Dry quenching process optimization system based on digital twinning and artificial intelligence
By optimizing the dry quenching system using digital twin and artificial intelligence technologies, the problems of comprehensive optimization of burn-off rate and steam production in the dry quenching system were solved, realizing efficient energy utilization and automated production control of the system, and improving production efficiency.
Patent Information
- Application Number
- CN202410071068.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-01-18
- Publication Date
- 2026-04-03
AI Technical Summary
Existing dry quenching systems struggle to achieve comprehensive optimization of dry quenching burn-off rate and steam output when optimizing production indicators, resulting in inefficient energy utilization.
A dry quenching process optimization system based on digital twins and artificial intelligence is adopted, including data acquisition, real-time calculation, prediction and intelligent control modules. Through data preprocessing, feature transformation, outlier removal and model verification, combined with multiple prediction models, the system can automatically adjust the parameters of the dry quenching furnace and waste heat boiler.
This system maximizes the overall benefits of the dry quenching system, reduces coal consumption, saves energy, improves production efficiency, and reduces manual intervention through automated control.
Smart Images
Figure CN121787208A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of digital twin and artificial intelligence technologies, and specifically to a dry quenching process optimization system based on digital twin and artificial intelligence. Background Technology
[0002] The main production indicators of a dry quenching system include the dry quenching loss rate and steam production. These two indicators have a certain correlation: a high dry quenching loss rate increases steam production, while a low dry quenching loss rate affects steam output. Simply reducing the dry quenching loss rate or increasing steam production alone cannot achieve global optimization. It is necessary to comprehensively consider the energy utilization system and, for the steam supply situation, coordinate the adjustment of the dry quenching loss rate and steam production to achieve optimal comprehensive utilization. Summary of the Invention
[0003] This invention relates to a dry quenching process optimization system based on digital twins and artificial intelligence, which can improve production efficiency and save energy.
[0004] To solve the above problems, the following technical solutions are provided:
[0005] The dry quenching process optimization system based on digital twins and artificial intelligence of the present invention is characterized by including a data acquisition module, a real-time calculation module, a prediction module, and an intelligent control module.
[0006] The data acquisition module includes a data acquisition unit and a data preprocessing unit. The data acquisition unit is used to acquire real-time data from the dry quenching furnace and the waste heat boiler and send it to the data preprocessing unit. The data preprocessing unit is used to remove outliers from the real-time data and send it to the real-time calculation module and the prediction module.
[0007] The real-time calculation module is used to calculate the actual burn-off rate and actual boiler heat exchange under the current parameters based on the real-time data after removing outliers, and then send the actual burn-off rate and actual boiler heat exchange to the prediction module.
[0008] The prediction module includes a feature transformation unit, a prediction model, an anomaly removal unit, a model validation unit, and a model selection and parameter tuning unit. The feature transformation unit transforms the real-time data after removing outliers and sends the transformed data to the anomaly removal unit. The anomaly removal unit removes outliers from the transformed data and sends the outlier-removed data to the model validation unit. The model validation unit includes a comparison model and multiple calculation models. Each calculation model calculates the burn-off rate and boiler heat exchange rate based on the outlier-removed data. The comparison model compares the calculated burn-off rate and calculated boiler heat exchange rate with the actual burn-off rate and actual boiler heat exchange rate, identifying the calculation models with the lowest deviation rates as base models. The model selection and parameter tuning unit assigns weights to each base model to form a composite model and adjusts the parameters of the composite model to obtain the prediction model. The prediction model obtains the predicted burn-off rate and predicted boiler heat exchange rate under the current parameters and sends them to the intelligent control module.
