A coal ash characteristic analysis-based mixed coal blending optimization method and system

Abstract

This application belongs to the field of coal-fired power generation technology and discloses a method and system for optimizing coal blending based on coal ash characteristic analysis. The method includes: acquiring coal ash composition data of various coal samples to be blended; calculating the ash fusion point characteristic temperature and coking risk index of each coal sample based on the coal ash composition data using a pre-trained coking tendency prediction model; determining the target constraint condition for the blending ratio according to the target boiler coking risk threshold; using minimizing fuel cost or maximizing combustion efficiency as the objective function, and combining the ash fusion point characteristic temperature, coking risk index, and target constraint condition, solving the blending ratio that meets the coking risk control requirements through an optimization algorithm and outputting the result to guide actual operation; and monitoring boiler operating parameters in real time, updating the blending ratio when the coking tendency increases. This application can improve the rationality of coal blending, effectively reduce boiler coking risk, ensure stable combustion, and balance economy and safety.

Description

Technical Field

[0001] This application relates to the field of coal-fired power generation technology, specifically to a method and system for optimizing coal blending based on coal ash characteristic analysis. Background Technology

[0002] Currently, in coal-fired power generation and other coal-fired applications, blending coal is often used to optimize combustion and control fuel costs. However, existing methods for optimizing blending coal often focus solely on reducing fuel costs or improving combustion efficiency, failing to adequately analyze the ash characteristics of the coal sample. This makes it difficult to accurately grasp the coking characteristics during coal combustion. Furthermore, when determining the blending ratio, it is impossible to set appropriate constraints based on the boiler's coking control requirements. This can easily lead to coking during the combustion of the blended coal in the boiler, affecting the stable operation of the boiler. It also makes it difficult to achieve a balance between effective control of coking risks, reasonable management of fuel costs, and optimization of combustion efficiency, thus failing to meet the comprehensive needs of actual coal-fired applications. Summary of the Invention

[0003] To solve, or at least partially solve, the above-mentioned technical problems, this application provides a method and system for optimizing coal blending based on coal ash characteristic analysis.

[0004] In a first aspect, this application provides a method for optimizing coal blending based on coal ash characteristic analysis, comprising the following steps: S1. Obtain the coal ash composition data of the various coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. S2. Based on the coal ash composition data, calculate the characteristic temperature of ash melting point and coking risk index of each coal sample when it is burned alone using a pre-trained coking tendency prediction model. S3. Determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. S4. Taking the minimization of fuel cost or the maximization of combustion efficiency as the objective function, and combining the ash melting point characteristic temperature, the coking risk index and the objective constraints, the coal blending ratio that meets the coking risk control requirements is solved by an optimization algorithm. S5. Output the blending ratio of coal that meets the coking risk control requirements to the coal control system to guide the actual blending operation. S6. Monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, return to S1 to S5 and update the blending ratio of coal.

[0005] Optionally, the process of constructing the coking tendency prediction model includes: Collect historical coal ash composition data and corresponding historical ash fusion point experimental data from historical coal samples; Characteristic parameters are extracted from historical coal ash composition data of historical coal samples. These characteristic parameters include acid-base ratio, silica-alumina ratio, iron-calcium ratio, and total amount of alkaline oxides. A basic prediction model is trained using a random forest model, with the aforementioned feature parameters as input and the historical ash melting point feature temperature corresponding to the historical ash melting point experimental data as output. Based on the historical ash melting point characteristic temperature output by the basic prediction model, the corresponding coking risk index is calculated and integrated to form the coking tendency prediction model.

[0006] Optionally, before training the coking tendency prediction model, the method further includes: Call the thermodynamic simulation tool, input the historical coal ash composition data of historical coal samples and the corresponding historical combustion atmosphere conditions, and generate the historical liquid phase generation temperature for each historical coal sample. The historical liquid phase formation temperature corresponding to the historical coal sample is correlated and merged with historical coal ash composition data and historical ash melting point experimental data to obtain an enhanced training dataset. The enhanced training dataset is used to train the coking tendency prediction model.

[0007] Optionally, the coking tendency prediction model adopts a hybrid architecture, the specific structure and operation process of which include: The first layer is the random forest model, which takes the feature parameters or the optimal feature subset as input and outputs the preliminary ash melting point feature temperature and coking risk index. The liquid phase formation temperature of the corresponding coal sample is obtained by calling thermodynamic simulation tools, and the temperature difference between the preliminary ash melting point characteristic temperature and the liquid phase formation temperature is calculated. The second layer is a linear regression correction model, which takes the temperature difference as input and outputs the corresponding correction amount. The initial ash melting point characteristic temperature and coking risk index are combined with the correction amount to obtain the final ash melting point characteristic temperature and coking risk index.

[0008] Optionally, after extracting the feature parameters, the method further includes: The importance of each of the aforementioned feature parameters to the prediction results of ash melting point characteristic temperature and coking risk index is evaluated based on the random forest model. Remove feature parameters whose importance is lower than a preset importance standard; A recursive feature elimination method is used to iteratively remove the feature parameters that contribute the least to the model prediction until the optimal feature subset is obtained. The optimal feature subset is then used as the input to the coking tendency prediction model.

[0009] Optionally, after calculating the ash melting point characteristic temperature and coking risk index of each coal sample during individual combustion using a pre-trained coking tendency prediction model, the method further includes: By calling a thermodynamic simulation tool and inputting the coal ash composition data and the combustion atmosphere conditions of the boiler, the mineral phase composition and liquid phase formation temperature of the ash residue at different temperatures can be determined. Based on the mineral phase composition and the liquid phase formation temperature, the ash melting point characteristic temperature and coking risk index output by the coking tendency prediction model are corrected to obtain the corrected ash melting point characteristic temperature and coking risk index.

[0010] Optionally, the optimization algorithm is a multi-objective genetic algorithm, where the objective function simultaneously includes minimizing fuel cost and maximizing combustion efficiency, wherein: The fuel cost is calculated by summing the products of the unit price of each type of coal and the corresponding blending ratio; The combustion efficiency is calculated based on the volatile matter, calorific value, and corresponding blending ratio of each type of coal.

[0011] Optionally, the process of solving the blending ratio of mixed coal using the multi-objective genetic algorithm further includes: The multi-objective genetic algorithm outputs a Pareto optimal solution set after solving the problem. The Pareto optimal solution set includes multiple sets of coal blending ratios that satisfy the objective constraints, as well as the fuel cost and combustion efficiency corresponding to each set of coal blending ratios. The TOPSIS method or a preset decision preference is used to comprehensively evaluate the blending ratios of each group of coal in the Pareto optimal solution set; Based on the evaluation results, a compromise solution is selected, and the coal blending ratio corresponding to the compromise solution is determined as the final coal blending ratio.

[0012] Optionally, after obtaining the blending ratio of the mixed coal, the process also includes: Determine whether the coal ash after mixing, corresponding to the blending ratio, meets the target constraint conditions. If the target constraints are not met, an additive addition suggestion is generated. The additive is used to help increase the flow temperature of the mixed coal ash and reduce the risk of coking. The additive addition recommendations, along with the coal blending ratio, are output to the coal combustion control system to guide the actual blending and additive addition operations.

[0013] Secondly, this application also provides a coal blending optimization system based on coal ash characteristic analysis, comprising: The acquisition module is used to acquire the coal ash composition data of multiple coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. The calculation module is used to calculate the characteristic temperature of ash melting point and coking risk index of each coal sample when it is burned alone, based on the coal ash composition data and a pre-trained coking tendency prediction model. The determination module is used to determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. The solution module is used to determine the blending ratio of coal that meets the coking risk control requirements by using an optimization algorithm, with the objective function of minimizing fuel cost or maximizing combustion efficiency, combined with the ash melting point characteristic temperature, the coking risk index and the objective constraints. The output module is used to output the blending ratio of coal that meets the requirements for coking risk control to the coal control system to guide the actual blending operation. The update module is used to monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, the acquisition module is re-executed to the output module to update the blending ratio of coal.

