Energy-saving control system and control method for dynamic proportioning optimization of multiple gas sources of a heating furnace

Abstract

The present application belongs to the technical field of intelligent control of industrial kiln and metallurgical energy saving and emission reduction, and provides an energy-saving control system and control method for dynamic proportioning optimization of multi-gas sources of a heating furnace, which comprises the following steps: firstly, cross-frequency alignment and feature fusion are performed on the collected high-frequency and low-frequency thermal and energy consumption data; in the multi-gas source mixed combustion condition, the traditional equal heat value static replacement principle is broken, a pre-trained machine learning engine is used for real-time multi-objective optimization, and a dynamic coal gas proportioning instruction that minimizes the predicted unit consumption is output; at the same time, feature sensitivity analysis is performed based on the Pearson correlation coefficient, and segmented targeted oxygen-enriched control is implemented to accurately dispatch the oxygen-enriched resources to the high-sensitivity heating section; in addition, the oxygen-enriched combustion is used to actively compress the high-temperature zone residence time, and the descaling water pressure is self-adaptively adjusted in linkage with the downstream hot rolling system. The present application effectively overcomes the hysteresis and oxygen-enriched blindness of the traditional regulation, significantly reduces the comprehensive unit consumption of the whole system, and initiates the cross-process quality closed loop, thereby greatly reducing the defect rate of the iron oxide scale on the surface of the billet.

Description

Technical Field

[0001] This invention belongs to the field of intelligent control of industrial kilns and energy conservation and emission reduction in metallurgy. It relates to the coordinated control of dynamic optimization of gas ratio, calorific value compensation and billet heating quality in hot rolling heating furnaces under multi-gas source co-firing and oxygen-enriched combustion conditions. Specifically, it is an energy-saving control system and control method for dynamic optimization of multi-gas source ratio in heating furnaces. Background Technology

[0002] In iron and steel metallurgical enterprises, hot rolling heating furnaces are core energy-consuming equipment. With the advancement of energy conservation and emission reduction, modern heating furnaces usually adopt a multi-source co-firing mode (such as mixing coke oven gas, blast furnace gas, converter gas and natural gas) to make full use of surplus gas. Existing heating furnace combustion control mostly relies on manual experience or conventional PID feedback control, that is, the operator manually or semi-automatically adjusts the valve opening of gas and combustion air according to the set target furnace temperature or the current temperature deviation.

[0003] However, in actual production conditions, existing control methods have the following significant drawbacks:

[0004] 1. Control lag and energy efficiency disconnect under complex gas source co-firing;

[0005] When the supply of a gas source fluctuates in real time (e.g., a sudden reduction in the supply of high-calorific-value coke oven gas, leading to a passive increase in the replacement ratio of converter gas), traditional manual regulation often only allows for simple gas volume replacement based on "equivalent calorific value." However, the combustion characteristics (such as flame length, emissivity, and radiative heat transfer rate) of different gas components vary greatly. This static and lagging feedback regulation cannot handle the nonlinear coupling relationship between various gas ratios and multiple furnace temperatures, often resulting in abnormal disconnects such as "the nominal calorific value of the mixed gas meets the standard, but the combustion efficiency decreases and the actual consumption per ton of steel increases dramatically."

[0006] 2. The overall blind spot and waste of resources in oxygen-enriched combustion

[0007] While some advanced heating furnaces have introduced oxygen-enriched combustion technology to improve the utilization rate of low-calorific-value gas, existing oxygen enrichment methods are mostly global diffuse oxygenation or uniform oxygen distribution at a fixed ratio. Current technologies fail to identify and quantify the different sensitivities of each heating section (such as the preheating section, primary heating section, and soaking section) to oxygen flow and temperature increases. This results in a significant waste of oxygen in the preheating section, where the correlation is extremely low, while the primary heating section, which truly needs temperature increases, cannot receive targeted, precise, and high-intensity oxygen enrichment compensation. Consequently, the overall energy-saving effect of the entire system is greatly reduced.

[0008] 3. Isolated disconnect between thermal parameters and downstream surface quality

[0009] The existing thermal control system of heating furnaces and the surface quality of downstream products (especially the initial defect assessment in the cold rolling process) are often two isolated information systems. In traditional processes, to ensure thorough heating, the billet is often held in the high-temperature zone of the furnace for an extended period, leading to excessive thickening of the original iron oxide scale (iron scale) on the stainless steel surface. Existing combustion control systems lack the technical means to establish a closed-loop linkage between "oxygen-enriched heat intensity," "furnace holding time," and "downstream descaling and iron scale indentation defects," resulting in persistently high levels of surface quality defects (such as iron scale indentations and rough bands).

