Method and system for self-adaptive control of calorific value of mixed gas

Abstract

The application discloses a kind of methods and systems of mixed coal gas heat value self-adaptive control, it is related to mixed fuel gas power generation field, it includes obtaining blast furnace gas and converter gas historical operation data, constructs first prediction model and second prediction model, and collects real-time pressure temperature data, generates temperature-pressure time sequence characteristics and then inputs first prediction model and second prediction model, obtains mixing parameter;Equivalent dust load is generated according to historical operation data, and target mixing ratio is determined;Valve control instruction is generated according to target mixing ratio, and feedforward prediction mixed heat value sequence is generated;Actual heat value sequence of mixed coal gas is obtained, dynamic feedback time delay is determined, and heat value residual error is generated;First prediction model and second prediction model are corrected based on heat value residual error, and are used for next control cycle.The application has the effect of realizing mixed coal gas heat value locking, guaranteeing the efficient and stable combustion operation of unit based on pressure temperature feedforward and big data model prediction, combined with double closed-loop adaptive control.

Description

Technical Field

[0001] This application relates to the field of mixed gas power generation, and in particular to a method and system for adaptive control of the calorific value of mixed gas. Background Technology

[0002] In industrial applications where blast furnace gas and converter gas are mixed to power internal combustion engines, stable control of the mixed gas's calorific value is a core prerequisite for ensuring the safe, efficient, and economical operation of the generator set. Internal combustion engines are extremely sensitive to the calorific value of the input gas, with the optimal economic operating range typically locked within a narrow range of 860 kcal per standard cubic meter. If the mixed calorific value is too high, it often means an excessively large proportion of high-calorific-value but high-dust-content converter gas, directly leading to a sharp increase in gas dust load, which in turn causes problems such as increased dust accumulation inside the unit, pipe wear and blockage, and excessive emissions. If the mixed calorific value is too low, the gas's ability to perform work is insufficient, resulting in a decrease in the unit's power output, and in severe cases, even major operational failures such as combustion instability, flameout, or inability to start normally.

[0003] Currently, the industry generally adopts a traditional control mode for regulating the calorific value of mixed gas, which uses gas flow measurement data as the basis for regulation and combines it with manual experience. However, this mode has inherent technical defects: on the one hand, the flow meter measurement response is severely lagging, and the obtained data has extremely poor real-time performance, which cannot truly reflect the rapid fluctuations in calorific value under the instantaneous changes in operating parameters such as gas pressure and temperature, and therefore cannot serve as an effective basis for closed-loop dynamic regulation; on the other hand, due to the complex operating conditions of blast furnace gas and converter gas sources, the calorific value of the two gas sources will fluctuate significantly and randomly at any time. The method of manually adjusting the mixing ratio based on experience after observing the calorific value meter is not only lagging and inaccurate, but also lacks the ability to predict in advance, and cannot adapt to such drastic and disorderly changes in operating conditions.

[0004] The aforementioned defects ultimately lead to a long-term and significant deviation of the calorific value of the mixed gas from the optimal economic point, making it impossible for the unit to operate stably within the optimal efficiency range. This results in poor power generation economics, severe equipment wear and tear, and frequent malfunctions, urgently requiring solutions through more intelligent and responsive control methods. Summary of the Invention

[0005] To address the issues of lag, low precision, uncontrollable dust load, and unstable operation of internal combustion engine generator sets in current mixed gas power generation systems, this application provides a method and system for adaptive control of the calorific value of mixed gas.

[0006] In the first aspect, the method for adaptive control of calorific value of mixed gas provided in this application adopts the following technical solution: S1, acquire historical operating data of blast furnace gas and converter gas, construct a first prediction model corresponding to blast furnace gas and a second prediction model corresponding to converter gas, and collect real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series characteristics. S2. Input the temperature and pressure time series characteristics into the first prediction model and the second prediction model to obtain the mixing parameters; the mixing parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary. S3. Generate an equivalent dust load based on the historical operating data, and determine the target blending ratio based on the equivalent dust load, the blending parameters, and the preset target mixing calorific value; S4. Generate valve control commands based on the target blending ratio to adjust the actual blending ratio, and generate a feedforward predicted blending calorific value sequence based on the target blending ratio and the blending parameters. S5. Obtain the actual calorific value sequence of the mixed gas, and determine the dynamic feedback delay based on the feedforward predicted mixed calorific value sequence, and generate the calorific value residual based on the dynamic feedback delay; S6. Based on the calorific value residual, correct the first prediction model and the second prediction model, and use the corrected first prediction model and the corrected second prediction model for the next control cycle.

[0007] By adopting the above technical solution, a first and second prediction model are constructed using historical operating data of blast furnace gas and converter gas. Real-time pressure and temperature data are collected to generate temperature and pressure time-series characteristics. The prediction model yields mixing parameters including predicted calorific value and calorific value confidence boundaries. The target mixing ratio is then determined by combining the equivalent dust load with the target mixed calorific value, thereby generating valve control commands and a feedforward predicted mixed calorific value sequence. The dynamic feedback delay is determined by obtaining the actual calorific value sequence of the mixed gas to generate the calorific value residual. Based on this residual, the first and second prediction models are corrected for use in the next control cycle. This solution organically integrates feedforward prediction, feedback correction, and dust load constraints, overcoming the shortcomings of traditional control response lag and single calorific value target optimization. It can promptly and accurately adaptively adjust the calorific value of the mixed gas while actively adapting to the load limitations of the dust removal system, achieving synergistic optimization of calorific value stability and dust emission control, and ensuring stable downstream production.

[0008] Optionally, the construction of the first prediction model corresponding to blast furnace gas and the second prediction model corresponding to converter gas specifically includes removing anomalies, aligning time, and filtering stable operating conditions from historical operating data to obtain valid historical operating data; Based on the aforementioned valid historical operating data, a sample of operating conditions relating pressure, temperature, and gas calorific value is constructed. The first and second prediction models are constructed and trained based on the operating condition samples.

[0009] By employing the aforementioned technical solution, valid historical operating data is obtained through anomaly removal, time alignment, and stable operating condition screening. Based on this data, operating condition samples are constructed, and the first and second prediction models are trained. This process eliminates non-steady-state disturbances and noise, ensuring the high quality and representativeness of the modeling data. This allows the prediction models to more accurately reflect the mapping relationship between pressure, temperature, and gas calorific value, significantly improving the accuracy and generalization ability of calorific value prediction, and laying a solid foundation for subsequent precise blending control.

[0010] Optionally, the step of constructing and training the first prediction model and the second prediction model based on the operating condition samples specifically includes identifying the gas supply conditions based on the operating condition samples and obtaining the operating condition weights. The operating condition sample is input into the expert prediction sub-model corresponding to the gas supply operating condition to obtain the expert predicted calorific value. Based on the operating condition weights and the expert-predicted heat values, the model output rules of the first prediction model and the second prediction model are obtained.

[0011] By adopting the above technical solution, the gas supply conditions are identified based on the operating condition samples, and the operating condition weights are obtained. The operating condition samples are then input into the corresponding expert prediction sub-model to generate expert predicted calorific values. Finally, the models are fused according to the operating condition weights to form the model output rules. This multi-expert sub-model fusion method uses operating condition weights to achieve soft switching, overcoming the problem of decreased accuracy of a single global model when operating conditions change over a wide range. It achieves adaptive fitting to complex gas supply characteristics and improves the robustness and accuracy of calorific value prediction under varying load and gas quality conditions.

[0012] Optionally, the step of collecting real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series features specifically includes collecting real-time pressure data and real-time temperature data of blast furnace gas and converter gas. Based on the real-time pressure data and the real-time temperature data, pressure change characteristics, temperature change characteristics, and equivalent density characteristics are generated. By integrating the real-time pressure data, the real-time temperature data, the pressure change characteristics, the temperature change characteristics, and the equivalent density characteristics, a temperature-pressure time series characteristic is obtained.

[0013] By employing the above technical solution, real-time pressure and temperature data of blast furnace gas and converter gas are collected, generating pressure change characteristics, temperature change characteristics, and equivalent density characteristics, which are then fused to obtain temperature-pressure time-series characteristics. These multi-dimensional features fully capture the dynamic information of gas calorific value changes with thermodynamic state, reflecting calorific value fluctuation trends more sensitively than simple instantaneous values. This provides more discriminative input for prediction models, thereby improving the response speed and accuracy of calorific value prediction.

[0014] Optionally, S2 specifically includes the first prediction model and the second prediction model outputting initial predicted calorific value and prediction uncertainty based on the temperature and pressure time series characteristics; Based on the type of gas, the trend of calorific value change, and the pressure and temperature disturbance state, the initial predicted calorific value is corrected by physical property constraints to obtain the predicted calorific value of blast furnace gas and the predicted calorific value of converter gas. The predicted calorific value of the blast furnace gas, the predicted calorific value of the converter gas, and the prediction uncertainty are used to generate a calorific value confidence boundary; these are then combined to obtain the blending parameters.

[0015] By adopting the above technical solution, the first and second prediction models output initial predicted calorific values ​​and prediction uncertainties based on the temperature and pressure time series characteristics. Then, physical property constraints are corrected according to the type of gas, the trend of calorific value change, and the pressure and temperature disturbance state to obtain the predicted calorific values ​​of blast furnace gas, converter gas, and calorific value confidence boundaries. This physical property constraint correction step utilizes the inherent physical properties of gas to suppress abnormal outputs of the model under disturbances, ensuring that the predicted calorific value remains within a reasonable physical range. This improves the reliability of the calorific value confidence boundaries and the safety of decision-making, effectively avoiding valve malfunctions caused by drastic jumps in predicted values.

[0016] Optionally, S3 specifically includes S31, determining the converter gas demand range based on the calorific value confidence boundary and the target mixed calorific value; S32. Generate an equivalent dust load based on the historical operating data, and determine the allowable range of converter gas based on the equivalent dust load; S33. Determine the target blending ratio based on the converter gas demand range and the converter gas allowable range.

[0017] By adopting the above technical solution, the converter gas demand range is determined based on the calorific value confidence boundary and the target mixed calorific value. The allowable range of converter gas is determined based on the equivalent dust load generated from historical operating data. Finally, the target blending ratio is determined comprehensively. This solution incorporates the calorific value adjustment requirements and the dust load capacity of the dust removal system into a unified decision-making framework. This ensures that while prioritizing the satisfaction of the target calorific value, the allowable dust handling boundary is not exceeded. It avoids overloading of dust removal equipment or exceeding emission standards due to improper blending ratios, thus achieving a dynamic balance between production indicators and environmental protection requirements.