[0014] The method provided in this application has the following beneficial effects: This method acquires specific ash composition data from multiple coal samples to be blended, providing a foundation for subsequently determining the coking-related characteristics of coal combustion. Based on the aforementioned ash composition data, a pre-trained coking tendency prediction model is used to calculate the ash melting point characteristic temperature and coking risk index of each coal sample during individual combustion. This effectively grasps the coking-related characteristics of a single coal sample during combustion, providing a targeted reference for coal blending and making coal selection before blending more targeted.

[0015] The target constraints for the blending ratio of coal are determined based on the coking risk threshold of the target boiler. These constraints are tailored to the actual operational needs of the target boiler, rather than using a uniform constraint standard. This ensures that the subsequently calculated blending ratio is more suitable for the target boiler and meets the coking control requirements during actual boiler operation. Using the minimization of fuel cost or the maximization of combustion efficiency as the objective function, and combining the ash melting point characteristic temperature, coking risk index, and target constraints, an optimization algorithm is used to solve the blending ratio. This allows the obtained blending ratio to satisfy both coking risk control requirements and fuel cost optimization or combustion efficiency improvement, solving the problem of balancing coking control, economy, and combustion effect in traditional coal blending.

[0016] Real-time monitoring of boiler combustion parameters allows for timely updates to the blending ratio by reverting to previous steps when an increased tendency to coking is detected. This enables prompt responses to changes in coking risk during boiler operation, preventing the accumulation of coking problems from impacting boiler operation, ensuring stable boiler operation under mixed coal combustion conditions, and reducing the risk of equipment failure or decreased operating efficiency due to coking. Attached Figure Description

[0017] Figure 1A schematic diagram of a coal blending optimization method based on coal ash characteristic analysis is provided for an embodiment of this application. Figure 2 This is a schematic diagram of the doping ratio-flow temperature relationship curve provided in an embodiment of this application. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, specific embodiments of this application will be described in further detail below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are merely for explaining this application and not for limiting it. It should also be noted that, for ease of description, only the parts relevant to this application are shown in the drawings, not all of them. Before discussing exemplary embodiments in more detail, it should be mentioned that some exemplary embodiments are described as processes or methods depicted as flowcharts. Although the flowcharts describe operations (or steps) as sequential processes, many of these operations can be performed in parallel, concurrently, or simultaneously. Furthermore, the order of the operations can be rearranged. The process can be terminated when its operation is completed, but may also have additional steps not included in the drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] See Figure 1 This application provides a method for optimizing coal blending based on coal ash characteristic analysis, including the following steps: S1. Obtain the coal ash composition data of the various coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. S2. Based on coal ash composition data, the characteristic temperature of ash melting point and coking risk index of each coal sample during individual combustion are calculated using a pre-trained coking tendency prediction model. S3. Determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. S4. Taking the minimization of fuel cost or the maximization of combustion efficiency as the objective function, and combining the ash melting point characteristic temperature, coking risk index and objective constraints, the coal blending ratio that meets the coking risk control requirements is solved by optimization algorithm. S5. Output the blending ratio of coal that meets the coking risk control requirements to the coal control system to guide the actual blending operation. S6. Monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, return to S1 to S5 and update the blending ratio of coal.

[0021] Specifically, coal blending optimization must be based on coal ash characteristics. Silica, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide in coal ash are key components affecting its melting and coking properties. Therefore, optimization work must first obtain data on these coal ash components from the various coal samples to be blended. In actual production, single-type coal combustion is prone to coking problems, or results in high fuel costs and poor combustion efficiency. Coal blending can regulate coking characteristics while also considering economic efficiency or combustion performance. Therefore, the optimization method revolves around coal ash characteristic analysis, first clarifying the coking-related indicators when each coal sample is burned individually, then determining blending constraints based on boiler requirements, and finally solving for the appropriate blending ratio.

[0022] After obtaining coal ash composition data, the ash melting point characteristic temperature and coking risk index of each coal sample are calculated by using a pre-trained coking tendency prediction model. The ash melting point characteristic temperature reflects the ease of coal ash melting, while the coking risk index reflects the possibility of coking during coal sample combustion. Mastering these indicators can determine whether a single coal sample is suitable for boiler combustion and can also provide a reference benchmark for coal blending.

[0023] Different boilers have different structural designs and combustion conditions, and their tolerance to coking varies. Therefore, based on the coking risk threshold of the target boiler, the target constraint condition for the blending ratio of coal is determined. The target constraint condition needs to ensure that the coking risk after the blended coal combustion does not exceed the boiler's tolerance range, so as to avoid coking affecting the normal operation of the boiler.

[0024] When determining the blending ratio of mixed coal, the objective function is to minimize fuel cost or maximize combustion efficiency. This is combined with previously obtained ash fusion point characteristic temperature, coking risk index, and defined target constraints. An optimization algorithm is used to select a suitable blending ratio. This approach ensures that the combustion of mixed coal does not generate coking risks exceeding control requirements, while also meeting the producer's needs for fuel cost or combustion efficiency, balancing safety and economy. Specifically, when determining the blending ratio, the target constraints are the core premise. The core indicator of the ash fusion point characteristic temperature is the flow temperature. A higher flow temperature reflects the flow properties of the molten coal ash; a higher flow temperature reduces the likelihood of coking. The target constraints clearly state that the flow temperature of the mixed coal ash must meet the corresponding requirements, while the coking risk index must be lower than the preset risk threshold set by the boiler. When combining the samples, the characteristic temperature of ash melting point and the coking risk index of each coal sample are first converted into corresponding indicators of mixed coal ash according to the blending ratio. Then, the compliance of the two indicators is taken as a hard premise and incorporated into the solution logic of the optimization algorithm. During the iteration of the optimization algorithm, the mixed coal ash indicators corresponding to each blending ratio are checked simultaneously to see if they meet the constraints. If they do not meet the constraints, they are directly eliminated. Finally, the blending ratio of mixed coal that meets the constraints of ash melting point and coking risk and is in line with the goal of minimizing fuel cost or maximizing combustion efficiency is selected.

[0025] After obtaining the blending ratio of the coal that meets the requirements for coking risk control, the ratio is output to the coal combustion control system, which can then use this ratio to blend various coal samples.

[0026] During the actual combustion process of the boiler, the coal quality may fluctuate, and the combustion conditions may also change due to load adjustments. These factors may cause changes in the tendency to coke. Therefore, it is necessary to monitor the operating parameters of the boiler combustion process in real time and grasp the combustion status. Once an increase in the tendency to coke is detected, it means that the current blending ratio can no longer effectively control the risk of coking. At this time, the process of acquiring coal ash composition data, calculating coking-related indicators, confirming constraints, and solving the blending ratio is restarted to update the blending ratio of mixed coal and continuously adapt to changes in combustion.

[0027] This optimization method can control the coking risk of mixed coal combustion, effectively avoid serious coking in the boiler, ensure stable and continuous boiler operation, and at the same time, by setting the objective function, it can take into account fuel cost control or combustion efficiency improvement, improve the economics of coal-fired boiler operation, and the design of real-time updated blending ratio can cope with various changes in actual combustion, maintain the continuity of optimization effect, reduce various operational problems caused by coking, improve equipment operation reliability, avoid shutdown and maintenance due to coking, meet the various needs of actual coal production, and adapt to the boiler combustion needs under different operating conditions.