[0010] In summary, the industry urgently needs a real-time automatic control method that can overcome the limitations of human experience, predict unit consumption trends in a forward-looking manner, and synergistically optimize gas ratio, targeted oxygen enrichment, and billet heating quality. Summary of the Invention

[0011] The purpose of this invention is to provide an energy-saving control system and method for dynamic ratio optimization of multiple gas sources in a heating furnace. Addressing the shortcomings of existing technologies, such as lag in gas ratio adjustment, waste of oxygen-enriched resources, and disconnect between thermal control and surface quality, this invention achieves dynamic thermal energy optimization under complex gas source co-firing conditions. It eliminates oxygen waste and overcomes the temperature unevenness and resource waste caused by existing globally diffused oxygen-enriched combustion. Furthermore, it establishes a cross-process closed loop of "furnace dwell time - surface quality," realizing intelligent control from "lagging feedback" to "feedforward guidance."

[0012] The technical solution adopted by the present invention to achieve the above objectives is as follows:

[0013] An energy-saving control system for dynamic optimization of multi-gas source ratio in a heating furnace includes a data acquisition module, a machine learning prediction and optimization engine, a collaborative execution and adaptive closed-loop module, and a downstream quality feedback module. The data acquisition module collects real-time data through sensors. The machine learning prediction and optimization engine uses a random forest regression algorithm to establish a nonlinear coupled prediction model for multi-gas source ratio, segmented oxygen enrichment, and multi-segment furnace temperature. The collaborative execution and adaptive closed-loop module includes a gas regulating valve and an oxygen regulating valve. The downstream quality feedback module is a high-pressure descaling system.

[0014] The control method of the energy-saving control system for dynamic ratio optimization of multiple gas sources in the heating furnace includes the following steps:

[0015] S1. Multi-source data cross-frequency alignment and perception: Collect high-frequency thermal parameters and low-frequency energy consumption indicators of the heating furnace, extract features and perform cross-frequency alignment through time window sliding algorithm, and construct a standard training sample set;

[0016] S2, Dynamic Gas Ratio Reconstruction and Unit Consumption Optimization: Real-time collected operating data is input into a pre-trained machine learning prediction and optimization engine. When fluctuations in the supply of high-calorific-value gas are detected, the machine learning prediction and optimization engine performs multi-objective simulation calculations in the background, outputs a dynamic conditioning ratio instruction that minimizes the predicted unit consumption, and sends it to the gas / oxygen regulating valve of the collaborative execution and adaptive closed-loop module.

[0017] S3. Based on characteristic sensitivity, target oxygen enrichment control calculates the correlation coefficient between the temperature and oxygen enrichment flow rate of each heating zone and the total furnace consumption. Based on this, the oxygen enrichment compensation priority zone is divided, and the oxygen enrichment flow rate is concentrated and allocated to the high-sensitivity zone. S4. Cross-process surface quality closed-loop control calculates the heat exchange efficiency increment based on the oxygen enrichment enhanced combustion state in step S3, actively outputs the high-temperature zone furnace dwell time compression command, and feeds forward the theoretical iron oxide scale state change to the high-pressure descaling system of the downstream quality feedback module, and adjusts the water pressure and injection parameters of the high-pressure descaling system in a linkage manner.

[0018] Furthermore, the high-frequency thermal parameters include at least the temperature of each heating zone and the oxygen enrichment flow rate of each zone, and the low-frequency energy consumption indicators include at least the total gas supply from each gas source and the energy consumption per ton of steel per shift.

[0019] Furthermore, the time window sliding algorithm traces backward based on the shift settlement time to extract the mean or time series features of furnace temperature and oxygen flow rate in each segment within that period, thereby achieving dynamic alignment of high- and low-frequency data.

[0020] The specific algorithm is as follows: Let the end time of the shift be T, and the time window length be Δt. Then, the effective polymerization temperature characteristic of the i-th heating segment within this time window is... The calculation formula is:

[0021]

[0022] in, For time windows Sampling time points within, This represents the total number of samplings within the time window. Similarly, the oxygen-enriched flow rate and gas flow rate of each segment are time-aligned using this formula to form a standard feature vector that corresponds one-to-one with the unit consumption of each shift.