[0018] Optionally, S33 specifically includes determining a target blending ratio based on the converter gas demand range and generating a dust overdraft amount when the converter gas demand range exceeds the allowable range and the mixed gas has a risk of insufficient calorific value. When the converter gas demand range and the converter gas allowable range meet the dust repayment conditions, the target blending ratio is reduced according to the dust overdraft amount to obtain the reduced target blending ratio. The equivalent dust load is updated based on the reduced target blending ratio.

[0019] By adopting the above technical solution, when the converter gas demand exceeds the allowable range and there is a risk of insufficient calorific value, a dust overdraft is generated, and a target blending ratio is determined accordingly. When the repayment conditions are met, the target blending ratio is reduced to repay the overdraft, and the equivalent dust load is updated. This overdraft and repayment mechanism ensures the continuity and stability of the calorific value supply of the mixed gas while implementing closed-loop management of the cumulative load of the dust removal system, preventing the dust load from exceeding the design capacity for a long period of time. This not only ensures the calorific value supply under emergency conditions but also extends the service life of the dust removal equipment and reduces maintenance costs.

[0020] Optionally, S5 specifically includes obtaining the actual calorific value sequence of the mixed gas and determining the dynamic feedback delay based on the correlation between the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas. Based on the dynamic feedback delay, the actual calorific value sequence of the mixed gas is time-aligned with the feedforward predicted mixed calorific value sequence to obtain aligned calorific value data; The calorific value residual is generated based on the aligned calorific value data.

[0021] By employing the above technical solution, the actual calorific value sequence of the mixed gas is obtained. The dynamic feedback delay is determined by calculating its correlation with the feedforward predicted mixed calorific value sequence, and the two sequences are then aligned in time to generate the calorific value residual. This dynamic delay identification method can adaptively track the lag time drift caused by changes in pipeline length and flow rate, ensuring that the residual calculation is performed on the correct time segment. This avoids misaligned comparisons under the fixed delay assumption, significantly improving the effectiveness of the calorific value residual, making model correction more accurate, and enhancing the reliability and adaptability of feedback control.

[0022] Optionally, S6 specifically includes generating blast furnace gas calorific value bias and converter gas calorific value bias based on the calorific value residual; The first prediction model is corrected based on the calorific value bias of the blast furnace gas, and the second prediction model is corrected based on the calorific value bias of the converter gas. The modified first prediction model and the modified second prediction model are output to the next control cycle.

[0023] By adopting the above technical solution, calorific value biases for blast furnace gas and converter gas are generated based on the calorific value residuals. These biases are then used to correct the first and second prediction models, respectively, and the corrected models are applied to the next control cycle. This dual-channel bias correction strategy can specifically compensate for the independent drift of each gas calorific value, avoiding coupling errors caused by a single total bias. This allows the model to maintain prediction accuracy over long-term operation, effectively overcoming performance degradation caused by gradual changes in gas composition or sensor characteristic shifts, and achieving online adaptive calorific value control.

[0024] In a second aspect, the present invention provides an adaptive control system for the calorific value of mixed coal gas, including a data acquisition and modeling module, a model prediction module, a blending ratio determination module, a valve control module, a calorific value feedback module, and a model correction module; The data acquisition and modeling module is used to acquire historical operating data of blast furnace gas and converter gas, construct a first prediction model for blast furnace gas and a second prediction model for converter gas, and acquire real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series features. The model prediction module is connected to the data acquisition and modeling module, and is used to input the temperature and pressure time series features into the first prediction model and the second prediction model to obtain the mixing parameters; the mixing parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary. The blending ratio determination module is connected to the model prediction module and is used to generate an equivalent dust load based on the historical operating data, and to determine the target blending ratio based on the equivalent dust load, the blending parameters and the preset target mixing calorific value. The valve control module is connected to the blending ratio determination module and is used to generate valve control commands according to the target blending ratio to adjust the actual blending ratio, and to generate a feedforward predicted blending calorific value sequence based on the target blending ratio and the blending parameters. The calorific value feedback module is connected to the valve control module and is used to obtain the actual calorific value sequence of the mixed gas, determine the dynamic feedback delay based on the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas, and generate the calorific value residual based on the dynamic feedback delay. The model correction module is connected to the calorific value feedback module and the data acquisition and modeling module respectively, and is used to correct the first prediction model and the second prediction model based on the calorific value residual, and use the corrected first prediction model and the corrected second prediction model in the next control cycle.

[0025] The technical effects of the adaptive control system for the calorific value of mixed gas provided in the second aspect of the present invention are the same as those in the first aspect, and will not be repeated here.

[0026] In summary, this application includes at least one of the following beneficial technical effects: 1. By adopting a calorific value prediction model for blast furnace gas and converter gas based on temperature and pressure time sequence characteristics, combined with a closed-loop correction architecture that combines feedforward prediction of mixed calorific value sequence and dynamic feedback delay compensation, the influence of gas transmission and detection lag on control accuracy can be overcome, enabling rapid and accurate adaptive adjustment of mixed gas calorific value. This effectively solves the problem of mixed calorific value fluctuation caused by slow response of conventional feedback control and model mismatch, thereby improving the operational stability and thermal efficiency of combustion equipment. 2. Due to the adoption of an equivalent dust load generation and a coordinated decision-making mechanism for the demand range and allowable range of converter gas, and the introduction of dust overdraft and repayment strategies, the dust removal system's capacity can be considered in real time during the adjustment of calorific value. This avoids excessive dust emissions or overload of dust removal equipment caused by excessive use of converter gas, and achieves synergistic optimization of mixed gas calorific value control and environmental dust removal. It solves the problem that calorific value targets and environmental constraints are difficult to balance in traditional control. 3. By employing multi-condition expert sub-model fusion, property constraint correction, and calorific value confidence boundary generation methods, combined with online calorific value residuals to perform gas type bias correction on the prediction model, it can adapt to a wide range of changes and disturbances in the gas source characteristics of blast furnace gas and converter gas, continuously maintain high-precision calorific value prediction and decision-making, effectively solve the problem of poor adaptability of single models or fixed parameter control under varying operating conditions, and thus achieve robust self-adaptation and low maintenance costs in long-term system operation. Attached Figure Description

[0027] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 This is a flowchart of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 2 This is a flowchart of step S13 of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 3 This is a flowchart of step S14 of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 4 This is a flowchart of step S2 of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 5This is a flowchart of step S33 of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 6 This is a flowchart of step S5 of the method for adaptive control of the calorific value of mixed gas provided in the embodiments of this application; Figure 7 This is an overall architecture diagram of the adaptive control system for the calorific value of mixed gas provided in the embodiments of this application.

[0029] Attached reference numerals: 1. Data acquisition and modeling module; 2. Model prediction module; 3. Blending ratio determination module; 4. Valve control module; 5. Calorific value feedback module; 6. Model correction module. Detailed Implementation

[0030] The following is in conjunction with the appendix Figure 1 -Appendix Figure 7 This application will be described in further detail.

[0031] This application discloses a method and system for adaptive control of the calorific value of mixed gas.

[0032] An adaptive control method for the calorific value of mixed gas is applied to a gas supply system for an internal combustion engine generator set, which supplies gas after mixing blast furnace gas and converter gas. The gas supply system may include blast furnace gas pipelines, converter gas pipelines, mixed gas pipelines, blast furnace gas regulating valves, converter gas regulating valves, pressure detection devices, temperature detection devices, online calorific value detection devices for mixed gas, and a control device. The control device is communicatively connected to the pressure detection device, temperature detection device, online calorific value detection device for mixed gas, blast furnace gas regulating valve, and converter gas regulating valve, respectively, and is used to execute the adaptive control method for the calorific value of mixed gas described in this embodiment.

[0033] In this embodiment, blast furnace gas is a gas with a relatively low calorific value, generally between 700-900 kcal / standard cubic meter in the industry; converter gas is a gas with a relatively high calorific value and relatively high dust content, generally between 1670-2000 kcal / standard cubic meter. Mixed gas refers to the gas formed by mixing blast furnace gas and converter gas in a certain proportion in a mixed gas pipeline and supplied to the internal combustion engine generator set. Since the calorific values ​​of blast furnace gas and converter gas fluctuate randomly with upstream production conditions, gas holder status, pipeline pressure fluctuations, temperature changes, and changes in the gas recovery stage, relying solely on manual observation of the mixed gas calorific value and manual adjustment of the blast furnace gas regulating valve and converter gas regulating valve can easily lead to problems such as lag in regulation, low regulation accuracy, uncontrollable dust load, and unstable operation of the internal combustion engine generator set.

[0034] This embodiment collects real-time pressure and temperature data of blast furnace gas and converter gas to construct temperature and pressure time series characteristics. It then uses a first prediction model and a second prediction model to predict the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary. Combined with the equivalent dust load and the preset target mixing calorific value, it determines the target blending ratio. Subsequently, it outputs valve control commands according to the target blending ratio to adjust the actual blending ratio. At the same time, it uses the feedforward predicted mixing calorific value sequence and the actual calorific value sequence of the mixed gas to determine the dynamic feedback delay. Based on the dynamic feedback delay, it generates the calorific value residual and uses the calorific value residual to correct the first prediction model and the second prediction model, thereby forming an adaptive control closed loop that combines temperature and pressure feedforward prediction, dust-constrained blending, and hysteresis calorific value feedback correction.

[0035] See attached document Figure 1 Specifically, the adaptive control method for the calorific value of mixed gas includes the following steps.

[0036] S1. Obtain historical operating data of blast furnace gas and converter gas, construct the first prediction model corresponding to blast furnace gas and the second prediction model corresponding to converter gas, and collect real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series characteristics.

[0037] In this embodiment, historical operating data refers to the data set recorded and stored by field instruments, actuators, control systems, and operation management systems during the historical operation of blast furnace gas, converter gas, and mixed gas. Historical operating data includes blast furnace gas pressure data, blast furnace gas temperature data, blast furnace gas calorific value data, converter gas pressure data, converter gas temperature data, converter gas calorific value data, actual calorific value data of mixed gas, blast furnace gas regulating valve opening data, converter gas regulating valve opening data, mixed gas pipeline pressure data, pipeline pressure drop data, internal combustion engine generator set load data, internal combustion engine generator set operating status data, and operating records related to ash removal, purging, and maintenance. Historical operating data can originate from the field distributed control system, data acquisition system, energy management system, or independently configured edge controller.