[0028] Figure 2This is a schematic diagram of the doping ratio-flow temperature relationship curve provided in an embodiment of this application. In practical applications, the correlation between the blending ratio and the characteristic temperature of the ash melting point of the blended coal can be presented through this curve. The curve uses the blending ratio (unit: %) of the coal sample to be blended in the blended coal as the abscissa and the flow temperature (unit: ℃) of the characteristic temperature of the ash melting point of the blended coal as the ordinate, showing the variation law of the characteristic temperature of the ash melting point of the blended coal under different blending ratios. It can intuitively see the trend of the influence of the blending ratio adjustment on the characteristic temperature of the ash melting point of the blended coal, and the flow temperature shows regular fluctuations corresponding to the ratio change. Such curve changes also confirm the scientific nature of this blending optimization method based on coal ash characteristic analysis. The correspondence between the ratio and the flow temperature shown by the curve can intuitively reflect the regulatory effect of the blending ratio adjustment on the coking risk, further illustrating that the blending ratio determined by this method can accurately control the characteristic temperature of the ash melting point of the blended coal within a safe range. Meanwhile, the curve can also help verify the rationality of the objective function setting. Whether focusing on controlling fuel costs or improving combustion efficiency, the selected blending ratio can correspond to the range in the curve that meets the ash melting point characteristic temperature requirements. This allows the blended coal combustion to meet both the coking risk control requirements and the economic or combustion performance requirements of actual production, indicating that the optimization method can provide a reliable guarantee for the stable operation of coal-fired boilers.

[0029] In some implementations, the process of constructing a coking tendency prediction model includes: Collect historical coal ash composition data and corresponding historical ash fusion point experimental data from historical coal samples; Characteristic parameters were extracted from historical coal ash composition data of historical coal samples. These parameters included acid-base ratio, silica-alumina ratio, iron-calcium ratio, and total amount of alkaline oxides. A random forest model was used, with feature parameters as input and historical ash melting point feature temperatures corresponding to historical ash melting point experimental data as output, to train a basic prediction model. Based on the historical ash melting point characteristic temperature output by the basic prediction model, the corresponding coking risk index is calculated and integrated to form a coking tendency prediction model.

[0030] Specifically, historical coal samples of different coal types and under different combustion conditions are collected first. For each historical coal sample, historical coal ash composition data are obtained. The detection range needs to cover key components such as silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. Historical ash melting point experimental data of the corresponding historical coal sample are also obtained.

[0031] After data acquisition, the next step is feature parameter extraction. Feature parameters reflecting coal ash melting and coking characteristics are screened and calculated from historical coal ash composition data. These parameters include acid-base ratio, silica-alumina ratio, iron-calcium ratio, and total amount of alkaline oxides. The acid-base ratio is calculated by statistically analyzing the total amount of acidic oxides (silicon dioxide, aluminum oxide, etc.) and alkaline oxides (calcium oxide, magnesium oxide, sodium oxide, potassium oxide, etc.) in historical coal ash composition, and then calculating the ratio between the two. The silica-alumina ratio is the ratio of silica content to aluminum oxide content in historical coal ash composition. The iron-calcium ratio is the ratio of ferric oxide content to calcium oxide content. The total amount of alkaline oxides is the sum of the contents of calcium oxide, magnesium oxide, sodium oxide, potassium oxide, etc. Through these calculations, the original coal ash composition data is transformed into more targeted feature parameters, providing effective input for model training.

[0032] Subsequently, a basic prediction model was trained, using a random forest model as the basic model architecture. The extracted feature parameters were used as input data, and the historical ash fusion point characteristic temperatures corresponding to historical ash fusion point experimental data were used as output data. The dataset was appropriately divided: one part was used as the training set for model parameter learning, and the other part as the validation set for model performance verification. During training, the basic learning parameters of the basic prediction model were gradually adjusted, allowing the model to learn the intrinsic correlation between the feature parameters and the ash fusion point characteristic temperatures, enabling it to accurately output the corresponding ash fusion point characteristic temperatures based on the input feature parameters. After multiple iterations of training, the basic prediction model's output on the validation set stabilized and the deviation met expectations, thus completing the training of the basic prediction model.

[0033] After the basic prediction model is trained, a coking risk index is calculated based on its output historical ash fusion point characteristic temperature. Combining the correlation between ash fusion point characteristic temperature and coking probability, a corresponding calculation rule is set: the lower the ash fusion point characteristic temperature, the higher the coking risk index. This rule transforms the historical ash fusion point characteristic temperature output by the basic prediction model into a coking risk index that intuitively reflects the likelihood of coking. Finally, the calculation logic of the basic prediction model and the coking risk index is integrated to form a complete coking tendency prediction model. This model, upon receiving input data, can output both the ash fusion point characteristic temperature and the corresponding coking risk index, providing data support for subsequent optimization of coal blending ratios.

[0034] Through this construction process, the coking tendency prediction model can accurately capture the correlation between coal ash composition and coking characteristics, providing accurate judgment basis for coking-related indicators for coal blending optimization, and ensuring the scientificity and effectiveness of subsequent blending schemes.

[0035] In some implementations, prior to training the coking tendency prediction model, the following steps are also included: Call the thermodynamic simulation tool, input the historical coal ash composition data of historical coal samples and the corresponding historical combustion atmosphere conditions, and generate the historical liquid phase generation temperature for each historical coal sample. The historical liquid phase formation temperature corresponding to the historical coal sample is correlated and merged with historical coal ash composition data and historical ash melting point experimental data to obtain an enhanced training dataset. The enhanced training dataset was used to train the coking tendency prediction model.

[0036] Data augmentation can be performed before training a coking tendency prediction model. This approach can compensate for insufficient historical data or limited scenario coverage, making the training dataset more representative and thus improving the model's generalization ability and prediction accuracy. The core of data augmentation is to generate supplementary data using thermodynamic simulation tools and then merge it with existing historical data to form an augmented dataset.

[0037] First, the simulation data generation operation is performed. A suitable thermodynamic simulation tool for coal ash characteristic analysis is selected, capable of simulating the melting characteristics of coal ash with different compositions under specific combustion atmospheres. During the operation, the historical coal ash composition data of collected historical coal samples is completely input into the thermodynamic simulation tool. The input must include all key component data such as silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. Simultaneously, based on the actual combustion conditions of the historical coal samples, the corresponding historical combustion atmosphere conditions are determined, including oxygen and carbon dioxide concentrations during combustion, and this is also input into the thermodynamic simulation tool. The thermodynamic simulation tool simulates the temperature change history of the historical coal samples during actual combustion, calculating and generating the historical liquid phase formation temperature for each historical coal sample under the corresponding combustion atmosphere. The historical liquid phase formation temperature reflects the critical state at which coal ash begins to form a liquid phase at different temperatures, supplementing data related to coking characteristics.

[0038] After the simulation data is generated, data association is performed. Using the unique identifier of each historical coal sample as the association basis, the historical liquid phase formation temperature corresponding to each historical coal sample is matched with the existing historical coal ash composition data and historical ash fusion point experimental data for that coal sample, ensuring a one-to-one correspondence between various data types for the same coal sample. After matching, the three types of data are merged and integrated to form an enhanced training dataset with a complete structure and comprehensive information. Compared to the original historical dataset, this dataset adds the dimension of liquid phase formation temperature, which can more comprehensively reflect the correlation between coal ash composition and melting and coking characteristics.