[0023] Furthermore, the dynamic gas ratio reconstruction rule is as follows: when the coke oven gas supply is insufficient, the dynamic conditioning ratio instruction output by the machine learning prediction engine is constrained by a specific optimization domain, specifically: forcibly introducing natural gas with a calorific value of about 6% as the calorific value conditioning benchmark, and cooperating with converter gas with a calorific value of more than 50%, so as to lock the predicted unit consumption of the entire system within the target minimum range.

[0024] Furthermore, the system has a built-in pre-warning mechanism for unit consumption. Based on the Pearson correlation coefficient, it is determined that the average temperature of the preheating section is strongly negatively correlated with the unit consumption of the entire furnace, with a correlation coefficient threshold of approximately -0.8. When the average temperature of the preheating section is below 1000°C, the system determines it as a leading indicator of increased unit consumption and immediately triggers a high-calorific-value energy replenishment feedforward command.

[0025] Furthermore, the targeted oxygen enrichment control based on characteristic sensitivity specifically includes: explicitly excluding oxygen enrichment supply to the preheating section with low sensitivity; identifying a heating section and a homogenizing section whose response coefficients to oxygen enrichment flow are between 0.5 and 0.6, setting them as the first-level priority zone for oxygen enrichment compensation, and prioritizing the scheduling of oxygen enrichment resources to the first-level priority zone.

[0026] Furthermore, the active output of the high-temperature zone furnace dwell time compression command dynamically controls the time compression ratio between 2% and 3% to suppress the excessive growth of primary iron oxide scale from the source and reduce the iron scale indentation defect rate; after the furnace dwell time is compressed, it is determined that the oxide scale thickness has become thinner but the bonding force may change, and an adjustment command is automatically sent to the hot rolling high-pressure descaling system to dynamically match the descaling water pressure and spray angle.

[0027] The beneficial effects of this invention are:

[0028] 1. Overcoming the limitations of static isocalorific value regulation, significantly reducing the overall system consumption per unit area. Existing technologies often rely on manual experience to blindly replace gas volumes when facing fluctuations in high-calorific-value gas sources (such as coke oven gas), which can easily lead to a precipitous drop in thermal efficiency. This method introduces a random forest algorithm and a multidimensional nonlinear optimization model to simulate different "natural gas + oxygen-enriched" combinations at high frequency in the background. When the system detects a limitation in high-calorific-value gas sources, it can automatically output dynamic ratio reconstruction instructions (e.g., precisely locking the natural gas ratio at around 6%, combined with more than 50% converter gas). This mechanism fundamentally eliminates the lag in regulation, successfully stabilizing the overall system consumption per ton of steel in an extremely low range of around 0.90 GJ / t, with an overall reduction in consumption of 3%-5% under complex operating conditions.

[0029] 2. Achieving segmented targeted oxygen enrichment eliminates resource waste caused by indiscriminate oxygen addition. Existing oxygen-enriched combustion technologies mostly employ global diffused oxygen mixing, which not only easily leads to uneven furnace temperature but also wastes a large amount of high-value oxygen in inefficient heat exchange zones. This invention creatively introduces a characteristic sensitivity algorithm to calculate and clearly define the heating section and soaking section as the primary priority zone for oxygen enrichment compensation (correlation coefficient between 0.5 and 0.6), and eliminates ineffective oxygen supply in the preheating section. This "precise drip irrigation" style targeted oxygen enrichment control greatly saves industrial oxygen consumption while ensuring that effective heat is transferred to the interior of the steel billet with maximum efficiency.

[0030] 3. A pioneering closed-loop system integrating thermal control and surface quality significantly improves product yield. Traditional furnace control focuses solely on thermal parameters, often extending the billet's residence time in the high-temperature zone to ensure thorough heating, leading to frequent oxide scale indentation defects in subsequent cold rolling processes. This invention utilizes the high thermal intensity from targeted oxygen enrichment to proactively reduce the billet's residence time in the high-temperature zone by 2%-3%, suppressing excessive growth of primary iron oxide scale at its source. Simultaneously, the system automatically adjusts the water pressure and spray angle of the downstream hot rolling high-pressure descaling system based on theoretical oxide layer thickness predictions. Actual measurements show that this closed-loop mechanism reduces the incidence of "iron scale indentation" by approximately 63.9%, and "rough band" defects by approximately 42.9%, resulting in a breakthrough improvement in the first-grade yield of cold-rolled products.