[0038] Among them, blast furnace gas pressure data refers to the pressure data collected by pressure detection devices installed in the blast furnace gas pipeline or at the blast furnace gas inlet. Blast furnace gas temperature data refers to the temperature data collected by temperature detection devices installed in the blast furnace gas pipeline or at the blast furnace gas inlet. Converter gas pressure data refers to the pressure data collected by pressure detection devices installed in the converter gas pipeline or at the converter gas inlet. Converter gas temperature data refers to the temperature data collected by temperature detection devices installed in the converter gas pipeline or at the converter gas inlet. Blast furnace gas calorific value data and converter gas calorific value data can be obtained from historical calorific value detection devices, offline test records, upstream gas quality statistics, model inversion results during stable operation, or a validated operating database. Mixed gas actual calorific value data refers to the mixed gas calorific value detected by the online mixed gas calorific value detection device during historical operation. Internal combustion engine generator set operating status data is used to distinguish between the start-up process, shutdown process, stable operation process, load adjustment process, and abnormal fault process.

[0039] The first prediction model is used to predict the calorific value of blast furnace gas based on its temperature and pressure time-series characteristics. The second prediction model is used to predict the calorific value of converter gas based on its temperature and pressure time-series characteristics. Neither the first nor the second prediction model is a simple regression model that directly outputs calorific value based on pressure and temperature at a specific moment. Instead, they are prediction models that respectively include operating condition identification, expert prediction, and physical property constraints. Operating condition identification identifies the gas supply conditions based on pressure, temperature, and their time-varying characteristics. Expert prediction outputs predicted calorific values ​​from corresponding expert prediction sub-models under different gas supply conditions. Physical property constraints make reasonable corrections to the model output based on gas type, calorific value variation trends, and pressure and temperature disturbances.

[0040] Real-time pressure and temperature data refer to the data collected in real time by the blast furnace gas pressure detection device, blast furnace gas temperature detection device, converter gas pressure detection device, and converter gas temperature detection device within the current control cycle. Temperature and pressure time-series characteristics refer to the set of features formed by the real-time pressure and temperature data and their changes within a time window. These characteristics include not only the real-time pressure and temperature data themselves, but also pressure change characteristics, temperature change characteristics, and equivalent density characteristics. Since the calorific value of gas is mainly affected by its composition, and online detection of gas composition often suffers from lag or high costs in industrial settings, this embodiment introduces temperature and pressure time-series characteristics. The trajectory of pressure and temperature changes over time is used as an indirect representation of the gas supply status and gas quality changes, thus providing a data foundation for subsequent predictions of the calorific value of blast furnace gas and converter gas.

[0041] In this embodiment, step S1 includes S11: obtaining historical operating data, performing anomaly removal, time alignment, and stable operating condition filtering on the historical operating data to obtain valid historical operating data.

[0042] Anomaly removal refers to removing data from historical operating data that clearly does not conform to the operating patterns of the equipment or the measurement patterns of the instruments. Anomaly data can include fixed extreme value data caused by disconnection of pressure detection devices, missing data caused by communication interruption of temperature detection devices, invalid data generated during the calibration of calorific value detection devices, jump data caused by abnormal valve feedback, and abnormal data generated during the shutdown of the internal combustion engine generator set. Time alignment refers to unifying data from different sources, with different sampling periods and different transmission delays onto the same time base, so that pressure data, temperature data, calorific value data, valve opening data, and unit status data at the same time segment can correspond to the same operating state. Stable operating condition screening refers to filtering data segments that reflect normal and stable gas supply relationships based on the operating status of the internal combustion engine generator set, load change trends, valve action status, and gas pressure fluctuation status.

[0043] In practice, the control device first reads historical operating data from the historical database and adds a data source identifier and timestamp to each piece of historical operating data according to its source. For data with obvious gaps, the control device determines whether the data can be reasonably supplemented based on data from adjacent time periods; if it can be supplemented, it performs trend supplementation based on adjacent valid data; if it cannot be supplemented, the time period corresponding to the data is marked as invalid. For pressure and temperature data, the control device identifies abnormal jumps based on instrument range, historical operating range, and the variation amplitude of adjacent sampling points. For calorific value data, the control device excludes abnormal detection data based on the maintenance records, calibration records, and communication status records of the calorific value detection device. For valve opening data, the control device determines the validity of the valve data based on the consistency between the valve opening command and the valve opening feedback.

[0044] After anomaly removal, the control unit performs time alignment of historical operating data according to a unified time reference. Since pressure and temperature detection devices typically respond quickly, while calorific value detection devices have detection and analysis delays, and there is also a delivery delay from the valve position to the online calorific value detection device for the mixed gas after valve action, the control unit does not simply pair data from the same moment during time alignment. Instead, it performs time correction based on the temporal relationships between valve action events, pressure change events, and calorific value response events in the historical operating data. For data with short sampling periods, the control unit can resample according to the unified time reference; for data with long sampling periods, the control unit can interpolate or hold based on adjacent valid data; for significantly lagging calorific value data, the control unit can correct it to the corresponding preceding operating condition time point based on historical correlation.

[0045] After time alignment, the control unit performs stable operating condition screening. During stable operating condition screening, the control unit excludes data from the initial startup phase of the internal combustion engine generator set, the shutdown process, abnormal combustion processes, drastic load adjustments, and upstream gas supply interruptions. For the retained data, the control unit further determines whether the pressure change trend, temperature change trend, and valve opening change trend are suitable for modeling. When pressure and temperature data exhibit drastic abnormal fluctuations and cannot establish a stable correspondence with calorific value changes, the corresponding data can be marked as disturbed operating condition samples rather than stable operating condition samples; when there is a traceable relationship between pressure, temperature, valve opening, and the actual calorific value of the mixed gas, the corresponding data is used as valid historical operating data. Valid historical operating data serves as the basis for subsequent construction of operating condition samples.

[0046] S12. Construct a working condition sample based on valid historical operating data to correlate pressure, temperature and gas calorific value.

[0047] The operating condition sample refers to the data sample formed by processing valid historical operating data and used to train the first and second prediction models. The operating condition sample includes at least input features and label data. Input features include pressure-related features, temperature-related features, pressure-temperature coupling features, valve operation features, and unit status features. Label data includes blast furnace gas calorific value labels, converter gas calorific value labels, or corrected gas calorific value estimation labels. For operating periods where single-channel gas calorific value labels cannot be directly obtained, inversion can be performed using the actual calorific value of the mixed gas, valve opening status, and gas pipeline status to obtain calorific value estimation labels for training.

[0048] In practice, the control device performs windowing processing on valid historical operating data. Windowing processing means using pressure and temperature data over a continuous time period as a sample unit, rather than using pressure and temperature data at a single moment as a sample unit. For each sample unit, the control device extracts the average pressure, direction of pressure change, amplitude of pressure fluctuation, rate of pressure change, average temperature, direction of temperature change, amplitude of temperature fluctuation, rate of temperature change, and the equivalent density feature calculated jointly from the pressure and temperature data. The equivalent density feature is used to characterize the impact of changes in gas state on the energy contribution and transport status of a unit volume of gas under the same pipeline conditions.

[0049] When constructing the operating condition sample for blast furnace gas, the control device reads blast furnace gas pressure data, blast furnace gas temperature data, blast furnace gas regulating valve data, and relevant calorific value labels to form the blast furnace gas operating condition sample. Similarly, when constructing the operating condition sample for converter gas, the control device reads converter gas pressure data, converter gas temperature data, converter gas regulating valve data, and relevant calorific value labels to form the converter gas operating condition sample. Because blast furnace gas and converter gas differ in their sources, calorific value ranges, dust characteristics, and fluctuation patterns, the control device constructs separate operating condition samples for blast furnace gas and converter gas, rather than completely mixing the two types of gas to train the same model. Therefore, the first prediction model can learn the relationship between the pressure and temperature state of blast furnace gas and its calorific value, while the second prediction model can learn the relationship between the pressure and temperature state of converter gas and its calorific value.

[0050] Furthermore, the control device can add operating condition labels to each operating condition sample. These labels describe the corresponding gas supply condition, such as stable gas supply, pressure rise, pressure fall, temperature drift, pressure disturbance, rapid calorific value change, or other abnormal disturbance conditions. Operating condition labels can be automatically generated using historical operating rules, or generated jointly from unsupervised clustering results and manual verification results. These labels are used for subsequent operating condition identification and training of the expert prediction sub-model.

[0051] S13. Construct and train the first and second prediction models based on the working condition samples.

[0052] In this embodiment, both the first and second prediction models include a condition identification section, an expert prediction section, a property constraint section, and an uncertainty output section. The condition identification section identifies the current gas supply condition based on the condition sample and outputs the condition weight. The expert prediction section includes multiple expert prediction sub-models, each corresponding to a different gas supply condition. The property constraint section constrains the model output based on the gas type, calorific value variation law, and pressure and temperature disturbance state. The prediction uncertainty characterizes the reliability of the current model output and is used to subsequently generate calorific value confidence boundaries.

[0053] See attached document Figure 2 Specifically, S13 includes S131, identifying the gas supply conditions based on the operating condition samples and obtaining the operating condition weights.

[0054] Among them, the gas supply condition refers to the supply status of blast furnace gas or converter gas within the current time window. The gas supply condition can be characterized by pressure changes, temperature changes, pressure-temperature coupling changes, valve opening status, and unit load status. The condition weight refers to the weight value of the current condition sample belonging to different gas supply conditions. The condition weight is used for subsequent fusion of multiple expert-predicted calorific values. Unlike simple single classification labels, the condition weight can express the transitional state between multiple conditions, avoiding abrupt changes in prediction results due to hard switching during condition transitions.

[0055] In practical implementation, the control device inputs operating condition samples into the operating condition identification section. The operating condition identification section can employ rule-based state machines, clustering models, classification models, or gating networks. For samples with relatively stable pressure and temperature changes, the operating condition identification section outputs a higher operating condition weight corresponding to a stable gas supply condition. For samples where pressure rises continuously over a short period and temperature changes simultaneously, the operating condition identification section outputs a higher operating condition weight corresponding to a pressure rise condition. For samples where pressure drops accompanied by changes in pipeline pressure drop, the operating condition identification section outputs a higher operating condition weight corresponding to a pressure drop condition or a pressure disturbance condition. For samples where temperature shifts slowly but pressure changes are not significant, the operating condition identification section outputs a higher operating condition weight corresponding to a temperature drift condition. For samples where the calorific value may change rapidly during converter gas recovery, the operating condition identification section can output a higher operating condition weight corresponding to a rapid change in calorific value condition.

[0056] During the training phase, the operating condition identification component learns the mapping relationship between pressure and temperature characteristics and gas supply conditions based on historical operating condition labels. During the operation phase, the operating condition identification component outputs corresponding operating condition weights based on the real-time generated temperature and pressure time-series characteristics. These operating condition weights serve as inputs for subsequent expert-predicted calorific value fusion; they are not directly used as the final calorific value output, nor do they directly generate valve control commands.