[0039] Finally, the enhanced training dataset was used to train the coking tendency prediction model. During training, the same operational logic as the original training process was followed. First, the enhanced training dataset was preprocessed to unify the data format and remove outliers. Then, the preprocessed enhanced training dataset was divided into a training set and a validation set. The training set was used for model parameter learning, and the validation set was used to monitor performance changes during model training. Subsequently, the feature parameters were used as input, and the historical ash melting point feature temperatures corresponding to the historical ash melting point experimental data were used as output. These were then substituted into the random forest model for training. By iteratively adjusting the model parameters, the model learned various correlation patterns in the enhanced training dataset until the model's prediction results on the validation set were stable and met expectations.

[0040] Such data augmentation operations can enrich the information dimensions and coverage of training data, allowing the model to more fully learn the correlation between coking characteristics under different coal ash compositions and combustion conditions. This effectively improves the stability and accuracy of model predictions, providing a more reliable basis for judging coking-related indicators for subsequent optimization of coal blending ratios.

[0041] In some implementations, the coking tendency prediction model employs a hybrid architecture, the specific structure and operation of which include: The first layer is a random forest model, which takes feature parameters or the optimal feature subset as input and outputs preliminary ash melting point feature temperature and coking risk index. The liquid phase formation temperature of the corresponding coal sample is obtained by calling thermodynamic simulation tools, and the temperature difference between the preliminary ash melting point characteristic temperature and the liquid phase formation temperature is calculated. The second layer is a linear regression correction model, which takes the temperature difference as input and outputs the corresponding correction amount. The initial ash melting point characteristic temperature and coking risk index are combined with the correction amount to obtain the final ash melting point characteristic temperature and coking risk index.

[0042] Specifically, the coking tendency prediction model adopts a hybrid architecture, using two models working together to achieve accurate prediction results. Compared to a single model, it can better capture the correlation between coal ash composition and coking characteristics, making the output ash melting point characteristic temperature and coking risk index more consistent with actual combustion scenarios. The specific structure of this hybrid architecture revolves around the two-layer model, and its operation proceeds step by step according to a fixed process.

[0043] First, the first layer of the random forest model is run. The characteristic parameters corresponding to the coal sample are organized, including acid-base ratio, silicon-aluminum ratio, iron-calcium ratio, and total amount of alkaline oxides. These characteristic parameters are input into the random forest model. Based on the parameter correlation rules formed by pre-training, the random forest model processes the input characteristic parameters. By analyzing the intrinsic relationship between various characteristic parameters and ash fusion point characteristic temperature and coking risk index, the model outputs the preliminary ash fusion point characteristic temperature and coking risk index. This step does not require the introduction of other auxiliary data and only relies on the model's own training results to complete the preliminary prediction.

[0044] After completing the initial prediction, the liquid phase formation temperature acquisition and difference calculation are performed. First, a thermodynamic simulation tool suitable for coal ash characteristic analysis is selected, and the coal ash composition data of the corresponding coal sample is input into the simulation tool. The coal ash composition includes silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. At the same time, the atmospheric conditions during the actual combustion of the coal sample are simulated. The simulation tool calculates the melting state of the coal ash during temperature changes to generate the liquid phase formation temperature of the corresponding coal sample. After generating the liquid phase formation temperature, the temperature difference between the preliminary ash melting point characteristic temperature output from the first layer and the liquid phase formation temperature is calculated based on this preliminary ash melting point characteristic temperature.

[0045] Next, the second-level linear regression correction model is run. The calculated temperature difference is used as input data and substituted into the linear regression correction model. This linear regression correction model has been pre-trained and has mastered the correlation between the temperature difference and the prediction deviation. After the difference data is input, the linear regression correction model processes it through predetermined operation logic and outputs the corresponding correction amount. The correction amount will change accordingly based on the magnitude and sign of the temperature difference, matching the degree of deviation of the preliminary prediction result, so that the correction direction and magnitude are in line with the actual needs, avoiding over-correction or under-correction.

[0046] Finally, the preliminary results and correction values ​​are fused. For the preliminary ash melting point characteristic temperature output from the first layer, it is integrated with the correction value output from the linear regression correction model. The preliminary temperature value is adjusted according to the corresponding calculation rules to obtain the corrected ash melting point characteristic temperature. At the same time, for the preliminary coking risk index, it is adjusted in conjunction with the corrected ash melting point characteristic temperature to keep the coking risk index and the corrected temperature in match, thus obtaining the accurate final ash melting point characteristic temperature and coking risk index.

[0047] This hybrid architecture and operating process can effectively compensate for the bias of single model predictions, making the output coking-related indicators more consistent with the actual combustion of coal samples. This provides reliable data support for subsequent judgment of coal sample coking characteristics and optimization of blending ratios, avoiding unreasonable blending schemes due to prediction biases.

[0048] Alternatively, a third lightweight gradient boosting tree can be added to the existing first-layer random forest model and second-layer linear regression correction model. In operation, the first-layer random forest model outputs the preliminary ash melting point characteristic temperature and coking risk index, then calculates the temperature difference between these and the liquid phase formation temperature generated by thermodynamic simulation. This temperature difference, along with the preliminary prediction results from the first layer, is then input into the third-layer lightweight gradient boosting tree. This third layer, by pre-learning the complex nonlinear laws governing prediction deviations under different coal quality characteristics, performs feature fusion and preliminary deviation calibration on the input data, outputting the calibrated temperature difference. This calibrated temperature difference is then input into the second-layer linear regression correction model to calculate the precise correction amount. Finally, it is fused with the preliminary prediction results to obtain the final ash melting point characteristic temperature and coking risk index. This approach solves the problem that the linear correction in the original two-layer architecture is difficult to adapt to the complex nonlinear relationship between coal ash melting characteristics and coking risk, making the correction logic more consistent with the actual coal combustion laws and further improving prediction accuracy.

[0049] In addition, before calculating the corresponding indicators of coal ash composition for each coal sample (such as ash melting point characteristic temperature, coking risk index, etc.), it is also possible to first test the particle size distribution of the pulverized coal sample, classify the grade according to the particle size range of the actual blending, and different particle size ranges correspond to different correction coefficients. The finer the particle size, the more complete the combustion of the coal sample and the more complete the release of coal ash components, and the correction coefficient approaches 1. Coarser particle size coal samples are prone to ash residue, and the correction coefficient is appropriately increased to make up for the deviation.

[0050] The measured values ​​of coal ash composition for a single coal sample are adjusted based on a correction factor. Then, the ash fusion point characteristic temperature and coking risk index are calculated based on this adjustment. Subsequent weighted calculations of the mixed coal ash indicators (such as ash fusion point characteristic temperature and coking risk index) all use the corrected single coal data. This method avoids compositional distortion caused by particle size differences, allowing subsequent blending ratio optimization to better reflect the actual coal ash characteristics during combustion, further ensuring the accuracy of coking risk control. It is suitable for practical application scenarios involving fluctuations in the particle size of coal samples from power plants.

[0051] In some implementations, after extracting the feature parameters, the method further includes: The importance of each characteristic parameter to the prediction results of ash melting point characteristic temperature and coking risk index is evaluated based on the random forest model. Remove feature parameters whose importance is lower than the preset importance standard; A recursive feature elimination method is adopted to iteratively eliminate the feature parameters that contribute the least to the model prediction until the optimal feature subset is obtained. The optimal feature subset is then used as the input to the coking tendency prediction model.

[0052] After extracting feature parameters from historical coal ash composition data, feature filtering is performed. This step can remove redundant or weakly correlated parameters, reduce the interference of irrelevant information on model prediction, and make the features input to the model more targeted, thereby improving the training efficiency and prediction accuracy of the coking tendency prediction model.