[0031] 4. Establish a pre-emptive early warning mechanism for unit consumption to improve the stability of production rhythm and automation level. Existing technologies often only implement passive feedback adjustments after the temperature of the main heating section drops or the unit consumption statistics are obtained. This invention utilizes data mining to identify patterns and establish the preheating section temperature (e.g., dropping below 1000°C) as a highly negatively correlated feedforward early warning indicator. Once the threshold is triggered, the system immediately issues a high-calorific-value replenishment command. This shift from a "lagging feedback" to a "predictive feedforward" control mode significantly smooths out furnace temperature fluctuations, reduces the frequency of manual intervention, and ensures the rhythm stability of the continuous rolling production line. Attached Figure Description

[0032] Figure 1 This is a schematic diagram of the system structure of the present invention;

[0033] Figure 2 This is a flowchart of the multi-source high and low frequency data cross-time window alignment process described in this invention;

[0034] Figure 3 This is a graph showing the unit consumption characteristic weight and sensitivity analysis of each temperature control segment based on the Pearson correlation coefficient of this invention.

[0035] Figure 4 This is a flowchart of the dynamic gas ratio reconstruction and real-time optimization algorithm for unit consumption in this invention;

[0036] Figure 5 This is the cross-process quality control linkage logic diagram in this invention. Detailed Implementation

[0037] The present invention will be further described below with reference to the accompanying drawings and embodiments:

[0038] An energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace, such as... Figure 1As shown, the system includes a data acquisition module, a machine learning prediction and optimization engine, a collaborative execution and adaptive closed-loop module, and a downstream quality feedback module. The data acquisition module collects real-time data through sensors. The machine learning prediction and optimization engine uses a random forest regression algorithm to establish a nonlinear coupled prediction model for multi-gas source ratio, segmented oxygen enrichment, and multi-segment furnace temperature. The collaborative execution and adaptive closed-loop module includes a gas regulating valve and an oxygen regulating valve. The downstream quality feedback module is a high-pressure descaling system.

[0039] The control method of the energy-saving control system for dynamic ratio optimization of multiple gas sources in the heating furnace includes the following steps:

[0040] S1. Multi-source data cross-frequency alignment and perception: Collect high-frequency thermal parameters and low-frequency energy consumption indicators of the heating furnace, extract features and perform cross-frequency alignment through time window sliding algorithm, and construct a standard training sample set;

[0041] The high-frequency thermal parameters include at least the temperature of each heating zone and the oxygen-enriched flow rate of each zone, and the low-frequency energy consumption indicators include at least the total gas supply from each gas source and the energy consumption per ton of steel per shift.

[0042] Since gas consumption and unit consumption statistics are typically low-frequency (e.g., every 8 hours), while furnace temperature and oxygen flow rate are high-frequency (minute-level), the system uses a time window sliding algorithm to trace back from the shift settlement time, extracting the mean or time series features of furnace temperature and oxygen flow rate for each segment within that time period. This achieves dynamic alignment of high- and low-frequency data. Figure 2 As shown, this provides accurate training samples for the model.

[0043] The system performs cross-time window aggregation calculations on high- and low-frequency data. For minute-level furnace temperature and oxygen data, and shift-level (e.g., 8-hour) unit consumption data, the system employs a time window mean smoothing algorithm for feature extraction. Let the shift end time be T, and the time window length be Δt. Then, the effective polymerization temperature characteristics of the i-th heating segment within this time window are... The calculation formula is:

[0044]

[0045] in, For time windows Sampling time points within, This represents the total number of samplings within the time window. Similarly, the oxygen-enriched flow rate and gas flow rate of each segment are time-aligned using this formula to form a standard feature vector that corresponds one-to-one with the unit consumption of each shift.

[0046] S2, the process of dynamic gas ratio reconstruction and unit consumption optimization, such as Figure 4As shown, real-time collected operating data is input into a pre-trained machine learning prediction and optimization engine. When fluctuations in the supply of high-calorific-value gas are detected, the machine learning prediction and optimization engine performs multi-objective simulation calculations in the background to find the optimal combination of "natural gas conditioning + converter gas". It outputs a dynamic conditioning ratio instruction that minimizes the predicted unit consumption and sends it to the gas / oxygen regulating valve of the collaborative execution and adaptive closed-loop module to maintain the current static gas ratio, thereby achieving the closed-loop control logic with the lowest unit consumption.