[0009] The intelligent control module includes a predictive comprehensive benefit unit and an adjustment unit. The predictive comprehensive benefit unit derives the predicted comprehensive benefit based on the predicted burn-off rate and predicted boiler heat exchange, and then sends the predicted comprehensive benefit to the adjustment unit. The adjustment unit adjusts the parameters of the dry quenching furnace and waste heat boiler in real time. When the predicted comprehensive benefit approaches the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is correct, and the adjustment continues in that direction. When the predicted comprehensive benefit deviates significantly from the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is incorrect, and the adjustment direction is changed. Simultaneously, the adjustment unit accumulates experience in the adjustment direction, reducing errors through training for more precise control.
[0010] The data preprocessing unit removes outliers by using the minimum, 25%, 50%, 75%, maximum, and mean values of the data distribution as statistical indicators. It evaluates the distribution of real-time data from three aspects: standard deviation, difference from the mean, and normal distribution. Extreme outliers are removed using the 5 sigma method.
[0011] The feature transformation unit uses Box-Cox transformation to transform the features of the real-time data after outlier removal.
[0012] The process of removing outliers from the transformed data by the anomaly removal unit is as follows: the predicted burn-off rate and predicted boiler heat exchange under the current parameters are obtained by using the model prediction, and the residual between the predicted burn-off rate and predicted boiler heat exchange and the actual burn-off rate and actual boiler heat exchange is used as the basis to remove outliers.
[0013] The model screening and parameter tuning unit uses grid search for parameter tuning.
[0014] The evaluation metrics for the grid search parameter tuning are: MSE, R2, and MAE.
[0015] The principle for allocating weights to the model screening and parameter tuning unit is: the base model with the smaller the deviation, the greater the weight it occupies, and the weights of all base models are 1.
[0016] The process by which the real-time calculation module calculates the actual burn-off rate is as follows:
[0017] Based on the principle of carbon conservation in the dry quenching system, the theoretical coke burn-off T is obtained by measuring the CO and CO2 content in the gas emitted after the dry quenching furnace blower. C ,
[0018]
[0019] Among them, V 放 V represents the amount of gas released after the blower. 放 Measured directly by sensors or calculated by formulas; This represents the volume percentage of CO in the circulating gas. This represents the volume percentage of CO2 in the circulating gas.
[0020] Based on the theoretical coke burn-off T C The theoretical dry quenching coke loss rate was obtained.
[0021]
[0022] Among them, A d It is coke ash.
[0023] The formula for predicting comprehensive return by the predicting comprehensive return unit is:
[0024]
[0025] Where W is the comprehensive return, t 蒸汽 P is the theoretical steam quantity. 蒸汽 For the price of steam, t 焦炭处理量 For coke processing capacity, L 烧损 P represents the dry quenching coke loss rate. 焦炭 This refers to the price of coke.
[0026] The above approach has the following advantages:
[0027] The dry quenching process optimization system based on digital twins and artificial intelligence of this invention includes a data acquisition module, a real-time calculation module, a prediction module, and an intelligent control module. This system can maximize overall benefits under different burn-off and steam volumes. Furthermore, by improving production efficiency, it can reduce coal consumption while ensuring profitability, thus saving energy. Attached Figure Description
[0028] Figure 1 This is a schematic diagram of the principle of the dry quenching process optimization system based on digital twins and artificial intelligence of the present invention;
[0029] Figure 2 This is a schematic diagram of the feature transformation of the dry quenching process optimization system based on digital twins and artificial intelligence in the embodiment.
[0030] Figure 3 This is a schematic diagram of the process of removing outliers after feature transformation in the dry quenching process optimization system based on digital twins and artificial intelligence in the embodiment.
[0031] Figure 4 This is a schematic diagram of the computing unit verification of the dry quenching process optimization system based on digital twins and artificial intelligence in the embodiment. Detailed Implementation
[0032] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments.
[0033] like Figure 1 As shown, the dry quenching process optimization system based on digital twins and artificial intelligence of the present invention includes a data acquisition module, a real-time calculation module, a prediction module, and an intelligent control module.
[0034] The data acquisition module contains a data acquisition unit and a data preprocessing unit. The data acquisition unit is used to collect real-time data from the dry quenching furnace and waste heat boiler and send it to the data preprocessing unit. The data preprocessing unit is used to remove outliers from the real-time data using statistical principles and then send it to the real-time calculation module and the prediction module.