[0057] S132. Input the operating condition sample into the expert prediction sub-model corresponding to the gas supply operating condition to obtain the expert predicted calorific value.

[0058] Among them, the expert prediction sub-model refers to the calorific value prediction sub-model established for specific gas supply conditions. Since the relationship between pressure, temperature, and calorific value is not entirely the same for the same type of gas under different gas supply conditions, multiple expert prediction sub-models are used to learn the mapping relationship under different conditions. The expert predicted calorific value refers to the gas calorific value prediction result output by each expert prediction sub-model based on the operating condition sample.

[0059] In practice, the control device trains multiple expert prediction sub-models for blast furnace gas based on blast furnace gas operating condition samples, and trains multiple expert prediction sub-models for converter gas based on converter gas operating condition samples. For blast furnace gas, the expert prediction sub-models can correspond to stable gas supply conditions, pressure fluctuation conditions, temperature drift conditions, and low calorific value risk conditions, respectively. For converter gas, the expert prediction sub-models can correspond to stable gas supply conditions, gas recovery stage change conditions, cabinet pressure disturbance conditions, and high calorific value and high dust content risk conditions, respectively. Each expert prediction sub-model receives pressure and temperature-related input features and outputs the expert predicted calorific value for the corresponding operating condition.

[0060] The expert prediction sub-model can employ regression models, tree models, neural network models, time series models, or other prediction models suitable for industrial process data. During implementation, to avoid the black-box model outputting uninterpretable results under abnormal operating conditions, the training data for the expert prediction sub-model needs to undergo processing in steps S11 and S12, and input feature importance verification, sample coverage verification, and output range verification are incorporated during training. After training, each expert prediction sub-model can effectively express the nonlinear relationship between pressure and temperature time series characteristics and the calorific value of the gas under its corresponding gas supply conditions.

[0061] S133. Based on the weighted integration of expert-predicted calorific values ​​under operating conditions, the model output rules of the first and second prediction models are obtained.

[0062] The model output rule refers to the rules by which the first and second prediction models generate their outputs based on the operating condition weights and expert-predicted heat values ​​during runtime. The model output rule does not simply select the result of one expert prediction sub-model, but rather fuses multiple expert-predicted heat values ​​according to the operating condition weights. Through this fusion, the first and second prediction models can maintain smooth output during operating condition switching, while simultaneously highlighting the role of the corresponding expert prediction sub-model under specific operating conditions.

[0063] In practice, for each operating condition sample, the control device first outputs multiple operating condition weights from the operating condition identification section, and then inputs the operating condition sample into multiple expert prediction sub-models to obtain multiple expert predicted calorific values. The control device weights and fuses these multiple expert predicted calorific values ​​according to the operating condition weights to obtain an initial predicted calorific value. For the first prediction model, this initial predicted calorific value corresponds to the initial predicted calorific value of blast furnace gas. For the second prediction model, this initial predicted calorific value corresponds to the initial predicted calorific value of converter gas. Subsequently, the control device inputs the initial predicted calorific value into the physical property constraint section for correction, and uses the dispersion among the multiple expert predicted calorific values, the distribution of operating condition weights, and the historical prediction residuals as the basis for calculating the prediction uncertainty.

[0064] In the above manner, both the first and second prediction models can output not only the predicted calorific value but also the prediction uncertainty. A higher prediction uncertainty indicates that the operating condition sample corresponding to the current temperature and pressure time series characteristics has fewer similar samples in historical data, larger differences in outputs among multiple expert prediction sub-models, or larger recent calorific value residuals. Conversely, a lower prediction uncertainty indicates higher reliability of the current prediction result. The prediction uncertainty is used in subsequent step S2 to generate the calorific value confidence boundary and in step S3 for robust determination of the target blending ratio.

[0065] S14. Collect real-time pressure and temperature data of blast furnace gas and converter gas, and generate temperature and pressure time series characteristics.

[0066] The real-time pressure and temperature data include real-time pressure data for blast furnace gas, real-time temperature data for blast furnace gas, real-time pressure data for converter gas, and real-time temperature data for converter gas. The real-time pressure and temperature data for blast furnace gas are collected by pressure and temperature detection devices on the blast furnace gas pipeline, and the same applies to converter gas. The time-series temperature and pressure characteristics are the set of input features obtained by fusing the real-time pressure and temperature data with their variation characteristics within a preset time window.

[0067] Specifically, S14 includes S141, collecting real-time pressure and temperature data of blast furnace gas and converter gas.

[0068] See attached document Figure 3In practice, the control device reads real-time data from the pressure and temperature detection devices according to a preset acquisition cycle. For blast furnace gas, the control device reads real-time pressure and temperature data, and adds a timestamp and pipeline identifier to these data. For converter gas, the control device reads real-time pressure and temperature data, and adds a timestamp and pipeline identifier to these data. The control device can also read feedback from the blast furnace gas regulating valve and the converter gas regulating valve to confirm the current actual blending status.

[0069] During data acquisition, the control device assesses the validity of real-time pressure and temperature data. If a pressure or temperature sensor experiences communication failure, data freeze, or significant out-of-range behavior, the control device can perform short-term compensation based on data from adjacent time points and the corresponding pipeline's operating status, generating a data quality flag. This data quality flag can serve as a factor influencing subsequent prediction uncertainty. When real-time pressure and temperature data remain abnormal, the control device can enter a degraded control mode and send the abnormal status to the operation monitoring system.

[0070] S142. Generate pressure change characteristics, temperature change characteristics, and equivalent density characteristics based on real-time pressure data and real-time temperature data.

[0071] Among them, pressure change characteristics refer to the features formed by the changing trend, rate of change, fluctuation amplitude, and stability of real-time pressure data within a time window. Temperature change characteristics refer to the features formed by the changing trend, rate of change, fluctuation amplitude, and stability of real-time temperature data within a time window. Equivalent density characteristics refer to the features used to characterize the state of coal gas, calculated based on real-time pressure and real-time temperature data. Equivalent density characteristics do not require an accurate representation of the true density of coal gas, but rather are used to characterize the impact of coupled pressure and temperature changes on the state and transport capacity of coal gas under the same pipeline conditions.

[0072] In practice, the control device writes the collected real-time pressure and temperature data into a circulating buffer. The circulating buffer stores pressure and temperature data from the most recent control cycles in chronological order. The control device reads the real-time pressure and temperature data of blast furnace gas, converter gas, and converter gas within the current time window from the circulating buffer, and calculates the direction of pressure change, amplitude of pressure change, rate of pressure change, degree of pressure fluctuation, direction of temperature change, amplitude of temperature change, rate of temperature change, and degree of temperature fluctuation for each.

[0073] For blast furnace gas, the control device generates pressure change characteristics, temperature change characteristics, and equivalent density characteristics based on real-time pressure and temperature data. For converter gas, the control device generates the same characteristics. For cases with significant abrupt changes in pressure and temperature, the control device can also generate pressure abrupt change indicators and temperature disturbance indicators. These indicators are used for subsequent operating condition identification and property constraint correction.

[0074] S143. By integrating real-time pressure data, real-time temperature data, pressure change characteristics, temperature change characteristics, and equivalent density characteristics, temperature-pressure time series characteristics are obtained.

[0075] In specific implementation, the control device integrates the real-time pressure data, real-time temperature data, pressure change characteristics, temperature change characteristics, and equivalent density characteristics of blast furnace gas into a time-series characteristic of blast furnace gas temperature and pressure according to a preset characteristic order; and integrates the real-time pressure data, real-time temperature data, pressure change characteristics, temperature change characteristics, and equivalent density characteristics of converter gas into a time-series characteristic of converter gas temperature and pressure according to a preset characteristic order.

[0076] In the model input stage, the time-series characteristics of blast furnace coal temperature and pressure can be input into the first prediction model, and the time-series characteristics of converter coal temperature and pressure can be input into the second prediction model. To maintain consistency in the terminology of "temperature and pressure time-series characteristics," the temperature and pressure time-series characteristics referred to in this embodiment include both blast furnace coal temperature and pressure time-series characteristics and converter coal temperature and pressure time-series characteristics. The temperature and pressure time-series characteristics are used as input in step S2 to generate blending parameters.

[0077] Through step S1, this embodiment obtains effective historical operating data, a first prediction model, a second prediction model, and temperature and pressure time-series characteristics. The first prediction model, the second prediction model, and the temperature and pressure time-series characteristics are used in step S2; the effective historical operating data are used in step S3 to generate the equivalent dust load.

[0078] S2. Input the temperature and pressure time series characteristics into the first prediction model and the second prediction model to obtain the blending parameters; the blending parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary.

[0079] In this embodiment, the blending parameters refer to the set of parameters used to determine the target blending ratio of blast furnace gas and converter gas. The blending parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary. The predicted calorific value of blast furnace gas refers to the calorific value prediction result of blast furnace gas output by the first prediction model based on the corresponding temperature and pressure time series characteristics of blast furnace gas and corrected by physical property constraints. The predicted calorific value of converter gas refers to the calorific value prediction result of converter gas output by the second prediction model based on the corresponding temperature and pressure time series characteristics of converter gas and corrected by physical property constraints. The prediction uncertainty is a quantitative representation of the reliability of the current prediction result by the first and second prediction models. The calorific value confidence boundary refers to the upper and lower boundaries of the calorific value generated based on the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, and the prediction uncertainty. The calorific value confidence boundary is used to account for prediction errors in subsequent calculations of the target blending ratio, avoiding insufficient calorific value of the mixed gas or excessive blending of converter gas due to model prediction deviations.

[0080] See attached document Figure 4 In this embodiment, S2 specifically includes S21, where the first prediction model and the second prediction model output the initial predicted calorific value and prediction uncertainty based on the temperature and pressure time series characteristics.

[0081] The initial predicted calorific value refers to the calorific value prediction result output by the first or second prediction model before the physical property constraint correction is performed. The prediction uncertainty can be obtained by combining multiple factors, such as the dispersion of the output results of the expert prediction sub-model, the concentration of the operating condition weight distribution, the similarity between the current temperature and pressure time series characteristics and historical operating condition samples, the magnitude of the recent calorific value residual, and the data quality indicators of the real-time pressure and temperature data.

