consumption, thereby bringing considerable safety, environmental protection and economic benefits.

[0005] Patent application publication number CN 109055711 A describes a method for determining the energy consumption per ton of steel in the furnace section of a continuous annealing unit. By studying an energy model between process parameters and energy consumption, this method constructs a furnace section energy consumption model per unit time to determine Q (Q represents the total heat revenue of the furnace section per unit time). This method is used to calculate and evaluate energy consumption per ton of steel, providing a theoretical basis for optimizing the furnace section of the continuous annealing unit, saving production capacity, and reducing costs. This patent only measures the furnace section energy consumption per ton of steel through a model and does not address the problem of predicting gas consumption based on existing strip production plans.

[0006] Patent publication number CN119846957A, full-factor driven fusion mechanism tempering process steel temperature prediction model and its construction and application. The patent proposes a steel temperature prediction method that combines mechanism model and data-driven method. By establishing a target function with total heat absorption rate as the core, combining the improved GSK algorithm to optimize the model parameters, the interpretability and extrapolation of the steel temperature prediction are improved. Although this method combines physical mechanism and data-driven method, it is mainly used for steel temperature prediction in tempering furnace, and does not involve dynamic prediction of gas consumption; its parameter optimization depends on offline training, lacks real-time self-learning mechanism, and cannot adapt to dynamic changes of production conditions.

[0007] The root cause of the above problems lies in the fact that the traditional technology has not effectively combined the advantages of physical mechanism and data-driven technology: static physical models lack dynamic parameter optimization capability, and pure data-driven methods are prone to "black box" dilemma and rely on a large amount of labeled data. Therefore, the industry urgently needs to design a high-precision gas demand prediction method based on a physical model framework embedded with a data-driven self-learning mechanism to solve the problem of limited precision caused by parameter fixation, insufficient modeling of multi-factor coupling effects, and dynamic response lag in traditional gas prediction models. SUMMARY

[0008] In view of the problems existing in the prior art, the purpose of the present application is to provide a method for high-precision prediction of gas demand of a continuous annealing furnace, which embeds real-time self-learning and historical rule mining mechanism on the basis of retaining the physical model framework, realizes high-precision self-adaptive prediction of gas demand by dynamically correcting model parameters and quantifying complex coupling relationships.

[0009] The technical solution adopted by the present application to solve its technical problems is: a method for high-precision prediction of gas demand of a continuous annealing furnace, comprising the following steps:

[0010] S1, obtaining strip steel production plan data and real-time running data;

[0011] S2, constructing a dynamic correlation model and realizing self-learning correction, the system constructs a gas demand prediction framework, and its formula is:

[0012] S3, triggering the self-learning mechanism and dynamic correction of parameters;

[0013] S4, outputting high-precision gas demand prediction results, generating a future 24-hour gas demand curve based on the dynamically optimized model parameters.

[0014] Specifically, the basic information of the production plan data in step S1 includes but is not limited to the steel grade G, thickness H, width B, target annealing temperature T, and target running speed V of the strip steel.

[0015] Specifically, the real-time operation data in step S1 include but are not limited to the collected gas consumption, actual running speed of the strip, ambient temperature T env , flue gas temperature T flue , shielding gas injection volume q prot ;

[0016] Model initialization: Set initial parameters based on thermodynamic principles, including but not limited to specific heat capacity c(G), combustion efficiency η combustion , heat loss coefficient γ, as the benchmark for self-learning optimization.

[0017] Specifically, the steps of establishing the dynamic association model and self-learning correction in step S2 are:

[0018] S21. Basic heat demand calculation: Calculate the strip temperature rise requirement ΔT = T out -T in , T out is the output strip temperature, T in The default ambient temperature is T env The basic heat load is quantified by combining the strip mass M = ρ·B·H·L, where ρ is the strip density and L is the strip length;

[0019] S22, dynamic optimization of comprehensive efficiency correction factor: Introducing the influence of steel grade thermal conductivity k(G), width-to-thickness ratio H / B and running speed V, dynamically adjust self-learning coefficients α1, α2, and β through real-time data feedback to optimize heat transfer efficiency;

[0020] S23. Heat loss and shielding gas heat loss compensation: Quantify the change of furnace insulation efficiency with ambient temperature, and compensate for heat loss by calculating the shielding gas flow rate and temperature rise. Combined with historical data, optimize the heat loss coefficient γ and shielding gas specific heat capacity parameters.