[0035] The data preprocessing unit removes outliers using statistical principles as follows: It uses the minimum, 25%, 50%, 75%, maximum, and mean of the data distribution as statistical indicators. It evaluates the distribution of real-time data based on three aspects: standard deviation, difference from the mean, and normality. Extreme outliers are removed using the 5-sigma method. The 5-sigma method is relative to the 3-sigma method. For example, if the 3-sigma method defines a data range of 0.25-1, then the corresponding data range for the 5-sigma method needs to be larger than the 3-sigma range, say 0-1.2. The specific removal process is a common technique used by those skilled in the art and will not be elaborated here.
[0036] The real-time calculation module is used to calculate the actual burn-off rate and actual boiler heat exchange under the current parameters based on the real-time data after removing outliers, and then sends the actual burn-off rate and actual boiler heat exchange to the prediction module.
[0037] The process by which the real-time calculation module calculates the actual burn-off rate is as follows:
[0038] Based on the principle of carbon conservation in the dry quenching system, the theoretical coke burn-off T is obtained by measuring the CO and CO2 content in the gas emitted after the dry quenching furnace blower. C ,
[0039]
[0040] Among them, V 放 V represents the amount of gas released after the blower. 放 Measured directly by sensors or calculated by formulas; This represents the volume percentage of CO in the circulating gas. This represents the volume percentage of CO2 in the circulating gas.
[0041] Based on the theoretical coke burn-off T C The theoretical dry quenching coke loss rate was obtained.
[0042]
[0043] Among them, A d It is coke ash.
[0044] If there is no way to measure V 放 The sensor, then V 放 The calculation process is as follows:
[0045] Based on the local relative humidity, local atmospheric pressure, and saturated vapor pressure of water, the humidity H is calculated, and then the absolute dry air volume V is calculated. 干空气 and air moisture content V 空含水 ,
[0046] V 干空气 =V 空气导入量 / (1+H)
[0047] V 空含水 =V 空气导入量 ―V 干空气
[0048] Among them, V 空气导入量 This refers to the total amount of air entering the dry quenching furnace;
[0049] Based on the amount of coke processed by the dry quenching furnace, G 出焦 The volume V of residual volatile matter in the dry quenching furnace is calculated from the amount of residual volatile matter precipitated from the coke. 挥 ,
[0050] V 挥 =G 出焦 × Amount of residual volatile matter precipitated from coke
[0051] Through V 挥 The amount of oxygen in the residual volatiles converted into oxygen, V, was obtained.残氧 ,
[0052] V 残氧 =V 挥 ×0.5×10%
[0053] The amount of gas released, V 放 ,
[0054]
[0055] in, This represents the volume percentage of O2 in the circulating gas. This represents the volume percentage of H2O in the circulating gas.
[0056] The real-time calculation module calculates the actual heat exchange process as follows:
[0057] First, calculate the actual heat exchange Q of the dry quenching furnace. 吸 ,
[0058] Q 吸 =Q ica —Q oca
[0059] Among them, Q ica To obtain the sensible heat of the flue gas entering the furnace, Q oca The apparent heat of the flue gas after furnace discharge;
[0060] Q ica =V ica ×C ica ×t ica
[0061] Among them, V ica C is the air intake volume of the dry quenching furnace. ica The average isobaric specific heat of the inlet air of the dry quenching furnace, t ica It is the temperature of the air intake for the dry quenching furnace;
[0062] Q oca =V oca ×C oca ×t oca
[0063] Among them, V oca C is the air volume of the dry quenching furnace. oca The average isobaric specific heat of the exhaust air from the dry quenching furnace, t oca It is the temperature of the air outlet from the dry quenching furnace.