[0082] In practice, the control device inputs the temperature and pressure time-series characteristics of the blast furnace gas into the first prediction model. The operating condition identification part of the first prediction model outputs blast furnace gas operating condition weights based on the temperature and pressure time-series characteristics. Multiple expert prediction sub-models within the first prediction model output multiple expert-predicted calorific values ​​for the blast furnace gas based on the temperature and pressure time-series characteristics. The first prediction model integrates these multiple expert-predicted calorific values ​​based on the blast furnace gas operating condition weights to obtain the initial predicted calorific value for the blast furnace gas. Simultaneously, the first prediction model obtains the prediction uncertainty for the blast furnace gas based on the differences between the multiple expert-predicted calorific values, the distribution of blast furnace gas operating condition weights, and the residual historical calorific value of the blast furnace gas.

[0083] Similarly, the control device inputs the temperature and pressure time-series characteristics of the converter gas into the second prediction model. The operating condition identification component of the second prediction model outputs converter gas operating condition weights based on the temperature and pressure time-series characteristics. Multiple expert prediction sub-models within the second prediction model output multiple expert-predicted calorific values ​​for the converter gas based on the temperature and pressure time-series characteristics. The second prediction model integrates these multiple expert-predicted calorific values ​​based on the converter gas operating condition weights to obtain the initial predicted calorific value for the converter gas. Simultaneously, the second prediction model obtains the prediction uncertainty for the converter gas based on the differences between the multiple expert-predicted calorific values, the distribution of converter gas operating condition weights, and the residual historical calorific value of the converter gas.

[0084] In this process, if the current temperature and pressure time series characteristics belong to a condition that is rarely seen in historical data, or if the outputs of multiple expert prediction sub-models differ significantly, the prediction uncertainty will increase accordingly; if the current temperature and pressure time series characteristics belong to a stable condition and the outputs of multiple expert prediction sub-models are relatively consistent, the prediction uncertainty will decrease accordingly. The prediction uncertainty is used in step S23 to generate the calorific value confidence boundary.

[0085] S22. Based on the type of gas, the trend of calorific value change, and the pressure and temperature disturbance state, the initial predicted calorific value is corrected by physical property constraints to obtain the predicted calorific value of blast furnace gas and the predicted calorific value of converter gas.

[0086] Among them, the gas type is used to distinguish between blast furnace gas and converter gas. Because blast furnace gas and converter gas differ in their calorific value range, fluctuation patterns, and dust characteristics, their physical property constraint rules differ. The calorific value change trend refers to the direction and magnitude of change in the initially predicted calorific value within adjacent control cycles. The pressure and temperature disturbance state refers to the state of the current gas condition, characterized by pressure sudden change indicators, temperature disturbance indicators, pressure change characteristics, and temperature change characteristics, indicating whether a disturbance has occurred.

[0087] Property constraint correction refers to the rationality assessment and adjustment of the initial predicted calorific value, ensuring that the final predicted calorific values ​​of blast furnace gas and converter gas conform to the calorific value range, continuity of change, and operating condition disturbance patterns corresponding to the gas type. Property constraint correction is not a simple fixed limit, but rather a dynamic determination of the correction method based on the calorific value change trend and pressure and temperature disturbance states. For example, when the pressure and temperature disturbance states indicate a stable gas state, the property constraint part can limit excessively large abrupt changes in the initial predicted calorific value between adjacent control cycles; when the pressure and temperature disturbance states indicate significant gas state disturbances, the property constraint part can relax the calorific value change restrictions, allowing the model to reflect the actual calorific value changes in a timely manner.

[0088] In practice, the control device first reads the corresponding physical property constraint rules based on the type of gas. For blast furnace gas, the physical property constraint rules include the reasonable range of calorific value for blast furnace gas, the requirement for continuous change of calorific value for blast furnace gas, and correction strategies under pressure and temperature disturbances. For converter gas, the physical property constraint rules include the reasonable range of calorific value for converter gas, the requirement for continuous change of calorific value for converter gas, and correction strategies under changes in converter gas during the recovery stage or under pressure disturbances.

[0089] Subsequently, the control device determines whether the current initial predicted calorific value deviates from the reasonable range corresponding to the type of gas. If the initial predicted calorific value deviates from the reasonable range, the control device makes corrections based on the prediction uncertainty and pressure and temperature disturbance states. If the prediction uncertainty is high and the pressure and temperature disturbance states do not support sudden changes in calorific value, the control device reverts the initial predicted calorific value back to the historical stable range. If the pressure and temperature disturbance states indicate a significant change in the gas state, the control device allows the initial predicted calorific value to change rapidly within the allowable range of physical property constraints. For prediction results of adjacent control cycles, the control device can also smooth the results based on the trend of calorific value changes to avoid frequent changes in valve control commands due to fluctuations in prediction results.

[0090] After correction for physical property constraints, the first prediction model outputs the predicted calorific value of blast furnace gas, and the second prediction model outputs the predicted calorific value of converter gas. The predicted calorific values ​​of blast furnace gas and converter gas serve as the basis for generating the calorific value confidence boundary in step S23, and as the core input for determining the target blending ratio in step S3.

[0091] S23. Generate calorific value confidence boundaries based on the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, and the prediction uncertainty; combine them to obtain the blending parameters.

[0092] The calorific value confidence boundaries include those for blast furnace gas and converter gas. The blast furnace gas calorific value confidence boundary characterizes the credible range within which the predicted calorific value of blast furnace gas might fall under the influence of prediction errors. The converter gas calorific value confidence boundary characterizes the credible range within which the predicted calorific value of converter gas might fall under the influence of prediction errors. The calorific value confidence boundaries can include an upper boundary and a lower boundary. The lower boundary ensures the safety of the calorific value of the mixed gas, while the upper boundary prevents excessively high calorific values ​​in the mixed gas and excessive blending of converter gas.

[0093] In practice, the control device generates confidence boundaries for the calorific value of blast furnace gas based on the predicted calorific value and the corresponding prediction uncertainty of blast furnace gas, and generates confidence boundaries for the calorific value of converter gas based on the predicted calorific value and the corresponding prediction uncertainty of converter gas. When the prediction uncertainty is large, the range corresponding to the calorific value confidence boundary is wide, indicating that the current calorific value prediction has high uncertainty; when the prediction uncertainty is small, the range corresponding to the calorific value confidence boundary is narrow, indicating that the current calorific value prediction is relatively reliable.

[0094] Subsequently, the control device combines the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary into blending parameters. These blending parameters are written into the blending calculation buffer of the control device and, in step S3, are used together with the equivalent dust load and the preset target blending calorific value to determine the target blending ratio.

[0095] In step S2, the blending parameters are obtained in this embodiment. The blending parameters are used in step S3 to calculate the target blending ratio, and in step S4 to generate a feedforward predicted blending calorific value sequence.

[0096] S3. Generate the equivalent dust load based on historical operating data, and determine the target blending ratio based on the equivalent dust load, blending parameters and preset target calorific value.

[0097] In this embodiment, the equivalent dust load refers to a quantitative characterization of the potential dust accumulation risk caused by converter gas blending to the mixed gas pipeline, burner, and internal combustion engine generator set. Because converter gas has a relatively high dust content, excessive blending to increase the calorific value of the mixed gas can easily lead to an increase in the mixed gas dust load, resulting in dust accumulation in pipelines, burner blockage, equipment wear, and increased operation and maintenance costs. Since reliable online dust detection devices are often not installed on-site, this embodiment generates the equivalent dust load based on historical operating data. Historical operating data can be data stored prior to the control cycle and continuously updated, including converter gas blending status, converter gas regulating valve opening, converter gas pressure and temperature status, mixed gas pipeline pressure drop trend, ash cleaning and purging records, and equipment operating status data.

[0098] The target mixed calorific value refers to the preset target calorific value of the mixed gas suitable for the stable and economical operation of the internal combustion engine generator set. The target blending ratio refers to the target mixing ratio of blast furnace gas and converter gas determined by the control device, which is used to generate valve control commands subsequently. The target blending ratio includes the target ratio of blast furnace gas and the target ratio of converter gas. In order to maintain consistency in technical logic, the target blending ratio in this embodiment satisfies the calorific value control requirements while also being constrained by the equivalent dust load.

[0099] In this embodiment, S3 specifically includes S31, determining the converter gas demand range based on the calorific value confidence boundary and the target mixed calorific value.

[0100] The converter gas demand range refers to the required proportion of converter gas in the blending process to achieve the target calorific value or meet calorific value safety conditions. Since both the predicted calorific values ​​of blast furnace gas and converter gas have prediction uncertainties, using only a single-point prediction value to calculate the converter gas proportion may result in the actual calorific value of the blended gas being lower or higher than the target calorific value due to prediction errors. Therefore, this embodiment determines the converter gas demand range based on the calorific value confidence boundary.

[0101] In practice, the control device reads the calorific value confidence boundary obtained in step S2 and the preset target mixed calorific value. Based on the calorific value confidence boundaries of blast furnace gas and converter gas, the control device calculates the minimum converter gas demand boundary required to meet the calorific value safety conditions, and the maximum converter gas demand boundary corresponding to avoiding excessively high calorific value of the mixed gas. The minimum converter gas demand boundary is used to prevent insufficient calorific value of the mixed gas, while the maximum converter gas demand boundary is used to prevent excessive blending of converter gas and excessively high calorific value of the mixed gas. Through the minimum and maximum demand boundaries, the control device obtains the converter gas demand range.

[0102] When the confidence boundary of blast furnace gas calorific value indicates that blast furnace gas may be in a lower calorific value state, the overall demand range for converter gas shifts towards a higher proportion of converter gas. When the confidence boundary of converter gas calorific value indicates that the uncertainty of the predicted calorific value of converter gas is large, the control device can increase the calorific value safety margin to avoid insufficient calorific value of the mixed gas due to the actual calorific value of converter gas being lower than the predicted calorific value. When the predicted calorific value of blast furnace gas is high and the prediction uncertainty is low, the demand range for converter gas can be narrowed accordingly to reduce the unnecessary blending of high-dust-content converter gas.

[0103] The range of converter gas demand is used in step S33 to determine the target blending ratio.

[0104] S32. Generate the equivalent dust load based on historical operating data, and determine the allowable range of converter gas based on the equivalent dust load.

[0105] The permissible range for converter gas refers to the range of proportions of converter gas allowed to be blended under current dust accumulation risk conditions. Unlike the demand range for converter gas, which focuses on calorific value safety, the permissible range for converter gas emphasizes dust constraints and equipment protection. A higher equivalent dust load indicates a greater risk of dust accumulation in the system, and the tighter the permissible range for converter gas; a lower equivalent dust load indicates a lower risk of dust accumulation in the system, and the permissible range for converter gas can be appropriately relaxed.