[0053] First, an importance assessment of the feature parameters is performed. All extracted feature parameters (including acid-base ratio, silicon-aluminum ratio, iron-calcium ratio, and total amount of alkaline oxides) are input into the random forest model along with the corresponding historical ash fusion point experimental data. During the operation of the random forest model, based on the contribution of each feature parameter to the prediction results of the ash fusion point characteristic temperature and coking risk index, an importance score is output for each feature parameter. A higher importance score indicates a stronger correlation between the feature parameter and the prediction results, and a more significant improvement in prediction accuracy; conversely, a lower importance score indicates higher redundancy of the feature parameter and a smaller impact on the prediction results.

[0054] After completing the importance assessment, low-importance feature parameters are removed. First, based on the actual prediction needs and model training objectives, a reasonable importance standard is preset. This standard can be determined by combining historical data validation results to effectively distinguish between effective and redundant features. Then, the importance score of each feature parameter is compared one by one with the preset standard. If a feature parameter's score is lower than the preset standard, it indicates that its positive effect on the prediction result is limited, or it may even introduce interference. In this case, the feature parameter is removed from the feature set, retaining only those feature parameters whose scores meet the preset standard.

[0055] Next, recursive feature elimination is performed to further optimize the feature set. First, the feature parameters after initial elimination are input into the random forest model to complete one round of model training and evaluate the current model's predictive performance. Then, the parameters with the least contribution to prediction are selected from the remaining feature parameters and removed. The model is then retrained using the updated feature set, and its performance is evaluated. This iterative "training-evaluation-elimination" process is repeated, removing the least contributing parameter from the current feature set in each iteration, until the model's predictive performance no longer improves due to feature parameter removal, or even shows a downward trend, at which point the iteration stops.

[0056] After the iteration stops, the set of feature parameters remaining at this point is determined as the optimal feature subset. When training the clustering tendency prediction model subsequently, all the originally extracted feature parameters are no longer used; instead, this optimal feature subset is used as the model's input data. This feature selection process significantly reduces the data dimensionality input to the random forest model, decreases the computational load during model training, improves training efficiency, and avoids prediction interference caused by redundant features.

[0057] In some implementations, after calculating the ash melting point characteristic temperature and coking risk index of each coal sample during individual combustion using a pre-trained coking tendency prediction model, the method further includes: By using thermodynamic simulation tools and inputting coal ash composition data and boiler combustion atmosphere conditions, the mineral phase composition and liquid phase formation temperature of ash slag at different temperatures can be determined. Based on the mineral phase composition and liquid phase formation temperature, the ash melting point characteristic temperature and coking risk index output by the coking tendency prediction model are corrected to obtain the corrected ash melting point characteristic temperature and coking risk index.

[0058] After calculating the ash melting point characteristic temperature and coking risk index of each coal sample during individual combustion using a pre-trained coking tendency prediction model, auxiliary analysis and result correction can be performed using thermodynamic simulation tools. This step can improve the problem of insufficient consideration of the microscopic mechanism of coal ash melting in the model prediction process, and make the final ash melting point characteristic temperature and coking risk index more consistent with the coal ash characteristics under actual combustion scenarios.

[0059] First, a thermodynamic simulation analysis is performed. During the operation, the coal ash composition data of the coal sample to be analyzed is input into the simulation tool. This input includes components such as silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. Then, combined with the actual combustion conditions of the target boiler, the corresponding combustion atmosphere conditions are determined, including parameters such as oxygen concentration, carbon dioxide concentration, and water vapor content during combustion. These parameters are simultaneously input into the thermodynamic simulation tool to simulate the impact of the real combustion environment on the coal ash melting characteristics. After data input, the thermodynamic simulation tool is started, and a temperature range matching the actual boiler combustion is set. Using the built-in thermodynamic calculation model of the simulation tool, the melting process of coal ash under different temperature gradients is simulated, gradually outputting the ash mineral phase composition corresponding to each temperature node. Simultaneously, the critical temperature at which the coal ash begins to form a liquid phase, i.e., the liquid phase formation temperature, is determined, forming a complete simulation analysis result.

[0060] Subsequently, the predicted results were revised. First, the simulated mineral phase composition and liquid phase formation temperature were analyzed. Regarding the mineral phase composition, the proportion of easily fusible minerals (such as feldspar and vitreous bodies) was determined. A higher proportion of easily fusible minerals indicates that the coal sample is more prone to melting and coking at the corresponding temperature, requiring a corresponding downward adjustment of the model's predicted ash fusion point temperature and an upward adjustment of the coking risk index. Conversely, a higher proportion of refractory minerals (such as quartz and corundum) requires a corresponding upward adjustment of the ash fusion point temperature and a downward adjustment of the coking risk index. Regarding the liquid phase formation temperature, it was compared with the model's predicted ash fusion point temperature. If the liquid phase formation temperature was lower than the predicted ash fusion point temperature, the model's prediction was conservative, requiring an appropriate downward adjustment of the ash fusion point temperature and an upward adjustment of the coking risk index. If the liquid phase formation temperature was higher than the predicted ash fusion point temperature, the model may have overestimated the coking risk, requiring an appropriate upward adjustment of the ash fusion point temperature and a downward adjustment of the coking risk index.

[0061] Based on the above analysis results, specific correction rules were formulated to clarify the correction ranges corresponding to the proportions of different mineral phases and the deviations in liquid phase formation temperatures. According to the correction rules, the ash melting point characteristic temperature and coking risk index output by the coking tendency prediction model were adjusted one by one. During the adjustment process, the correction range was kept consistent with the degree of deviation in the simulation analysis results to avoid over-correction or under-correction. After the correction was completed, the final ash melting point characteristic temperature and coking risk index were obtained. This result combines the advantages of model prediction and thermodynamic simulation, and can more accurately reflect the coking characteristics of coal samples under actual boiler combustion conditions.

[0062] In some implementations, the optimization algorithm is a multi-objective genetic algorithm, where the objective function simultaneously includes minimizing fuel cost and maximizing combustion efficiency, wherein: Fuel costs are calculated by summing the products of the unit price of each type of coal and its corresponding blending ratio; Combustion efficiency is calculated based on the volatile matter, calorific value, and corresponding blending ratio of each type of coal.

[0063] After determining the target constraints for the blending ratio of mixed coal, a multi-objective genetic algorithm is used to solve for the blending ratio. This can simultaneously take into account the needs of fuel cost control and combustion efficiency improvement, so that the obtained blending ratio not only meets the requirements for coking risk control, but also meets the economic and combustion performance requirements of the producer.

[0064] First, the objective function is constructed and the basic data is prepared. The objective function includes two dimensions: minimizing fuel cost and maximizing combustion efficiency. The calculation logic for each dimension needs to be clearly defined and the corresponding basic data obtained. For fuel cost calculation, the unit price of each coal sample to be blended is collected through market research or procurement records, with each unit price corresponding to a coal sample type. Then, the calculation rule for fuel cost is determined, namely, the total fuel cost of blended coal is equal to the sum of the products of the unit price of each coal sample and the blending ratio of that coal sample. This rule is used to link the unit price with the blending ratio, forming the expression for calculating fuel cost. For combustion efficiency calculation, the volatile matter and calorific value data of each coal sample to be blended can be obtained through laboratory testing. The volatile matter test needs to simulate the pyrolysis conditions under actual combustion environment, and the calorific value test is completed using heat measurement equipment so that the data can truly reflect the combustion performance of the coal sample. Then, combined with the principle of coal combustion, the calculation logic of combustion efficiency is determined—the combustion efficiency of blended coal is positively correlated with the volatile matter and calorific value of each coal sample. The comprehensive combustion performance parameters of blended coal are calculated by weighted summation, and then converted into quantifiable combustion efficiency indicators to form a calculation expression that maximizes combustion efficiency.