[0064] like Figure 1 As shown, the prediction module includes a feature transformation unit, a prediction model, an anomaly removal unit, a model validation unit, and a model screening and parameter tuning unit. The feature transformation unit performs feature transformation on the real-time data after removing outliers and sends the transformed data to the anomaly removal unit.
[0065] In this embodiment, the Box-Cox transform is used for feature transformation, and the transformation process is as follows: Figure 2 As shown, the Box-Cox transform is widely used to adjust datasets to a form closer to a normal distribution, thereby improving model performance and accuracy. The unique feature of the Box-Cox transform is its ability to adapt to different dataset distributions, transforming them into a more normal shape by adjusting the exponent of the data. The Box-Cox transform has significant advantages in processing sensor data, particularly in handling skewed data, improving data fitting performance, and reducing the impact of outliers. Furthermore, the Box-Cox transform can also explore and optimize the skewness and correlation of sensor data, providing a more reliable foundation for subsequent modeling and analysis.
[0066] The anomaly removal unit is used to remove outliers from the transformed data and then send the outlier-removed data to the model validation unit.
[0067] The outlier removal unit removes outliers from the transformed data by: using the model to predict the burn-off rate and boiler heat exchange under the current parameters; and using the residuals between the predicted burn-off rate and predicted boiler heat exchange and the actual burn-off rate and actual boiler heat exchange as the basis for removing outliers. The removal process is as follows: Figure 3 As shown.
[0068] The model validation unit contains a comparison model and multiple computational models. In this embodiment, the multiple computational models are designated as computational model 1, computational model 2, ..., computational model n, and the computational models are RandomForestRegressor, GBDT, lightGBM, xgboost, LinearRegression, and / or SVR. Each computational model is used to derive the calculated burn-off rate and calculated boiler heat exchange rate based on the data after removing outliers. The comparison model is used to compare the calculated burn-off rate and calculated boiler heat exchange rate with the actual burn-off rate and actual boiler heat exchange rate, and to determine the computational models with the lowest deviation rates as the base models, such as... Figure 4 As shown. The model selection and parameter tuning unit is used to allocate weights to each base model, forming a composite model, and then uses grid search to adjust the parameters of the composite model to obtain the prediction model. The prediction model is used to obtain the predicted burn-off rate and the predicted boiler heat exchange rate, and then sends them to the intelligent control module. In this embodiment, the five models with the lowest deviation rates are selected as base models, namely W1, W2, W3, W4, and W5, and the deviation rate comparison of each model is W. 1偏差 >W 2偏差 >W 3偏差 >W 4偏差 >W 5偏差 The weights of these five models are as follows:
[0069] W 5权重 >W 4权重 >W 3权重 >W 2权重 >W 1权重
[0070] Furthermore, W 5权重 +W 4权重 +W 3权重 +W 2权重 +W 1权重 =1.
[0071] The evaluation metrics for grid search parameter tuning are: MSE, R2, and MAE.
[0072] The intelligent control module includes a predictive comprehensive benefit unit and an adjustment unit. The predictive comprehensive benefit unit is used to derive the predicted comprehensive benefit based on the predicted burn-off rate and the predicted boiler heat exchange, and then sends the predicted comprehensive benefit to the adjustment unit.
[0073] The formula for predicting comprehensive returns is as follows:
[0074]
[0075] Where W is the comprehensive return, t 蒸汽 P is the theoretical steam quantity. 蒸汽 For the price of steam, t 焦炭处理量 For coke processing capacity, L 烧损 P represents the dry quenching coke loss rate. 焦炭 This refers to the price of coke. 蒸汽 Based on the theoretical heat transfer value Q of the dry quenching furnace 吸 The theoretical steam volume can be calculated directly. The specific process is based on existing technology and will not be elaborated here.
[0076] The adjustment unit is used to adjust the parameters of the dry quenching furnace and waste heat boiler in real time. These parameters include the opening degree of the air inlet valve, the circulating air volume, and the coke discharge rate. When the predicted comprehensive benefit approaches the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is correct, and the adjustment should continue in that direction. When the predicted comprehensive benefit deviates significantly from the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is incorrect, and the adjustment direction should be changed. Simultaneously, the adjustment unit accumulates experience in the adjustment direction, reducing errors through training for more precise control.