[0106] In practice, the control device reads data related to dust accumulation from historical operating data. This dust accumulation-related data includes historical target blending ratios of converter gas, historical actual blending ratios, historical opening degrees of converter gas regulating valves, converter gas pressure and temperature status, pressure drop trends in the mixed gas pipeline, pressure difference trends before and after the burner, ash cleaning and purging records, and maintenance records. The control device generates an equivalent dust load based on this data.

[0107] In one implementation, the control device first generates a basic dust accumulation amount based on the historical mixing time and proportion of converter gas. Subsequently, the control device corrects the basic dust accumulation amount based on the trend of pipeline pressure drop changes. If the pipeline pressure drop continues to increase, it indicates a potential dust accumulation trend in the pipeline or filter components, and the equivalent dust load increases accordingly. If the cleaning, purging, or maintenance records show that the system has completed dust removal, the control device reduces the equivalent dust load based on the cleaning records. Furthermore, the control device can also estimate the equivalent converter gas throughput based on the converter gas pressure and temperature status, and use the equivalent converter gas throughput as an influencing factor on the equivalent dust load. Therefore, even if no online dust detection device is installed on-site, existing historical operating data can be used to form a dust risk characterization.

[0108] After obtaining the equivalent dust load, the control device determines the permissible range of converter gas based on the equivalent dust load. When the equivalent dust load is lower than the preset dust risk state, the permissible range of converter gas can be wider to ensure flexibility in calorific value adjustment. When the equivalent dust load is close to the preset dust risk state, the permissible range of converter gas is tightened to reduce the continued large-scale mixing of converter gas. When the equivalent dust load is at a high risk state, the control device can prioritize reducing the permissible range of converter gas and determine whether dust overdraft control is necessary in subsequent steps, taking into account calorific value safety requirements.

[0109] The allowable range of converter gas is used in step S33 together with the demand range of converter gas to determine the target blending ratio.

[0110] S33. Determine the target blending ratio based on the converter gas demand range and the allowable range of converter gas.

[0111] The target blending ratio refers to the desired blending ratio of blast furnace gas and converter gas within the current control cycle. The target blending ratio must simultaneously consider the target mixed calorific value, calorific value confidence boundary, prediction uncertainty, equivalent dust load, valve regulation capacity, and process safety limitations. The target blending ratio is not simply calculated directly based on the predicted calorific values ​​of blast furnace gas and converter gas, but rather adaptively coordinated between the converter gas demand range and the allowable range.

[0112] In practice, the control device compares the converter gas demand range with the converter gas allowable range. If the converter gas demand range falls entirely within the converter gas allowable range, the control device can select a target blending ratio within this overlapping range that results in a predicted calorific value of the mixed gas close to the target mixed calorific value while minimizing converter gas consumption. If the converter gas demand range partially exceeds the converter gas allowable range, the control device determines the target blending ratio based on the principles of calorific value safety priority and dust constraint. If there is no effective overlap between the converter gas demand range and the converter gas allowable range, the control device further determines whether there is a risk of insufficient calorific value in the mixed gas and implements dust overdraft or dust compensation control based on the determination result.

[0113] See attached document Figure 5 Specifically, S33 includes S331, when the demand range of converter gas exceeds the allowable range of converter gas and there is a risk of insufficient calorific value of the mixed gas, the target blending ratio is determined according to the demand range of converter gas, and the dust overdraft amount is generated.

[0114] The risk of insufficient calorific value refers to the risk that, when blending converter gas within the current permissible range, the actual calorific value of the mixed gas may be lower than the calorific value required for stable operation of the internal combustion engine generator set or lower than the target mixed calorific value. Dust overdraft refers to the cumulative amount of dust risk generated after the blending ratio of converter gas is temporarily allowed to exceed the permissible range for calorific value safety reasons. Dust overdraft does not indicate that the actual dust level has exceeded the standard, but rather that the system has used up its future dust budget to ensure calorific value safety.

[0115] In practice, the control device first determines whether the calorific value of the mixed gas, calculated based on the calorific value confidence boundary, might still be lower than the target mixed calorific value or the safe calorific value if blending is performed at the upper limit of the converter gas's permissible range. If there is a risk of insufficient calorific value, the control device determines the target blending ratio according to the converter gas's demand range, rather than mechanically limiting it to the converter gas's permissible range. This avoids problems such as power reduction, unstable combustion, or inability to start the internal combustion engine generator set due to overly rigid dust constraints.

[0116] Simultaneously, the control device generates a dust overdraft amount based on the degree and duration of the target blending ratio exceeding the allowable range of converter gas, as well as the current equivalent dust load. This dust overdraft amount is stored in the dust budget recording area and used in subsequent control cycles to determine whether to initiate dust repayment control. In this way, this embodiment designs the dust constraint as a constraint with dynamic management capabilities, rather than a simple fixed-ratio limit.

[0117] S332. When the converter gas demand range and the converter gas allowable range meet the dust repayment conditions, the target blending ratio is reduced according to the dust overdraft amount to obtain the reduced target blending ratio.

[0118] The dust repayment condition refers to the conditions under which the current calorific value and dust load status of the blast furnace gas allow for a reduction in the converter gas blending ratio, thereby reducing subsequent dust accumulation. Dust repayment conditions may include situations such as an increase in the predicted calorific value of the blast furnace gas, a decrease in the demand range of the converter gas, a stable actual calorific value of the mixed gas, a reduction in prediction uncertainty, a high equivalent dust load, or the existence of dust overdraft. The target blending ratio after reduction refers to the target blending ratio formed after reducing the converter gas ratio to repay the dust overdraft, while meeting the calorific value safety conditions.

[0119] In practice, the control device checks the dust budget recording area for any dust overdraft in each control cycle. If a dust overdraft exists, the control device determines whether the current converter gas demand range and the converter gas allowable range have sufficient overlap, and whether the target mixed calorific value or calorific value safety conditions can still be met after reducing the converter gas ratio. If the dust repayment conditions are met, the control device selects a lower converter gas ratio within the converter gas demand range as the reduced target blending ratio. For operating conditions that require avoiding drastic valve operation, the control device can gradually reduce the target blending ratio according to the valve adjustment rate limit.

[0120] During dust recovery, the control device does not directly sacrifice calorific value safety to reduce dust risk. Instead, it reduces the proportion of converter gas while still meeting calorific value safety requirements. This balances the stable operation of the internal combustion engine generator set with long-term dust control.

[0121] S333. Update the equivalent dust load based on the reduced target blending ratio.

[0122] In practice, the control device updates the dust budget recording area based on the reduced target blending ratio. If the reduced target blending ratio decreases the amount of converter gas blended, the dust overdraft decreases according to the preset repayment rule. If the reduced target blending ratio continues to be implemented, the equivalent dust load will also decrease accordingly or stop increasing. The control device uses the updated equivalent dust load as the input for step S3 of the next control cycle.

[0123] If dust replenishment control is not triggered in step S332, the control device will still update the equivalent dust load based on the current target mixing ratio and historical operating data to ensure that the equivalent dust load continuously reflects the dust status of the system. The updated equivalent dust load can also be used for operation monitoring and maintenance early warning.

[0124] In step S3, the target blending ratio is obtained in this embodiment. The target blending ratio is used in step S4 to generate valve control commands and to generate a feedforward predicted blending calorific value sequence.

[0125] S4. Generate valve control commands based on the target blending ratio to adjust the actual blending ratio, and generate a feedforward predicted blending calorific value sequence based on the target blending ratio and blending parameters.

[0126] Among them, valve control commands refer to the control instructions sent by the control device to the blast furnace gas regulating valve and the converter gas regulating valve. Valve control commands may include the target opening degree of the blast furnace gas regulating valve, the target opening degree of the converter gas regulating valve, the valve opening adjustment direction, the valve opening adjustment rate, and the valve action holding state. The actual blending ratio refers to the actual blending ratio of blast furnace gas and converter gas formed in the mixed gas pipeline. The actual blending ratio can be estimated from valve opening feedback, pressure and temperature status, pipeline status, and historical calibration relationships. The feedforward predicted mixed calorific value sequence refers to the time series of predicted mixed gas calorific value generated by the control device based on the target blending ratio, blending parameters, and prediction results of continuous control cycles. The feedforward predicted mixed calorific value sequence is used in step S5 to perform correlation analysis with the actual mixed gas calorific value sequence and determine the dynamic feedback delay.

[0127] In practice, the control device first reads the target blending ratio determined in step S3. The target blending ratio includes the target ratio of blast furnace gas and the target ratio of converter gas. Based on the target ratio of blast furnace gas, the control device generates the target opening or target adjustment amount for the blast furnace gas regulating valve, and based on the target ratio of converter gas, it generates the target opening or target adjustment amount for the converter gas regulating valve. To avoid frequent and large valve movements, the control device can limit the rate of change of the valve control command based on the valve's mechanical characteristics, pipeline pressure status, and the valve opening in the previous control cycle. If the difference between the target blending ratio and the current actual blending ratio is small, the control device can generate a fine-tuning command or a holding command; if the difference is large, the control device can output valve control commands in stages to smoothly approach the target blending ratio.

[0128] After the valve control command is sent, the blast furnace gas regulating valve and the converter gas regulating valve execute the corresponding actions. The control device reads the opening feedback of the blast furnace gas regulating valve and the converter gas regulating valve, and determines whether the valve action conforms to the valve control command based on the valve feedback. If there is a deviation in the valve action, the control device can make secondary corrections based on the deviation, or write the abnormal valve status into the fault log. The actual blending ratio varies with the valve action and is affected by gas pressure, temperature, and pipeline resistance.

[0129] Subsequently, the control device generates a feedforward predicted calorific value based on the target blending ratio and blending parameters. Specifically, the control device calculates the feedforward predicted calorific value for the current control cycle based on the predicted calorific values ​​of blast furnace gas, converter gas, and the target blending ratio. Simultaneously, the control device stores the feedforward predicted calorific value for the current control cycle and the feedforward predicted calorific values ​​for previous control cycles in chronological order, forming a feedforward predicted calorific value sequence. Each element in the feedforward predicted calorific value sequence is appended with a corresponding control cycle timestamp, target blending ratio, predicted calorific value of blast furnace gas, predicted calorific value of converter gas, and prediction uncertainty.

[0130] The purpose of the feedforward predicted calorific value sequence is to provide a calorific value prediction trajectory that is independent of the delayed output of the online calorific value detection device for the mixed gas. Since the actual calorific value of the mixed gas detected by the online calorific value detection device often corresponds to the valve action and gas mixing state at a previous moment, rather than the valve action and gas mixing state at the current moment, the feedforward predicted calorific value sequence is needed in step S5 to identify the dynamic feedback delay. By saving the feedforward predicted calorific value sequence as a continuous time series, the control device can time-align the actual calorific value sequence of the mixed gas with the feedforward predicted calorific value sequence in subsequent steps.