[0065] After constructing the objective function, an iterative solution using a multi-objective genetic algorithm is executed. The first step is to initialize the population. Based on the type of coal sample to be blended, the range of values ​​for the blending ratio of each coal sample is set (the sum of the blending ratios of all coal samples is 1, and the blending ratio of a single coal sample is not less than 0). A certain number of blending ratio combinations are generated randomly to form an initial population. Each individual in the population corresponds to a set of potential coal blending ratios.

[0066] The second step is to set the fitness function, which combines the two constructed objective functions with the objective constraints (ash melting point characteristic temperature corresponding to the coking risk threshold and coking risk index requirements). The core function of the fitness function is to determine whether the blending ratio corresponding to each individual in the population meets the coking risk control requirements, and at the same time quantifies the comprehensive performance of fuel cost and combustion efficiency under the blending ratio, generating the fitness value of each individual. The higher the fitness value, the closer the blending ratio is to the optimal target.

[0067] The third step executes the core iterations of the genetic algorithm, including selection, crossover, and mutation. The selection operation uses roulette wheel or tournament selection to select individuals with high fitness values ​​from the current population and retain them for the next generation, ensuring the inheritance of superior genes. The crossover operation randomly selects two individuals as parents and exchanges some blending ratio genes through single-point or multi-point crossover to generate new offspring individuals, enriching population diversity. The mutation operation randomly selects a specific coal sample blending ratio from some individuals in the population and makes minor adjustments within a set mutation range to avoid the population getting trapped in local optima. After each iteration, the fitness value of each individual in the new generation is recalculated, and it is determined whether the iteration termination condition is met (e.g., the number of iterations reaches a preset upper limit, or the fitness value of the best individual tends to stabilize).

[0068] After the iteration terminates, all individuals that meet the target constraints (i.e., the corresponding ash melting point characteristic temperature and coking risk index meet the coking risk threshold requirements) are selected from the final population. The blending ratios corresponding to these individuals constitute the preliminary set of blended coal blending ratios that meet the requirements. Through the solution operation of the above multi-objective genetic algorithm, the relationship between fuel cost and combustion efficiency can be fully balanced under the premise of controllable coking risk. This provides a rich and high-quality candidate blending ratio for the subsequent selection of compromise solutions, ensuring that the final output blending ratio has safety, economy, and combustion performance.

[0069] In some implementations, the process of solving the blending ratio of mixed coal using a multi-objective genetic algorithm also includes: After solving the problem using the multi-objective genetic algorithm, the Pareto optimal solution set is output. The Pareto optimal solution set includes multiple sets of coal blending ratios that satisfy the objective constraints, as well as the fuel cost and combustion efficiency corresponding to each set of coal blending ratios. The TOPSIS method or the preset decision preferences are used to comprehensively evaluate the blending ratios of each group of coal in the Pareto optimal solution set. Based on the evaluation results, a compromise solution is selected, and the coal blending ratio corresponding to the compromise solution is determined as the final coal blending ratio.

[0070] After the multi-objective genetic algorithm completes iterative solving and outputs the Pareto optimal solution set, a compromise solution can be determined through a specific evaluation and screening process. This allows the final solution that best meets actual production needs to be selected from multiple blending ratios that meet coking risk constraints. The core of this process is to quantify the comprehensive performance of each candidate blending ratio through scientific evaluation methods, so that the final result can balance the dual requirements of fuel cost and combustion efficiency.

[0071] First, the Pareto optimal solution set is organized. All candidate blending ratios output by the multi-objective genetic algorithm are extracted, clarifying the types of coal samples to be blended and their respective proportions for each ratio. The fuel cost calculation results and combustion efficiency evaluation results for each ratio are retrieved, and these three are organized into a structured dataset according to the correspondence between "blending ratio combination - fuel cost - combustion efficiency". During the organization process, the data correlation is checked one by one to confirm the matching relationship between each blending ratio and the corresponding cost and efficiency data, avoiding subsequent evaluation biases due to data mismatches. At the same time, candidate ratios with incomplete data or logical contradictions are eliminated.

[0072] A comprehensive evaluation is then conducted, which can be performed using the TOPSIS (Top-Approximation-Ideal-Solution Ranking) method or based on pre-defined decision preferences. If the TOPSIS method is chosen, the first step is to standardize the processed dataset, converting fuel cost (lower is better) and combustion efficiency (higher is better) into standardized values ​​with uniform dimensions, eliminating the difference in magnitude between different indicators. The second step is to determine the positive and negative ideal solutions. The positive ideal solution is the virtual optimal solution with the minimum standardized fuel cost and the maximum combustion efficiency, while the negative ideal solution is the virtual worst solution with the maximum standardized fuel cost and the minimum combustion efficiency. The third step is to calculate the Euclidean distance between each group of candidate blending ratios and the positive and negative ideal solutions. The closer the distance is to the positive ideal solution and the farther the distance is to the negative ideal solution, the better the overall performance of the solution. The fourth step is to calculate the closeness of each group of candidate ratios based on the distance calculation results. The closeness value ranges from 0 to 1, with a value closer to 1 indicating better overall performance.

[0073] If the evaluation is based on preset decision preferences, the core needs and preferences of the producers must first be clarified. For example, if the focus is on controlling fuel costs, then cost should be given a higher weight; if the focus is on improving combustion efficiency, then efficiency should be given a higher weight. The weighting settings need to be confirmed through consultation in conjunction with actual production needs. Then, the fuel costs and combustion efficiencies corresponding to each group of candidate blending ratios are weighted and summed according to the preset weights to calculate a comprehensive score. The higher the score, the more the scheme conforms to the preset preferences.

[0074] Finally, the compromise solution is selected and determined. If the TOPSIS method is used, all candidate blending ratios are sorted from highest to lowest similarity, and the ratio with the highest similarity is selected as the compromise solution. If multiple ratios have similar similarities and are all at a high level, further verification based on the actual production scenario is required, and the blending ratio suitable for the boiler's daily operating conditions should be prioritized. If a preset decision preference assessment is used, candidate ratios are sorted from highest to lowest comprehensive score, and the ratio with the highest score is selected as the compromise solution. If the differences between the multiple ratios with the highest scores are small, the solution with lower operational difficulty (such as the blending ratio of a certain type of coal sample being easier to control precisely) can be prioritized.

[0075] After identifying the compromise solution, its corresponding blending ratio is determined as the final scheme to guide actual blending operations. Through this evaluation and screening process, the optimal solution can be accurately identified from multiple candidate schemes that meet the constraints. This ensures that the coking risk of blended coal combustion is controllable, while also maximizing alignment with the core needs of the producer, making the final blending ratio safe, economical, and practical.

[0076] In some embodiments, after obtaining the blending ratio of the mixed coal, the method further includes: Determine whether the coal ash after mixing according to the blending ratio meets the target constraint conditions. If the target constraints are not met, an additive addition suggestion will be generated. The additive is used to help increase the flow temperature of the mixed coal ash and reduce the risk of coking. The additive addition recommendations, along with the coal blending ratio, are output to the coal control system to guide the actual blending and additive addition operations.

[0077] After obtaining the blending ratio of the coal, verifying the constraints of the coal ash after blending and adjusting it with additives is a bottom-line guarantee for coking risk control. This step can avoid coking problems caused by the inability of a single blending ratio to meet the target requirements, and make the blended coal finally put into combustion suitable for the coking risk control needs of the target boiler.