[0047] The dynamic gas ratio reconfiguration breaks the linear substitution rule of isocalorific value; when faced with insufficient supply of high-calorific-value gas (such as coke oven gas) or drastic fluctuations in calorific value, the ratio reconfiguration is initiated:

[0048] First, the optimization engine performs high-frequency simulation calculations in the background to find the lowest unit consumption balance point;

[0049] Secondly, guided by the model, the system proactively reduces the proportion of coke oven gas and increases the proportion of converter gas (LDG) to 50%-55% as basic heat support; at the same time, natural gas (NG) is introduced as a calorific value "control," precisely locking its proportion at around 6%. This specific combination of "converter gas + natural gas conditioning" can effectively lock the entire system's unit consumption within the optimal operating range of around 0.90 GJ / t.

[0050] The optimization engine is built upon a prediction objective function based on random forest regression. Let the theoretical prediction unit cost be... Its function mapping relationship is expressed as:

[0051]

[0052] in, This is a gas blending vector that includes coke oven, blast furnace, converter, and natural gas. The oxygen-enriched flow vector for each segment; This is the furnace temperature state vector for each section.

[0053] During real-time optimization, the system aims to find a control strategy that minimizes unit cost, i.e., to solve an optimization problem.

[0054]

[0055] The constraints include:

[0056] 1. Minimum calorific value constraint: (in (Theoretical calorific value for a single gas source).

[0057] 2. Key conditioning and blending ratio constraints: Under low calorific value compensation conditions, the natural gas blending ratio is forcibly constrained. ≈6%, converter gas ratio 50%.

[0058] The system has a built-in early warning mechanism for unit consumption; based on the Pearson correlation coefficient, it is determined that the average temperature of the preheating section is strongly negatively correlated with the unit consumption of the whole furnace, and the correlation coefficient threshold is about -0.8; when the average temperature of the preheating section is lower than 1000°C, the system judges it as a leading indicator of increased unit consumption and immediately triggers a high calorific value energy replenishment feedforward command.

[0059] A pre-warning mechanism for unit consumption based on preheating section temperature, such as Figure 3 As shown, feature weight analysis revealed that the average temperature of the preheating section and the final unit consumption of the entire furnace exhibit a very high negative correlation (correlation coefficient reaches approximately -0.8).

[0060] The system sets the preheating section as a "pre-warning zone for unit consumption". Figure 3 The impact weight of each process parameter on unit consumption is intuitively displayed. The strong negative correlation of preheating section temperature (as an early warning indicator) and the high positive correlation of primary heating section / soaking section temperature (as a target oxygen-enriched zone) are particularly highlighted, which are strong evidence for the core algorithm logic of this method.

[0061] When the average temperature of the preheating section is below 1000°C, the system determines that the heat load of the subsequent heating section will increase and the unit consumption will rise. It immediately triggers the "high calorific value energy supplement feedforward command" to increase the total calorific value input of the system in advance and avoid a significant loss of control over the unit consumption.

[0062] S3. Based on characteristic sensitivity, target oxygen enrichment control is used to calculate the correlation coefficient between the temperature and oxygen enrichment flow rate of each heating zone and the total furnace consumption. Based on this, the oxygen enrichment compensation priority zone is divided, and the oxygen enrichment flow rate is concentrated and allocated to the high sensitivity zone.

[0063] The targeted oxygen enrichment control based on characteristic sensitivity specifically includes: explicitly excluding oxygen enrichment supply to the preheating section with low sensitivity; identifying a heating section and a homogenizing section whose response coefficients to oxygen enrichment flow are between 0.5 and 0.6, setting them as the first-level priority zone for oxygen enrichment compensation, and prioritizing the scheduling of oxygen enrichment resources to the first-level priority zone.

[0064] Based on the Pearson correlation coefficient, the system calculates the target sensitivity by determining the Pearson correlation coefficient between each process parameter and the overall furnace consumption. This is to achieve the division between "preheating phase early warning" and "targeted oxygen-enriched priority zone":

[0065]

[0066] In the formula, For the first The average temperature or oxygen flow characteristics of a certain section of a sample; This corresponds to the actual consumption per ton of steel. This represents the total number of historical samples. and These are the sample means of the corresponding variables.