[0077] Assuming t is given 蒸汽 =102t / h (tons per hour), P 蒸汽 = 170 yuan, t 焦炭处理量 = 186 tons / hour, L 烧损 =2.5%~3%, P 焦炭 = 2440 yuan.
[0078] Under the current conditions, the predicted comprehensive benefit is 5703.08 yuan / h when the burn-off rate is 2.5%, and 3303.71 yuan / h when the burn-off rate is 3%. The theoretical maximum comprehensive benefit is 5703.08 yuan / h. If, at this point, the adjustment unit reduces the air inlet valve while keeping the circulating air volume and coke discharge volume unchanged, and the resulting predicted comprehensive benefit is closer to the theoretical maximum comprehensive benefit than the predicted benefit obtained from the parameters of the dry quenching furnace and waste heat boiler at the previous moment, then the adjustment direction is considered correct, and the adjustment continues until the predicted comprehensive benefit gets closer and closer to the theoretical maximum comprehensive benefit. Conversely, if the adjustment direction is incorrect, the parameters of the dry quenching furnace and waste heat boiler are adjusted in the opposite direction until the predicted comprehensive benefit gets closer and closer to the theoretical maximum comprehensive benefit. Simultaneously, the adjustment unit accumulates experience in the adjustment direction, reducing errors through training. That is, the next adjustment unit prioritizes adjusting in the direction of previously obtained correct experience, thereby reducing the number of adjustment errors and enabling more precise control.
[0079] This invention's dry quenching process optimization system, based on digital twins and artificial intelligence, leverages a large-scale digital twin-based data platform for data support and employs AI algorithms to predict dry quenching loss rate and boiler heat exchange. An intelligent control module performs a comprehensive benefit assessment based on the predicted dry quenching loss rate and boiler heat exchange, automatically adjusting the parameters of the dry quenching furnace and waste heat boiler according to the assessment. This integration of digital twins and artificial intelligence enables optimal boiler regulation without manual intervention. Considering production safety and the impact of shutdowns and maintenance, the system controls the air inlet valve, gradually regulating the circulating air volume and coke discharge rate. Automated control of these three parameters increases system usage frequency and further enhances production efficiency. Simultaneously, it incorporates factors such as coke and steam prices to provide the most economical loss rate control.
[0080] Furthermore, the predictive model of the dry quenching process optimization system based on digital twins and artificial intelligence is composed of multiple mathematical models, combining the advantages of multiple models to achieve complementarity among different models. The specific advantages are as follows:
[0081] 1. Diverse model selection: The ensemble model includes different types of base models with different structures, algorithms, or feature selection methods. This increases the diversity of the model ensemble and improves the overall robustness.
[0082] 2. Dynamic Weight Adjustment: Consider introducing a dynamic weight adjustment mechanism to adjust the weights of the model in the ensemble based on its performance on different data subsets. This approach allows the ensemble model to adapt more flexibly to different data distributions.
[0083] Model ensembles can improve the robustness of algorithms, making models more robust when facing complex and ever-changing real-world application scenarios.