[0078] First, the constraints on the mixed coal ash are determined. The core indicators of the target constraints are first identified: the ash fusion point characteristic temperature of the mixed coal ash must be higher than a preset temperature threshold, and the coking risk index must be lower than a preset risk threshold. Then, based on the obtained blending ratio, the comprehensive ash fusion point characteristic temperature and comprehensive coking risk index of the mixed coal ash are calculated through weighted calculation. In the weighted calculation, the blending ratio of each coal sample is used as the weight, and the ash fusion point characteristic temperature and coking risk index of each coal sample when burned individually are weighted and summed to obtain the corresponding indicators of the mixed coal ash. The calculated comprehensive indicators are then compared one by one with the corresponding thresholds of the target constraints to determine whether the mixed coal ash meets the constraints: if the comprehensive ash fusion point characteristic temperature is higher than the preset temperature threshold and the comprehensive coking risk index is lower than the risk threshold, the constraints are considered met; if either indicator fails to meet the requirements, the constraints are considered not met.

[0079] If the mixed coal ash fails to meet the target constraints, additive recommendations are generated. The first step is to analyze the core reasons for not meeting the constraints. If the failure is due to an excessively low overall ash melting point, an additive type that can increase the coal ash melting temperature should be selected. If the failure is due to an excessively high overall coking risk index, an additive type that can inhibit the coking reaction of the coal ash should be selected. Common suitable additives include alumina-based and magnesium oxide-based types. The specific selection needs to be determined based on the composition characteristics of the mixed coal ash—for example, if the proportion of alkaline oxides in the mixed coal ash is high, alumina-based additives should be prioritized to neutralize the alkaline components and increase the melting temperature.

[0080] The second step is to determine the range of additive dosage. Small-batch experiments are conducted to simulate the changes in the ash melting point and coking risk index of the mixed coal ash under different dosages. This identifies the minimum and safe dosage range that ensures the mixed coal ash indicators just meet the target constraints, avoiding insufficient dosage from failing to achieve the desired effect or excessive dosage from increasing costs. Simultaneously, the method of additive addition is clarified. Based on the operating procedures of the coal combustion control system, it is determined whether the additive is added separately to a certain type of coal sample before blending or added synchronously during the blending process. This ensures that the addition operation is compatible with the existing blending process and does not require significant adjustments to equipment operating parameters.

[0081] Finally, a joint output operation is performed, integrating the generated additive addition suggestions (including additive type, dosage range, and addition method) with the previously obtained coal blending ratio into a structured guidance scheme. This integrated guidance scheme is then output to the coal combustion control system. The control system, based on the coal blending ratio in the scheme, completes the mixing of various coal samples and simultaneously controls the dosage and timing of additive addition according to the additive addition suggestions, achieving synchronized execution of coal sample blending and additive addition.

[0082] This operational process effectively compensates for the limitations of single-coal blending ratios, ensuring that the blended coal ultimately entering the boiler for combustion meets the requirements for coking risk control, further improving the stability of coal-fired boiler operation, and avoiding equipment failures or decreased operating efficiency due to coking.

[0083] In some implementations, an iron-calcium synergistic factor can be added as a supplementary feature parameter in the feature parameter extraction stage to more accurately capture the influence of the synergistic fluxing effect of ferric oxide and calcium oxide on the coking characteristics of coal ash, making the feature parameters more comprehensive.

[0084] In practice, after calculating the acid-base ratio, silicon-aluminum ratio, iron-calcium ratio, and total amount of alkaline oxides, the calculation of the iron-calcium synergistic factor is added. Its expression is: Iron-calcium synergistic factor = Ferric oxide content × ln(calcium oxide content + 1). This formula correlates the contents of the two key oxides, transforming them into a quantitative indicator reflecting their synergistic effect. With the addition of this factor, the set of characteristic parameters expands to five categories: acid-base ratio, silicon-aluminum ratio, iron-calcium ratio, total amount of alkaline oxides, and iron-calcium synergistic factor.

[0085] When conducting subsequent importance assessments of feature parameters, the iron-calcium synergistic factor will be included in the assessment scope and input into the random forest model along with the other four types of feature parameters. The random forest model will simultaneously calculate its importance score for the prediction results of the ash melting point characteristic temperature and coking risk index. During the iterative screening process of eliminating low-importance feature parameters and recursively eliminating features, the iron-calcium synergistic factor will also participate in the comparison and screening according to the same standards. If its importance score meets the preset standard and its contribution to the prediction meets the requirements, it will be retained in the optimal feature subset and ultimately used as input to the coking tendency prediction model along with other core feature parameters. If the score does not meet the standard, it will be eliminated to ensure the relevance and effectiveness of the optimal feature subset.

[0086] By introducing the iron-calcium synergistic factor, the characteristic parameters can more comprehensively cover the key factors affecting coal ash coking, enabling the coking tendency prediction model to learn more fully the synergistic effect between coal ash components and further improve the accuracy of the prediction results.

[0087] In some implementations, during the stage of determining the target constraints for blending coal proportions, a new constraint on total sulfur content can be considered to balance coking risk control and environmental protection requirements. Specifically, sulfur content data for each coal sample to be blended can be obtained through laboratory testing. Simultaneously, in conjunction with local environmental emission standards and the environmental emission requirements of the target boiler, a maximum sulfur content threshold for the total sulfur content after blended coal combustion can be set. The constraint that the total sulfur content of the blended coal must not exceed this maximum sulfur content threshold is then established. This, along with the existing constraints on ash fusion point characteristic temperature and coking risk index corresponding to the coking risk threshold, constitutes a complete set of target constraints.

[0088] In the fitness function setting stage of the multi-objective genetic algorithm solution, the total sulfur content constraint needs to be incorporated into the fitness judgment logic. The total sulfur content of the blended coal corresponding to each candidate blending ratio is calculated as the sum of the products of the sulfur content of each coal sample and its corresponding blending ratio. Subsequently, when judging whether the candidate blending ratio meets the constraint requirements, in addition to verifying whether the ash melting point characteristic temperature and coking risk index meet the standards, an additional step is added to compare the total sulfur content with the preset highest sulfur content threshold. Only when all three conditions are met is the candidate blending ratio determined to meet the constraint conditions, and its fitness value calculation is valid.

[0089] After incorporating the total sulfur content constraint, the candidate blending ratios and the final compromise solutions selected by the multi-objective genetic algorithm will simultaneously meet the dual requirements of controllable coking risk and compliance with environmental emission standards. This ensures that the coal blending scheme not only guarantees stable boiler operation but also complies with environmental regulations, avoiding environmental problems caused by excessive sulfur emissions and improving the overall applicability of the scheme.

[0090] In some implementations, during process S6, while simultaneously monitoring the boiler combustion operating parameters in real time, rapid detection data of coal ash composition from incoming coal samples can be collected synchronously. This data is then compared with historical coal sample composition fluctuation ranges to predict whether subsequent incoming coal will show a shift in key coal ash components. If coal quality fluctuations are predicted, there is no need to wait for the coking tendency to increase. Instead, the ash melting point characteristic temperature and coking risk index of each coal sample are recalculated based on the predicted data. Using the original target constraints and optimization algorithms, a backup blending ratio is quickly generated. Once the actual incoming coal quality is confirmed, the process directly switches to the suitable backup blending ratio or slightly adjusts the existing ratio to avoid coking risks caused by coal quality fluctuations.

[0091] This approach allows blending optimization to shift from passive feedback adjustment to proactive predictive adjustment, making it suitable for scenarios where the quality of incoming coal is unstable in actual production. It further shortens the adjustment response time, continuously maintains controllable coking risks during blended coal combustion, and ensures stable fuel costs and combustion efficiency.