[0067] Based on real-time model calculations, the correlation coefficient between preheating section temperature and unit consumption is... -0.8

[0068] At that time, it was established as a leading reverse early warning indicator for unit consumption;

[0069] When the response coefficients of the heating section and the soaking section to the oxygen-enriched flow rate hour,

[0070] Based on this, the system designated it as a first-level priority zone for oxygen-enriched compensation.

[0071] Based on feature sensitivity, the segmented targeted oxygen enrichment control system abandons the conventional approach of blindly enriching oxygen globally, and performs targeted resource scheduling based on the Pearson correlation coefficient:

[0072] 1. The temperature of the heating section and the homogenization section are identified as being most sensitive to oxygen flow rate (correlation coefficient reaches 0.5-0.6), and these sections are set as the "primary priority zone for oxygen enrichment compensation".

[0073] 2. When the production pace accelerates or rapid heat compensation is required, the system automatically intercepts ineffective oxygen supply to the preheating section (correlation coefficient < 0.3) and concentrates the oxygen-enriched flow to the heating section and the uniform heating section to achieve precise drip irrigation of heat.

[0074] S4. Closed-loop control of surface quality across processes, such as Figure 5 As shown, the heat exchange efficiency increment is calculated based on the oxygen-enriched enhanced combustion state in step S3, and a command to compress the furnace dwell time in the high-temperature zone is actively output. The theoretical iron oxide scale state change is fed forward to the high-pressure descaling system of the downstream quality feedback module, and the water pressure and injection parameters of the high-pressure descaling system are adjusted in a coordinated manner.

[0075] The integrated closed-loop quality control, which combines furnace dwell time and descaling pressure, not only optimizes energy consumption but also extends thermal control to downstream surface quality management.

[0076] 1. Reduced furnace dwell time: By utilizing the high heat intensity generated by oxygen-enriched combustion in the "primary priority zone", the system actively reduces the furnace dwell time of the billet in the high-temperature zone by 2%-3%, effectively inhibiting the excessive growth of primary iron oxide scale.

[0077] 2. Adaptive descaling pressure: When the system senses that the furnace dwell time has been compressed, it determines that the oxide scale thickness has become thinner but the bonding force may change (fluctuation of iron scale indentation risk). It automatically sends an adjustment command to the hot rolling high-pressure descaling system to dynamically match the descaling water pressure and spray angle, ultimately achieving a comprehensive quality benefit of more than 60% reduction in "iron scale indentation" defects.

[0078] To achieve a cross-boundary closed-loop estimation of surface quality, the system incorporates an empirical formula relating furnace dwell time and iron scale thickness based on parabolic oxidation laws, which is used to estimate the dynamics of iron oxide scale in conjunction with furnace dwell time.

[0079]

[0080] In the formula, This refers to the thickness of the iron oxide scale; To be affected by the current furnace temperature With oxygen concentration The oxidation rate constant affected by the temperature; $\tau$ represents the furnace dwell time in the high-temperature zone.

[0081] After the system enhances the heat exchange rate through oxygen-enriched combustion, it actively outputs a time compression command, setting a new target furnace dwell time. satisfy:

[0082]

[0083] in, The system uses the base dwell time under conventional air combustion as a reference, and adjusts the compression ratio based on a real-time prediction model. The temperature is dynamically controlled between 2% and 3%, thereby ensuring the uniformity of the billet temperature while significantly reducing the calculated theoretical oxide scale thickness h. This result is then sent to the descaling system to adjust the spray water pressure.

[0084] Cross-process surface quality closed-loop control breaks down the information silos between the heating furnace and the rolling line. It compresses the furnace dwell time through oxygen-enriched combustion and adjusts the pressure of the downstream high-pressure descaling water accordingly, thus realizing an innovative mechanism for surface quality closed-loop control.

[0085] The system and its control method have been applied to the No. 1-3 heating furnaces in the hot rolling mill of this enterprise, reducing gas consumption by 12.7%.

Claims

1. An energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace, characterized in that: It includes a data acquisition module, a machine learning prediction and optimization engine, a collaborative execution and adaptive closed-loop module, and a downstream quality feedback module. The data acquisition module collects real-time data through sensors. The machine learning prediction and optimization engine uses a random forest regression algorithm to establish a nonlinear coupled prediction model for multi-gas source ratio, segmented oxygen enrichment, and multi-segment furnace temperature. The collaborative execution and adaptive closed-loop module includes a gas regulating valve and an oxygen regulating valve. The downstream quality feedback module is a high-pressure descaling system.