[0131] In step S4, this embodiment outputs valve control commands to adjust the actual blending ratio and generates a feedforward predicted calorific value sequence. This feedforward predicted calorific value sequence is used in step S5 to determine the dynamic feedback delay and generate the calorific value residual.

[0132] S5. Obtain the actual calorific value sequence of the mixed gas, and determine the dynamic feedback delay by combining the feedforward predicted mixed calorific value sequence, and generate the calorific value residual based on the dynamic feedback delay.

[0133] Correlation refers to the degree of consistency between the feedforward predicted calorific value sequence and the actual calorific value sequence of the mixed gas under different time offsets. By comparing the correlation between the two sequences under different time offsets, the lag time of the current actual calorific value sequence of the mixed gas relative to the feedforward predicted calorific value sequence can be estimated. The dynamic feedback delay is the time offset that best matches the changing trend of the feedforward predicted calorific value sequence and the actual calorific value sequence of the mixed gas.

[0134] In practice, the control device reads the actual calorific value of the mixed gas from the online calorific value detection device and stores it in the calorific value feedback buffer according to the timestamp, forming a sequence of actual calorific values ​​of the mixed gas. Simultaneously, the control device reads the feedforward predicted mixed calorific value sequence from the feedforward predicted mixed calorific value buffer generated in step S4. Because the actual calorific value sequence of the mixed gas contains detection noise and analysis lag, the control device can first smooth the sequence to avoid single-point noise affecting the identification of dynamic feedback delay.

[0135] Subsequently, within a sliding time window, the control device performs correlation calculations on the actual calorific value sequence of the mixed gas and the feedforward predicted calorific value sequence under various time offsets. Under each time offset, the control device compares the changing trends of the actual calorific value sequence of the mixed gas with the corresponding offset feedforward predicted calorific value sequence to obtain the corresponding correlation evaluation value. The control device selects the time offset with the optimal correlation evaluation value as the dynamic feedback delay. If the correlation evaluation value is unstable or the correlations corresponding to multiple time offsets are close, the control device can combine pipeline pressure, valve opening, and historical dynamic feedback delays to make a smooth judgment, avoiding frequent jumps in the dynamic feedback delay.

[0136] By determining the dynamic feedback delay through correlation, the originally lagging actual calorific value sequence of the mixed gas is no longer directly used for current valve control. Instead, it is transformed into calibration information corresponding to the historical feedforward predicted mixed calorific value sequence. This avoids the problem of control oscillation caused by directly using the current actual calorific value for current valve action.

[0137] See attached document Figure 6 In this embodiment, S5 specifically includes S51, obtaining the actual calorific value sequence of the mixed gas, and determining the dynamic feedback delay based on the correlation between the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas.

[0138] Correlation refers to the degree of consistency between the feedforward predicted calorific value sequence and the actual calorific value sequence of the mixed gas under different time offsets. By comparing the correlation between the two sequences under different time offsets, the lag time of the current actual calorific value sequence of the mixed gas relative to the feedforward predicted calorific value sequence can be estimated. The dynamic feedback delay is the time offset that best matches the changing trend of the feedforward predicted calorific value sequence and the actual calorific value sequence of the mixed gas.

[0139] In practice, the control device reads the actual calorific value of the mixed gas from the online calorific value detection device and stores it in the calorific value feedback buffer according to the timestamp, forming a sequence of actual calorific values ​​of the mixed gas. Simultaneously, the control device reads the feedforward predicted mixed calorific value sequence from the feedforward predicted mixed calorific value buffer generated in step S4. Because the actual calorific value sequence of the mixed gas contains detection noise and analysis lag, the control device can first smooth the sequence to avoid single-point noise affecting the identification of dynamic feedback delay.

[0140] Subsequently, within a sliding time window, the control device performs correlation calculations on the actual calorific value sequence of the mixed gas and the feedforward predicted calorific value sequence under various time offsets. Under each time offset, the control device compares the changing trends of the actual calorific value sequence of the mixed gas with the corresponding offset feedforward predicted calorific value sequence to obtain the corresponding correlation evaluation value. The control device selects the time offset with the optimal correlation evaluation value as the dynamic feedback delay. If the correlation evaluation value is unstable or the correlations corresponding to multiple time offsets are close, the control device can combine pipeline pressure, valve opening, and historical dynamic feedback delays to make a smooth judgment, avoiding frequent jumps in the dynamic feedback delay.

[0141] By determining the dynamic feedback delay through correlation, the originally lagging actual calorific value sequence of the mixed gas is no longer directly used for current valve control. Instead, it is transformed into calibration information corresponding to the historical feedforward predicted mixed calorific value sequence. This avoids the problem of control oscillation caused by directly using the current actual calorific value for current valve action.

[0142] S52. Based on the dynamic feedback delay, the actual calorific value sequence of the mixed gas is time-aligned with the feedforward predicted mixed calorific value sequence to obtain aligned calorific value data.

[0143] Aligned calorific value data refers to data formed by pairing the feedforward predicted calorific value and the actual calorific value of the same batch of mixed gas after considering the dynamic feedback delay. Aligned calorific value data includes the aligned feedforward predicted calorific value, the corresponding actual calorific value of the mixed gas, the corresponding target blending ratio, the corresponding predicted calorific value of the blast furnace gas, the corresponding predicted calorific value of the converter gas, and the corresponding timestamp.

[0144] In practice, the control device, based on the dynamic feedback delay obtained in step S51, searches for the historical feedforward predicted calorific value corresponding to the current actual calorific value of the mixed gas in the feedforward predicted calorific value sequence. In other words, the current actual calorific value of the mixed gas is not directly compared with the current feedforward predicted calorific value, but rather compared with the feedforward predicted calorific value before the dynamic feedback delay. The control device combines the found historical feedforward predicted calorific value with the current actual calorific value of the mixed gas to form a set of aligned calorific value data. If the time point corresponding to the dynamic feedback delay falls between two feedforward predicted calorific values, the control device can interpolate based on adjacent feedforward predicted calorific values ​​to obtain the aligned feedforward predicted calorific value for the corresponding time point.

[0145] Simultaneously, the control device reads the target blending ratio and blending parameters corresponding to the generation of the aligned feedforward predicted calorific value. This ensures that when generating the calorific value residual later, the residual can be mapped to the actual predicted calorific values ​​of the blast furnace gas, converter gas, and target blending ratio involved in the mixing, thus avoiding residual attribution errors.

[0146] S53. Generate calorific value residuals based on the aligned calorific value data.

[0147] The calorific value residual refers to the difference or deviation between the actual calorific value of the aligned mixed gas and the aligned feedforward predicted calorific value. The calorific value residual reflects the prediction deviation resulting from the combined effects of the first prediction model, the second prediction model, the target blending ratio, and the actual mixing process. The calorific value residual can be positive or negative. When the calorific value residual indicates that the actual calorific value of the mixed gas is higher than the feedforward predicted calorific value, it suggests that the model or mixing process may have underestimated the calorific value of the gas; when the calorific value residual indicates that the actual calorific value of the mixed gas is lower than the feedforward predicted calorific value, it suggests that the model or mixing process may have overestimated the calorific value of the gas.

[0148] In practice, the control device calculates the calorific value residual based on the aligned calorific value data obtained in step S52. To avoid measurement noise from individual mixed gas actual calorific value data affecting model correction, the control device can smooth or screen the calorific value residuals across multiple consecutive control cycles. If the actual calorific value data of the mixed gas corresponding to a certain calorific value residual is marked as low confidence, the control device can reduce the impact of that calorific value residual on model correction in step S6. If multiple consecutive calorific value residuals have the same direction and their amplitude gradually increases, it indicates that the first or second prediction model may have a systematic bias, requiring bias correction in step S6.

[0149] Through step S5, this embodiment obtains the dynamic feedback delay and calorific value residual. The calorific value residual is used in step S6 to correct the first prediction model and the second prediction model.

[0150] S6. Based on the residual calorific value, correct the first and second prediction models, and use the corrected first and second prediction models for the next control cycle.

[0151] In this embodiment, S6 specifically includes S61, generating blast furnace gas calorific value bias and converter gas calorific value bias based on the calorific value residual.

[0152] In practice, the control device reads the calorific value residual generated in step S5, and simultaneously reads the corresponding target blending ratio, predicted calorific value of blast furnace gas, predicted calorific value of converter gas, prediction uncertainty, and operating condition weight from the aligned calorific value data. The control device determines whether the calorific value residual is primarily affected by the prediction error of blast furnace gas or the prediction error of converter gas based on the target blending ratio. If the target ratio of converter gas is high within the alignment time period, and the prediction uncertainty corresponding to the second prediction model is high, then a larger portion of the calorific value residual can be allocated to the converter gas calorific value bias. If the target ratio of blast furnace gas is high within the alignment time period, and the prediction uncertainty corresponding to the first prediction model is high, then a larger portion of the calorific value residual can be allocated to the blast furnace gas calorific value bias. If the prediction uncertainties of the first and second prediction models are similar, the control device can allocate the calorific value residual according to the target ratios of blast furnace gas and converter gas.

[0153] To avoid over-correction of bias due to a single calorific value residual, the control device can weight and fuse calorific value residuals from multiple consecutive control cycles to generate blast furnace gas calorific value biases and converter gas calorific value biases. Newer calorific value residuals can have higher weights, while earlier ones can have lower weights. If the direction of the calorific value residuals changes frequently, it indicates strong disturbances in the actual operating conditions or high noise in calorific value detection. In this case, the control device can reduce the bias correction magnitude and increase the prediction uncertainty. If the direction of the calorific value residuals remains consistent, it indicates a stability bias in the model. In this case, the control device can gradually increase the corresponding calorific value bias correction.

[0154] S62. The first prediction model is corrected based on the calorific value bias of blast furnace gas, and the second prediction model is corrected based on the calorific value bias of converter gas.

[0155] In practice, the control device writes the calorific value bias of blast furnace gas into the bias correction unit of the first prediction model and the calorific value bias of converter gas into the bias correction unit of the second prediction model. When the first prediction model outputs the predicted calorific value of blast furnace gas in the next control cycle, it will correct the model output based on the calorific value bias of the blast furnace gas. Similarly, when the second prediction model outputs the predicted calorific value of converter gas in the next control cycle, it will correct the model output based on the calorific value bias of the converter gas.