[0092] This application also provides a coal blending optimization system based on coal ash characteristic analysis, including: The acquisition module is used to acquire the coal ash composition data of multiple coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. The calculation module is used to calculate the characteristic temperature of ash melting point and coking risk index of each coal sample when it is burned alone, based on coal ash composition data and a pre-trained coking tendency prediction model. The determination module is used to determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. The solution module is used to solve the blending ratio of coal that meets the coking risk control requirements by taking the minimization of fuel cost or the maximization of combustion efficiency as the objective function, combined with the ash melting point characteristic temperature, coking risk index and objective constraints. The output module is used to output the blending ratio of coal that meets the requirements for coking risk control to the coal control system to guide the actual blending operation. The update module is used to monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, the acquisition module is re-executed to the output module to update the blending ratio of coal.

[0093] The system provided in this application has the same technical features as the above-described method embodiments, and therefore can achieve the same technical effects and solve the same technical problems, which will not be repeated here.

[0094] The above description is merely a preferred embodiment and the technical principles employed in this application. This application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions that can be made by those skilled in the art will not depart from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of this application.

Claims

1. A method for optimizing coal blending based on coal ash characteristic analysis, characterized in that, Includes the following steps: S1. Obtain the coal ash composition data of the various coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. S2. Based on the coal ash composition data, calculate the characteristic temperature of ash melting point and coking risk index of each coal sample when it is burned alone using a pre-trained coking tendency prediction model. S3. Determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. S4. Taking the minimization of fuel cost or the maximization of combustion efficiency as the objective function, and combining the ash melting point characteristic temperature, the coking risk index and the objective constraints, the coal blending ratio that meets the coking risk control requirements is solved by an optimization algorithm. S5. Output the blending ratio of coal that meets the requirements for coking risk control to the coal control system to guide the actual blending operation. S6. Monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, return to S1 to S5 and update the blending ratio of coal.

2. The method according to claim 1, characterized in that, The construction process of the coking tendency prediction model includes: Collect historical coal ash composition data and corresponding historical ash fusion point experimental data from historical coal samples; Characteristic parameters are extracted from historical coal ash composition data of historical coal samples. These characteristic parameters include acid-base ratio, silica-alumina ratio, iron-calcium ratio, and total amount of alkaline oxides. A basic prediction model is trained using a random forest model, with the aforementioned feature parameters as input and the historical ash fusion point feature temperature corresponding to the historical ash fusion point experimental data as output. Based on the historical ash melting point characteristic temperature output by the basic prediction model, the corresponding coking risk index is calculated and integrated to form the coking tendency prediction model.

3. The method according to claim 2, characterized in that, Before training the coking tendency prediction model, the following steps are also included: Call the thermodynamic simulation tool, input the historical coal ash composition data of historical coal samples and the corresponding historical combustion atmosphere conditions, and generate the historical liquid phase generation temperature for each historical coal sample. The historical liquid phase formation temperature corresponding to the historical coal sample is correlated and merged with historical coal ash composition data and historical ash melting point experimental data to obtain an enhanced training dataset. The enhanced training dataset is used to train the coking tendency prediction model.

4. The method according to claim 2, characterized in that, The coking tendency prediction model adopts a hybrid architecture, and its specific structure and operation process include: The first layer is the random forest model, which takes the feature parameters or the optimal feature subset as input and outputs the preliminary ash melting point feature temperature and coking risk index. Thermodynamic simulation tools are used to obtain the liquid phase formation temperature of the corresponding coal sample, and the temperature difference between the preliminary ash melting point characteristic temperature and the liquid phase formation temperature is calculated. The second layer is a linear regression correction model, which takes the temperature difference as input and outputs the corresponding correction amount. The initial ash melting point characteristic temperature and coking risk index are combined with the correction amount to obtain the final ash melting point characteristic temperature and coking risk index.

5. The method according to claim 4, characterized in that, After extracting the feature parameters, the method further includes: The importance of each of the aforementioned feature parameters to the prediction results of ash melting point characteristic temperature and coking risk index is evaluated based on the random forest model. Remove feature parameters whose importance is lower than a preset importance standard; A recursive feature elimination method is used to iteratively remove the feature parameters that contribute the least to the model prediction until the optimal feature subset is obtained. The optimal feature subset is then used as the input to the coking tendency prediction model.

6. The method according to claim 1, characterized in that, After calculating the ash melting point characteristic temperature and coking risk index of each coal sample during individual combustion using a pre-trained coking tendency prediction model, the method further includes: By calling a thermodynamic simulation tool and inputting the coal ash composition data and the combustion atmosphere conditions of the boiler, the mineral phase composition and liquid phase formation temperature of the ash residue at different temperatures can be determined. Based on the mineral phase composition and the liquid phase formation temperature, the ash melting point characteristic temperature and coking risk index output by the coking tendency prediction model are corrected to obtain the corrected ash melting point characteristic temperature and coking risk index.

7. The method according to claim 1, characterized in that, The optimization algorithm is a multi-objective genetic algorithm, and the objective function simultaneously includes minimizing fuel cost and maximizing combustion efficiency, wherein: The fuel cost is calculated by summing the products of the unit price of each type of coal and the corresponding blending ratio; The combustion efficiency is calculated based on the volatile matter, calorific value, and corresponding blending ratio of each type of coal.

8. The method according to claim 7, characterized in that, The process of solving the blending ratio of mixed coal using the multi-objective genetic algorithm also includes: The multi-objective genetic algorithm outputs a Pareto optimal solution set after solving the problem. The Pareto optimal solution set includes multiple sets of coal blending ratios that satisfy the objective constraints, as well as the fuel cost and combustion efficiency corresponding to each set of coal blending ratios. The TOPSIS method or a preset decision preference is used to comprehensively evaluate the blending ratios of each group of coal in the Pareto optimal solution set; Based on the evaluation results, a compromise solution is selected, and the coal blending ratio corresponding to the compromise solution is determined as the final coal blending ratio.

9. The method according to claim 1, characterized in that, After obtaining the blending ratio of the mixed coal, the following is also included: Determine whether the coal ash after mixing, corresponding to the blending ratio, meets the target constraint conditions. If the target constraints are not met, an additive addition suggestion is generated. The additive is used to help increase the flow temperature of the mixed coal ash and reduce the risk of coking. The additive addition recommendations, along with the coal blending ratio, are output to the coal combustion control system to guide the actual blending and additive addition operations.

10. A coal blending optimization system based on coal ash characteristic analysis, characterized in that, include: The acquisition module is used to acquire the coal ash composition data of multiple coal samples to be blended. The coal ash composition includes at least silicon dioxide, aluminum oxide, ferric oxide, calcium oxide, magnesium oxide, sodium oxide, and potassium oxide. The calculation module is used to calculate the characteristic temperature of ash melting point and coking risk index of each coal sample when it is burned alone, based on the coal ash composition data and a pre-trained coking tendency prediction model. The determination module is used to determine the target constraints for the blending ratio of coal based on the coking risk threshold of the target boiler. The solution module is used to determine the blending ratio of coal that meets the coking risk control requirements by using an optimization algorithm, with the objective function of minimizing fuel cost or maximizing combustion efficiency, combined with the ash melting point characteristic temperature, the coking risk index and the objective constraints. The output module is used to output the blending ratio of coal that meets the requirements for coking risk control to the coal control system to guide the actual blending operation. The update module is used to monitor the operating parameters of the boiler combustion process in real time. If an increase in coking tendency is detected, the acquisition module is re-executed to the output module to update the blending ratio of coal.

Images