[0021] Specifically, in the formula in step S2: M is the mass of the strip steel, f eff is the comprehensive efficiency correction factor, Q loss and Q prot are heat loss and shielding gas heat loss compensation respectively, and ηcombustion is the combustion efficiency.

[0022] Specifically, the steps of triggering the self-learning mechanism and dynamic parameter correction in step S3 are:

[0023] S31, residual threshold trigger: real-time calculation of predicted gas volume Q gas When the residual error with the actual consumption exceeds 3%, the gradient descent algorithm is started to optimize key parameters such as combustion efficiency η combustion ;

[0024] S32, historical data transfer learning: build a steel grade model library, when a new steel grade is put into production, call the historical parameters of the similar steel grade to initialize the model, and combine online learning to converge the error to within 2% within 72 hours;

[0025] S33, long-term regular preloading: using LSTM network to mine seasonal periodicity characteristics, preloading environmental temperature compensation coefficient to the model.

[0026] The present application has the following beneficial effects:

[0027] The continuous annealing furnace gas demand high-precision prediction method designed by the present application, by fusing thermodynamic physical mechanism and data-driven technology, constructs a dynamic correlation model based on steel grade, strip steel specification, process parameters and environmental variables, realizes model parameter self-learning optimization combined with real-time data feedback and historical big data training, breaks through the limitations of static empirical coefficients, realizes high-precision adaptive prediction of gas demand, significantly improves the gas demand prediction accuracy and working condition adaptability, and provides technical support for energy dynamic balance and safety production of steel enterprises. BRIEF DESCRIPTION OF DRAWINGS

[0028] Figure 1 is a flow chart of the continuous annealing furnace gas demand high-precision prediction method. DETAILED DESCRIPTION

[0029] The technical solutions in the embodiments of the present application will be further clearly, completely and in detail explained in combination with the drawings in the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0030] As shown in Figure 1 , a continuous annealing furnace gas demand high-precision prediction method comprises the following steps:

[0031] 1, obtaining strip steel production plan data and real-time running data.

[0032] The basic information of the production plan data includes but is not limited to the steel grade G, thickness H, width B, target annealing temperature T and target running speed V of the strip steel.

[0033] The real-time running data includes but is not limited to collecting gas consumption, actual running speed of strip steel, environment temperature T env , flue gas temperature T flue , and protection gas injection amount q prot .

[0034] Model initialization: based on the principle of thermodynamics, set the initial parameters, including but not limited to specific heat capacity c(G), combustion efficiency η combustion, heat loss coefficient γ as the benchmark of self-learning optimization.

[0035] 2, build a dynamic correlation model and realize self-learning correction.

[0036] 1) Basic heat demand calculation: calculate the strip temperature rise demand ΔT = T out -T in , T out is the temperature of the output strip, T in , the value of the ambient temperature T env by default; and combined with the strip mass M = ρ·B·H·L quantification of basic heat load, ρ is the strip density, L is the strip length.

[0037] 2) Dynamic optimization of comprehensive efficiency correction factor: introduce the influence of steel thermal conductivity k(G), width-thickness ratio H / B and running speed V, dynamically adjust self-learning coefficients α1, α2, β through real-time data feedback, and optimize heat transfer efficiency.

[0038] 3) Heat loss and protective gas heat loss compensation: quantize the change rule of furnace insulation efficiency with ambient temperature, and compensate the heat loss through protective gas flow and temperature rise calculation, and optimize the heat loss coefficient γ and the specific heat capacity parameters of the protective gas combined with historical data.

[0039] The system constructs a gas demand prediction framework, and the formula is: In the formula: M is the strip mass, f eff is the comprehensive efficiency correction factor, Q loss and Q prot are heat loss and protective gas heat loss compensation items, and ηcombustion is the combustion efficiency.

[0040] 3, trigger the self-learning mechanism and dynamic correction of parameters.

[0041] 1) Residual threshold triggering: real-time calculation of the residual of the predicted gas quantity Q gas and the actual consumption, when the error exceeds 3%, start the gradient descent algorithm to optimize the key parameters such as combustion efficiency η combustion .