[0156] In addition to output bias correction, the control device can also update the prediction uncertainties of the first and second prediction models based on the calorific value residuals. When the calorific value residuals remain large, it indicates that the current model's prediction reliability is decreasing. The control device can increase the prediction uncertainty of the corresponding model, widening the calorific value confidence boundary generated in step S2, and enabling step S3 to adopt a more conservative strategy when determining the target blending ratio. When the calorific value residuals gradually decrease and stabilize within a small range, it indicates that the model correction is effective. The control device can gradually reduce the prediction uncertainty of the corresponding model, narrowing the calorific value confidence boundary and improving the economy and stability of converter gas blending control.

[0157] In some implementations, the control device can also store the calorific residual, aligned calorific data, and corresponding temperature and pressure time-series characteristics into the model update database. The model update database can be used for periodic retraining of the first and second prediction models, or for online updating of some expert prediction sub-models. To ensure system operational safety, online corrections prioritize bias correction and uncertainty correction. Periodic model retraining can be performed during non-critical runtime periods or in an offline environment. After training is complete, the updated model is deployed to the control device.

[0158] S63. Output the modified first prediction model and the modified second prediction model to the next control cycle.

[0159] In practice, the control device labels the first and second prediction models, after bias correction and prediction uncertainty correction, as the corrected first prediction model and the corrected second prediction model. The corrected first and second prediction models are stored in the model cache and invoked in step S2 of the next control cycle. In the next control cycle, the control device re-acquires real-time pressure and temperature data of blast furnace gas and converter gas, regenerates the temperature and pressure time series characteristics, and inputs the new temperature and pressure time series characteristics into the corrected first and second prediction models to obtain updated blending parameters.

[0160] Through the aforementioned closed loop, the calorific value residual will not directly cause a lag in the current valve control command, but will instead be used to correct the predictive model for the next control cycle. This allows for the utilization of real feedback information provided by the actual calorific value sequence of the mixed gas, while avoiding valve oscillation caused by lag in the online calorific value detection device for the mixed gas. As the control cycle continues to operate, the first and second predictive models can be continuously corrected based on actual operating data, and the target blending ratio can maintain a dynamic balance between calorific value safety, dust control, and operational economy.

[0161] This application also provides a mixed gas calorific value adaptive control system, which is applied to the method of any of the above embodiments.

[0162] See attached document Figure 7 An adaptive control system for the calorific value of mixed coal gas includes a data acquisition and modeling module 1, a model prediction module 2, a blending ratio determination module 3, a valve control module 4, a calorific value feedback module 5, and a model correction module 6.

[0163] The data acquisition and modeling module 1 is used to acquire historical operating data of blast furnace gas and converter gas, construct a first prediction model for blast furnace gas and a second prediction model for converter gas, and collect real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series characteristics.

[0164] Model prediction module 2 is connected to data acquisition and modeling module 1. It is used to input temperature and pressure time series characteristics into the first prediction model and the second prediction model to obtain mixing parameters. The mixing parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary.

[0165] The blending ratio determination module 3 is connected to the model prediction module 2. It is used to generate the equivalent dust load based on historical operating data, and determine the target blending ratio based on the equivalent dust load, blending parameters and preset target mixing calorific value.

[0166] The valve control module 4 is connected to the blending ratio determination module 3 and the model prediction module 2. It is used to generate valve control commands based on the target blending ratio to adjust the actual blending ratio, and to generate a feedforward predicted blending calorific value sequence based on the target blending ratio and blending parameters.

[0167] The calorific value feedback module 5 is connected to the valve control module 4 and is used to obtain the actual calorific value sequence of the mixed gas, determine the dynamic feedback delay based on the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas, and generate the calorific value residual based on the dynamic feedback delay.

[0168] The model correction module 6 is connected to the calorific value feedback module 5 and the data acquisition and modeling module 1 respectively. It is used to correct the first prediction model and the second prediction model based on the calorific value residual, and to use the corrected first prediction model and the corrected second prediction model in the next control cycle.

[0169] Unless otherwise defined, the technical or scientific terms used in this application shall have the ordinary meaning understood by one of ordinary skill in the art to which this application pertains. The terms "first," "second," "third," and similar terms used in this application specification and claims do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms "an" or "a" and similar terms do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms "comprising" or "including" and similar terms mean that the elements or objects preceding "comprising" or "including" encompass the elements or objects listed following "comprising" or "including" and their equivalents, and do not exclude other elements or objects. "Above," "below," "left," "right," etc., are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0170] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. A method for adaptive control of the calorific value of mixed gas, characterized in that: This includes S1, acquiring historical operating data of blast furnace gas and converter gas, constructing a first prediction model for blast furnace gas and a second prediction model for converter gas, and collecting real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series characteristics. S2. Input the temperature and pressure time series characteristics into the first prediction model and the second prediction model to obtain the mixing parameters; the mixing parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty, and the calorific value confidence boundary. S3. Generate an equivalent dust load based on the historical operating data, and determine the target blending ratio based on the equivalent dust load, the blending parameters, and the preset target mixing calorific value; S4. Generate valve control commands based on the target blending ratio to adjust the actual blending ratio, and generate a feedforward predicted blending calorific value sequence based on the target blending ratio and the blending parameters. S5. Obtain the actual calorific value sequence of the mixed gas, and determine the dynamic feedback delay based on the feedforward predicted mixed calorific value sequence, and generate the calorific value residual based on the dynamic feedback delay; S6. Based on the calorific value residual, correct the first prediction model and the second prediction model, and use the corrected first prediction model and the corrected second prediction model for the next control cycle.

2. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: The construction of the first prediction model for blast furnace gas and the second prediction model for converter gas specifically includes anomaly removal, time alignment, and stable operating condition screening of historical operating data to obtain effective historical operating data. Based on the aforementioned valid historical operating data, a sample of operating conditions relating pressure, temperature, and gas calorific value is constructed. The first and second prediction models are constructed and trained based on the operating condition samples.

3. The method for adaptive control of the calorific value of mixed gas according to claim 2, characterized in that: The step of constructing and training the first prediction model and the second prediction model based on the working condition samples specifically includes identifying the gas supply working conditions based on the working condition samples and obtaining the working condition weights. The operating condition sample is input into the expert prediction sub-model corresponding to the gas supply operating condition to obtain the expert predicted calorific value. Based on the operating condition weights and the expert-predicted heat values, the model output rules of the first prediction model and the second prediction model are obtained.

4. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: The process of collecting real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series characteristics specifically includes collecting real-time pressure data and real-time temperature data of blast furnace gas and converter gas. Based on the real-time pressure data and the real-time temperature data, pressure change characteristics, temperature change characteristics, and equivalent density characteristics are generated. By integrating the real-time pressure data, the real-time temperature data, the pressure change characteristics, the temperature change characteristics, and the equivalent density characteristics, a temperature-pressure time series characteristic is obtained.

5. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: S2 specifically includes the first prediction model and the second prediction model outputting the initial predicted calorific value and prediction uncertainty based on the temperature and pressure time series characteristics; Based on the type of gas, the trend of calorific value change, and the pressure and temperature disturbance state, the initial predicted calorific value is corrected by physical property constraints to obtain the predicted calorific value of blast furnace gas and the predicted calorific value of converter gas. The predicted calorific value of the blast furnace gas, the predicted calorific value of the converter gas, and the prediction uncertainty are used to generate a calorific value confidence boundary; these are then combined to obtain the blending parameters.

6. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: S3 specifically includes S31, determining the converter gas demand range based on the calorific value confidence boundary and the target mixed calorific value; S32. Generate an equivalent dust load based on the historical operating data, and determine the allowable range of converter gas based on the equivalent dust load; S33. Determine the target blending ratio based on the converter gas demand range and the converter gas allowable range.

7. The method for adaptive control of the calorific value of mixed gas according to claim 6, characterized in that: Specifically, S33 includes determining a target blending ratio based on the converter gas demand range and generating a dust overdraft amount when the converter gas demand range exceeds the allowable range and there is a risk of insufficient calorific value in the mixed gas. When the converter gas demand range and the converter gas allowable range meet the dust repayment conditions, the target blending ratio is reduced according to the dust overdraft amount to obtain the reduced target blending ratio. The equivalent dust load is updated based on the reduced target blending ratio.

8. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: S5 specifically includes obtaining the actual calorific value sequence of the mixed gas, and determining the dynamic feedback delay based on the correlation between the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas; Based on the dynamic feedback delay, the actual calorific value sequence of the mixed gas is time-aligned with the feedforward predicted mixed calorific value sequence to obtain aligned calorific value data; The calorific value residual is generated based on the aligned calorific value data.

9. The method for adaptive control of the calorific value of mixed gas according to claim 1, characterized in that: Specifically, S6 includes generating blast furnace gas calorific value offset and converter gas calorific value offset based on the calorific value residual; The first prediction model is corrected based on the calorific value bias of the blast furnace gas, and the second prediction model is corrected based on the calorific value bias of the converter gas. The modified first prediction model and the modified second prediction model are output to the next control cycle.

10. A mixed gas calorific value adaptive control system, characterized in that: It includes a data acquisition and modeling module (1), a model prediction module (2), a blending ratio determination module (3), a valve control module (4), a calorific value feedback module (5), and a model correction module (6); The data acquisition and modeling module (1) is used to acquire historical operating data of blast furnace gas and converter gas, construct a first prediction model corresponding to blast furnace gas and a second prediction model corresponding to converter gas, and acquire real-time pressure and temperature data of blast furnace gas and converter gas to generate temperature and pressure time series features. The model prediction module (2) is connected to the data acquisition and modeling module (1) and is used to input the temperature and pressure time series characteristics into the first prediction model and the second prediction model to obtain the mixing parameters; the mixing parameters include the predicted calorific value of blast furnace gas, the predicted calorific value of converter gas, the prediction uncertainty and the calorific value confidence boundary; The blending ratio determination module (3) is connected to the model prediction module (2) and is used to generate an equivalent dust load based on the historical operating data, and to determine the target blending ratio based on the equivalent dust load, the blending parameters and the preset target mixing calorific value. The valve control module (4) is connected to the blending ratio determination module (3) and is used to generate valve control commands according to the target blending ratio to adjust the actual blending ratio, and to generate a feedforward predicted blending calorific value sequence based on the target blending ratio and the blending parameters. The calorific value feedback module (5) is connected to the valve control module (4) and is used to obtain the actual calorific value sequence of the mixed gas, determine the dynamic feedback delay based on the feedforward predicted mixed calorific value sequence and the actual calorific value sequence of the mixed gas, and generate the calorific value residual based on the dynamic feedback delay. The model correction module (6) is connected to the calorific value feedback module (5) and the data acquisition and modeling module (1) respectively, and is used to correct the first prediction model and the second prediction model based on the calorific value residual, and use the corrected first prediction model and the corrected second prediction model for the next control cycle.

Images