2. The control method of the energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace as described in claim 1, characterized in that: Includes the following steps: S1. Multi-source data cross-frequency alignment and perception: Collect high-frequency thermal parameters and low-frequency energy consumption indicators of the heating furnace, extract features and perform cross-frequency alignment through time window sliding algorithm, and construct a standard training sample set; S2, Dynamic Gas Ratio Reconstruction and Unit Consumption Optimization: Real-time collected operating data is input into a pre-trained machine learning prediction and optimization engine. When fluctuations in the supply of high-calorific-value gas are detected, the machine learning prediction and optimization engine performs multi-objective simulation calculations in the background, outputs a dynamic conditioning ratio instruction that minimizes the predicted unit consumption, and sends it to the gas / oxygen regulating valve of the collaborative execution and adaptive closed-loop module. S3. Based on characteristic sensitivity, target oxygen enrichment control calculates the correlation coefficient between the temperature and oxygen enrichment flow rate of each heating zone and the total furnace consumption. Based on this, the oxygen enrichment compensation priority zone is divided, and the oxygen enrichment flow rate is concentrated and allocated to the high-sensitivity zone. S4. Cross-process surface quality closed-loop control calculates the heat exchange efficiency increment based on the oxygen enrichment enhanced combustion state in step S3, actively outputs the high-temperature zone furnace dwell time compression command, and feeds forward the theoretical iron oxide scale state change to the high-pressure descaling system of the downstream quality feedback module, and adjusts the water pressure and injection parameters of the high-pressure descaling system in a linkage manner.

3. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The high-frequency thermal parameters include at least the temperature of each heating zone and the oxygen enrichment flow rate of each zone, and the low-frequency energy consumption indicators include at least the total gas supply from each gas source and the energy consumption per ton of steel per shift.

4. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The time window sliding algorithm traces backward from the shift settlement time to extract the average value or time series features of furnace temperature and oxygen flow rate in each segment within that time period, thereby achieving dynamic alignment of high- and low-frequency data. The specific algorithm is as follows: Let the end time of the shift be T, and the time window length be Δt. Then, the effective polymerization temperature characteristic of the i-th heating segment within this time window is... The calculation formula is: in, For time windows Sampling time points within, This represents the total number of samplings within the time window. Similarly, the oxygen-enriched flow rate and gas flow rate of each segment are time-aligned using this formula to form a standard feature vector that corresponds one-to-one with the unit consumption of each shift.

5. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The dynamic gas ratio reconstruction rule is as follows: when the coke oven gas supply is insufficient, the dynamic conditioning ratio instruction output by the machine learning prediction engine is constrained by a specific optimization domain. Specifically, it is to forcibly introduce natural gas, accounting for about 6% of the calorific value, as the calorific value conditioning benchmark, and combine it with converter gas, accounting for more than 50%, so as to lock the predicted unit consumption of the entire system within the target minimum range.

6. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The system has a built-in early warning mechanism for unit consumption. Based on the Pearson correlation coefficient, it is determined that the average temperature of the preheating section is strongly negatively correlated with the unit consumption of the whole furnace, and the correlation coefficient threshold is about -0.

8. When the average temperature of the preheating section is lower than 1000°C, the system judges it as a leading indicator of increased unit consumption and immediately triggers a high calorific value energy replenishment feedforward command.

7. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The targeted oxygen enrichment control based on characteristic sensitivity specifically includes: explicitly excluding oxygen enrichment supply to the preheating section with low sensitivity; identifying a heating section and a homogenizing section whose response coefficients to oxygen enrichment flow are between 0.5 and 0.6, setting them as the first-level priority zone for oxygen enrichment compensation, and prioritizing the scheduling of oxygen enrichment resources to the first-level priority zone.

8. The control method for an energy-saving control system for dynamic ratio optimization of multiple gas sources in a heating furnace according to claim 2, characterized in that: The active output command for compressing the furnace dwell time in the high-temperature zone is dynamically controlled between 2% and 3% to suppress the excessive growth of primary iron oxide scale and reduce the iron scale indentation defect rate from the source. After the furnace dwell time is compressed, it is determined that the oxide scale thickness has become thinner but the bonding force may have changed. An adjustment command is automatically sent to the hot rolling high-pressure descaling system to dynamically match the descaling water pressure and spray angle.

Images