[0042] 2) Historical data transfer learning: build a sub-model library for each steel grade, when a new steel grade is put into production, call the historical parameters of similar steel grades to initialize the model, and combine online learning to converge the error to within 2% within 72 hours.

[0043] 3) Long-term rule preloading: use LSTM network to mine seasonal periodicity characteristics, preload the ambient temperature compensation coefficient to the model, and reduce the prediction deviation under extreme weather conditions.

[0044] 4. Output high-precision gas demand prediction results, generate future 24-hour gas demand curve based on dynamically optimized model parameters, provide dynamic scheduling recommendations to guide energy balance and production plan adjustment.

[0045] Embodiments of the present application: existing gas prediction methods rely on physical models of static parameters, which cannot dynamically adapt to changes in equipment state, fluctuations in raw materials and complex working conditions, resulting in high prediction errors. The present application proposes a method that combines physical mechanisms and data-driven self-learning, dynamically optimizes model parameters through real-time data feedback and historical rule mining, realizes high-precision gas demand prediction, and provides decision support for coal gas dynamic scheduling. The gas demand prediction system of the embodiment of the present application includes a data acquisition layer, a dynamic model layer, a self-learning correction layer, and a decision output layer, and the specific implementation process is as follows.

[0046] The system acquires two types of key data in real time through industrial communication protocols: one is strip production plan data, including steel grade, thickness, width, target annealing temperature and running speed; the other is real-time running data, including gas consumption, environmental temperature, flue gas temperature, and protection gas injection amount. Based on the principle of thermodynamics, the model parameters (such as specific heat capacity and combustion efficiency) are initialized, and the system constructs a gas demand prediction framework, the formula of which is:

[0047] In the formula, the steel grade is A, M is the mass of the strip, f eff is the comprehensive efficiency correction factor, Q loss and Q prot are the heat loss and protection gas heat loss compensation terms, and ηcombustion is the combustion efficiency.

[0048] All parameter initial values are based on theoretical calculations and are dynamically optimized through the following self-learning mechanism.

[0049] Real-time residual feedback optimization is the core of model dynamic adaptation. The system calculates the residual error between the predicted value and the actual gas consumption in real time, and when the error exceeds the preset threshold (such as 3%), the parameter correction program is automatically triggered. In production, it is found that the combustion efficiency decreases due to the accumulation of carbon in the burner, and the model uses the gradient descent algorithm to propagate the error in reverse, and η combustion is gradually corrected from the initial value of 0.88 to 0.895, and within 24 hours, the error is converged to within 1%. This process is fully automated and does not require human intervention, significantly improving the response speed to fluctuations in equipment state.

[0050] The historical data migration learning solves the problem of rapid adaptation of new steel grades. The system constructs a sub-model library for each steel grade to store the historical optimization parameters of different steel grades. When a new steel grade A is produced for the first time, the model calls the parameter set of a similar steel grade B for initialization, and combines online learning with real-time data. The prediction error is converged from the initial 4.2% to 1.5% within 72 hours, which is 80% shorter than the traditional manual debugging period. This mechanism is particularly suitable for multi-variety and small-batch production scenarios, and supports enterprises to quickly seize high-value-added markets.

[0051] In the output layer, the system generates a future 24-hour gas demand curve and provides dynamic scheduling recommendations accordingly. When a gas demand peak occurs, the system starts the standby gas tank to store pressure in advance, reducing the emission rate. At the same time, process parameters can be optimized, and the model dynamically adjusts the annealing temperature setting based on real-time prediction values, reducing gas consumption while ensuring the mechanical properties of the strip.