Claims
1. A dry quenching process optimization system based on digital twins and artificial intelligence, characterized in that, It includes a data acquisition module, a real-time computing module, a prediction module, and an intelligent control module; The data acquisition module includes a data acquisition unit and a data preprocessing unit. The data acquisition unit is used to acquire real-time data from the dry quenching furnace and the waste heat boiler and send it to the data preprocessing unit. The data preprocessing unit is used to remove outliers from the real-time data and send it to the real-time calculation module and the prediction module. The real-time calculation module is used to calculate the actual burn-off rate and actual boiler heat exchange under the current parameters based on the real-time data after removing outliers, and send the actual burn-off rate and actual boiler heat exchange to the prediction module. The prediction module includes a feature transformation unit, a prediction model, an anomaly removal unit, a model verification unit, and a model selection and parameter tuning unit. The feature transformation unit transforms the real-time data after removing outliers and sends the transformed data to the anomaly removal unit. The anomaly removal unit removes outliers from the transformed data and sends the outlier-removed data to the model verification unit. The model verification unit includes a comparison model and multiple calculation models. Each calculation model calculates the burn-off rate and boiler heat exchange rate based on the outlier-removed data. The comparison model compares the calculated burn-off rate and calculated boiler heat exchange rate with the actual burn-off rate and actual boiler heat exchange rate, identifying the calculation models with the lowest deviation rates as base models. The model selection and parameter tuning unit allocates weights to each base model to form a composite model and adjusts the parameters of the composite model to obtain the prediction model. The prediction model obtains the predicted burn-off rate and predicted boiler heat exchange rate under the current parameters and sends them to the intelligent control module. The intelligent control module includes a predictive comprehensive benefit unit and an adjustment unit. The predictive comprehensive benefit unit is used to derive the predicted comprehensive benefit based on the predicted burn-out rate and predicted boiler heat exchange, and then sends the predicted comprehensive benefit to the adjustment unit. The adjustment unit is used to adjust the parameters of the dry quenching furnace and the waste heat boiler in real time. When the predicted comprehensive benefit approaches the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is correct, and the adjustment continues in that direction. When the predicted comprehensive benefit deviates from the theoretical maximum comprehensive benefit, it indicates that the adjustment direction is incorrect, and the adjustment direction is changed. At the same time, the adjustment unit accumulates experience in the adjustment direction and reduces errors through training to achieve more precise control.
2. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The data preprocessing unit removes outliers by using the minimum, 25%, 50%, 75%, maximum, and mean values of the data distribution as statistical indicators. The distribution of real-time data is evaluated from three aspects: the standard deviation of the data, the difference from the mean, and the normal distribution pattern. Extreme outliers are removed using the 5 sigma method.
3. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The feature transformation unit uses Box-Cox transformation to transform the features of the real-time data after outlier removal.
4. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The process of removing outliers from the transformed data by the anomaly removal unit is as follows: the predicted burn-off rate and predicted boiler heat exchange under the current parameters are obtained by using the model prediction, and the residual between the predicted burn-off rate and predicted boiler heat exchange and the actual burn-off rate and actual boiler heat exchange is used as the basis to remove outliers.
5. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The model screening and parameter tuning unit uses grid search for parameter tuning.
6. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 5, characterized in that, The evaluation metrics for the grid search parameter tuning are: MSE, R2, and MAE.
7. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 5, characterized in that, The principle for allocating weights to the model screening and parameter tuning unit is: the base model with the smaller the deviation, the greater the weight it occupies, and the weights of all base models are 1.
8. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The process by which the real-time calculation module calculates the actual burn-off rate is as follows: Based on the principle of carbon conservation in the dry quenching system, the theoretical coke burn-off T is obtained by measuring the CO and CO2 content in the gas emitted after the dry quenching furnace blower. C , Among them, V 放 V represents the amount of gas released after the blower. 放 Measured directly by sensors or calculated by formulas; This represents the volume percentage of CO in the circulating gas. This represents the volume percentage of CO2 in the circulating gas. Based on the theoretical coke burn-off T C The theoretical dry quenching coke loss rate was obtained. Among them, A d It is coke ash.
9. The dry quenching process optimization system based on digital twins and artificial intelligence as described in claim 1, characterized in that, The formula for predicting comprehensive return by the predicting comprehensive return unit is: Where W is the comprehensive return, t 蒸汽 P is the theoretical steam quantity. 蒸汽 For the price of steam, t 焦炭处理量 For coke processing capacity, L 烧损 P represents the dry quenching coke loss rate. 焦炭 This refers to the price of coke.