[0052] The present application establishes a correlation measurement model of gas consumption and influencing factors such as gas calorific value, strip heating temperature, strip grade, strip thickness, strip width, strip thermal conductivity, strip running speed, flue gas emission temperature, and furnace insulation capacity. In actual production, the measured gas consumption and production coil working condition are self-learned and corrected, continuously improving the model's measurement accuracy, and solving the precision limitation problem caused by the traditional prediction method's parameter fixation, dynamic response lag, and insufficient modeling of complex working conditions. By accurately predicting gas demand, the system can further guide the optimization of coal gas resources, supporting energy balance and production plan adjustment. The model prediction error is reduced from the traditional physical model's average of 5.2% to 0.8%, with an accuracy improvement of 84%. The annual gas cost savings are about 15 million yuan. In terms of quality, annealing temperature fluctuations are reduced by 16%, and automobile panel surface defect rates are reduced by 5%, directly driving a 1.8 percentage point increase in yield. At the same time, the process debugging period of new steel grades is compressed from 20 days in traditional methods to 7 days. The product quality stability of high-end cold-rolled automobile panels and household appliance panels in the Shandong base is improved, supporting Shandong Iron and Steel Group to seize high-value-added markets.

[0053] The present application is not limited to the above embodiments, and any structural changes made under the inspiration of the present application should be known. Any technical solution with the same or similar technical solutions as the present application falls within the scope of the present application.

[0054] The technical, shape, and structure parts not described in detail in the present application are well-known technologies.

Claims

1. A method for high-precision forecasting of gas demand for a continuous annealing furnace, characterized in that: The following steps are involved: S1. Obtaining strip steel production plan data and real-time operation data; S2. Build a dynamic correlation model and implement self-learning correction. The system builds a gas demand prediction framework. The formula is: S3, triggering the self-learning mechanism and dynamic parameter correction; S4. Output high-precision gas demand forecast results and generate the gas demand curve for the next 24 hours based on the dynamically optimized model parameters.

2. The method for high-precision forecasting of continuous annealing furnace gas demand according to claim 1, characterized in that: The basic information of the production plan data in step S1 includes but is not limited to the steel type G, thickness H, width B, target annealing temperature T, and target running speed V of the strip.

3. The method for high-precision forecasting of continuous annealing furnace gas demand according to claim 2, characterized in that: The real-time operation data in step S1 include but are not limited to the collected gas consumption, actual running speed of the strip, ambient temperature T env , flue gas temperature T flue , shielding gas injection volume q prot ; Model initialization: Set initial parameters based on thermodynamic principles, including but not limited to specific heat capacity c(G), combustion efficiency η combustion , heat loss coefficient γ, as the benchmark for self-learning optimization.

4. The method for high-precision forecasting of continuous annealing furnace gas demand according to claim 3, characterized in that: The steps of establishing the dynamic association model and self-learning correction in step S2 are: S21. Basic heat demand calculation: Calculate the strip temperature rise requirement ΔT = T out -T in , T out is the output strip temperature, T in The default ambient temperature is T env The basic heat load is quantified by combining the strip mass M = ρ·B·H·L, where ρ is the strip density and L is the strip length; S22, dynamic optimization of comprehensive efficiency correction factor: Introducing the influence of steel grade thermal conductivity k(G), width-to-thickness ratio H / B and running speed V, dynamically adjust self-learning coefficients α1, α2, and β through real-time data feedback to optimize heat transfer efficiency; S23. Heat loss and shielding gas heat loss compensation: Quantify the change of furnace insulation efficiency with ambient temperature, and compensate for heat loss by calculating the shielding gas flow rate and temperature rise. Combined with historical data, optimize the heat loss coefficient γ and shielding gas specific heat capacity parameters.

5. The method for high-precision forecasting of continuous annealing furnace gas demand according to claim 1, characterized in that: In the formula in step S2: M is the mass of the strip steel, f eff is the comprehensive efficiency correction factor, Q loss and Q prot are heat loss and shielding gas heat loss compensation respectively, and ηcombustion is the combustion efficiency.

6. The method for high-precision forecasting of continuous annealing furnace gas demand according to claim 5, characterized in that: The steps of triggering the self-learning mechanism and dynamic parameter correction in step S3: S31, residual threshold trigger: real-time calculation of predicted gas volume Q gas When the residual error with the actual consumption exceeds 3%, the gradient descent algorithm is started to optimize key parameters such as combustion efficiency η combustion ; S32, Historical Data Transfer Learning: Build a steel seed model library. When a new steel grade is put into production, call the historical parameters of similar steel grades to initialize the model. Combined with online learning, the error is converged to within 2% within 72 hours. S33. Long-term regularity preloading: Use LSTM network to mine seasonal periodic characteristics and preload ambient temperature compensation coefficients into the model.

Images