Multi-coal-type coal blending optimization method

Abstract

The invention relates to the technical field of coal washing and processing, and discloses a multi-coal-type coal blending optimization method, which comprises the following steps: S1, a data processing step for collecting raw coal floating and sinking experimental data, S2, a curve fitting step for fitting a B spline selectable curve by using a differential evolution algorithm, S3, a yield estimation step, S4, a quality evaluation step, and S5, a coal blending step. S5, an optimization decision-making step, which is used for constructing a mixed coal selection proportion stochastic optimization model, taking clean coal yield maximization under a target clean coal ash content given condition as a target function, and carrying out online learning and updating on distribution of clean coal product quality indexes under the condition that index information is unknown, and S5, an optimization decision-making step, which is used for constructing a mixed coal selection proportion stochastic optimization model. And under the constraint condition of ensuring the qualified rate of the quality index of the clean coal product, the optimal choice of the multi-coal-type coal blending scheme is obtained. According to the method, the problem of coal blending optimization of multiple coal types in the prior art is effectively solved, and the method has important application value for improving the intelligent level and economic benefits of the coal washing industry.

Description

Technical Field

[0001] This invention relates to the field of coal washing and processing technology, specifically to a method for optimizing the blending of multiple coal types. Background Technology

[0002] Due to differences in quality indicators such as ash and sulfur content, raw coal of different types has different washability curves (i.e., the relationship curve between ash content and cumulative yield of floating matter). In actual production, in order to optimize resource allocation, reduce costs and improve overall efficiency, it is often necessary to mix raw coal of different types in a certain proportion before washing. However, existing technologies lack an effective method to quickly and accurately determine the optimal mixing and washing ratio for multiple types of raw coal. Traditional methods often rely on empirical estimation or simple linear superposition, which is difficult to accurately reflect the washability characteristics of mixed coal. This results in the determined coal blending scheme not maximizing benefits, and may even cause economic losses due to substandard product quality. Therefore, how to achieve efficient and coordinated washing of raw coal of different types and provide a method to quickly and accurately determine the mixed coal scheme that maximizes benefits is the technical problem to be solved by this invention.

[0003] The shortcomings of existing multi-coal blending optimization methods are: In patent document CN107008566B, a dry deep classification method is used by double-layer elastic grading screen, combined with gravity separator separation, to achieve efficient separation of thermal coal. It has the advantages of high feed ratio, high separation accuracy and strong adaptability. However, this scheme is mainly for thermal coal and does not involve the multi-coal blending optimization problem of coking coal. It does not fully consider the impact of raw coal quality fluctuations on the washing effect, resulting in the blending scheme being out of touch with actual production. In patent document CN118406511A, a coking economic evaluation system for coal is constructed by principal component analysis and entropy weight method. Combined with a coal blending coke quality prediction system, coking coal is comprehensively optimized from both economic value and product quality perspectives. However, this method does not consider the impact of raw coal quality fluctuations on washing and beneficiation effects, lacks direct optimization of clean coal yield, and lacks dynamic prediction and qualification rate estimation of clean coal product quality indicators, making it difficult to guarantee the quality stability of coal blending schemes. In patent document CN115392561A, a coal blending and washing model based on coking characteristics determines the optimal coal blending ratio based on the principle of maximizing clean coal yield and ensuring that clean coal indicators are qualified. However, this method does not solve the problem of dynamic prediction when coal quality information is unknown, and the optimization of the coal blending ratio depends on fixed constraints, which is difficult to adapt to the fluctuations in coal quality in actual production. The optimization model is mostly based on fixed constraints and cannot adapt to the real-time updates and uncertainties of coal quality information. In patent document CN105131996B, the quality of coke was stabilized by classifying and blending single types of coal through microstructure analysis of coking coal. However, this method is only applicable to specific types of blended coking coal and has limited versatility. Summary of the Invention

[0004] This invention provides a method for optimizing the blending of multiple coal types, in order to solve the technical problems of inaccurate blending, unstable quality, and poor adaptability in the existing methods for optimizing the blending of multiple coal types proposed in the background.

[0005] This invention provides a method for optimizing the blending of multiple coal types, the optimization method comprising the following steps: S1, Data Processing Steps, used to collect raw coal float-sink test data and establish an optionality curve between ash content and cumulative yield of floating matter using B-spline basis functions; S2. Curve fitting step: Use the differential evolution algorithm to fit the B-spline selectivity curve to obtain the parameters corresponding to the optimal curve; S3. Yield estimation step: Under the given specific value of target clean coal ash content, estimate the yield of clean coal for single coal type and the yield of clean coal for mixed coal type based on the parameters corresponding to the optimal curve. S4. Quality assessment step: The incremental learning Bayesian estimation method is used to learn and update the distribution of clean coal product quality indicators online when the indicator information is unknown, and to estimate the pass rate of clean coal product quality indicators under different coal blending schemes. The quality assessment step is based on the yield of clean coal from mixed coal. S5. Optimize the decision-making process by constructing a stochastic optimization model for the proportion of mixed coal to be selected. With the objective function of maximizing the yield of clean coal under the given condition of the target clean coal ash content, the optimal choice of multi-coal blending scheme is obtained under the constraint of ensuring the pass rate of clean coal product quality indicators. The optimization decision-making process is based on the pass rate of clean coal product quality indicators.

[0006] Preferably, the domain of the gray node sequence of the B-spline basis function is: ,in This represents the minimum ash content sample value in the buoyancy and sinking test data. For the maximum gray sample value, the B-spline order is set to 2.

[0007] Preferably, when establishing the optional curve, constraints need to be applied to ensure that the curve is monotonically increasing. Among the constraints, when the function is continuously differentiable, its derivative is required to be greater than or equal to 0; when the function is discrete, the difference between adjacent points is required to be greater than or equal to 0.

[0008] Preferably, in the S2 curve fitting step, the process of fitting the B-spline selectivity curve using the differential evolution algorithm is as follows: generate the B-spline curve based on the control points, calculate the sum of squared errors between the estimated and actual values ​​of the floating material yield, and use the differential evolution algorithm to optimize the control points of the B-spline curve to minimize the objective function.

[0009] Preferably, during the algorithm execution, S2 uses an exponential crossover scheme when performing crossover operations on individuals in the population. That is, it randomly selects a starting point k and a length L, and assigns the values ​​of L consecutive gene positions starting from position k to the experimental individual.

[0010] Preferably, during curve fitting, individuals need to be sorted by fitness, following the feasibility criterion: Individuals that satisfy the monotonically increasing constraint are better than those that do not. If all conditions are met, the one with higher fitness is better; If none of the conditions are met, the one with the lower degree of constraint violation is better.

[0011] Preferably, the yield of washed clean coal from mixed coal is calculated by weighting the yield of washed clean coal from each type of coal with its corresponding mixed washing ratio.

[0012] Preferably, the quality indicators of refined coal products specifically include sulfur content, volatile matter, moisture content, and caking degree.

[0013] Preferably, the distribution of online learning updates is performed when the index information is unknown. Specifically, when a new detection value of the d-th quality index of raw coal C is obtained, the posterior distribution parameters are updated using the following formula: and .

[0014] Preferably, in the S2 curve fitting step, the initial population size of the differential evolution algorithm is set to 20, the differential weight is set to 0.6, and the termination threshold is set to 1e-5. In the S5 optimization decision step, the initial population size of the differential evolution algorithm is set to 20, the differential weight F is set to 0.6, the crossover probability CR is set to 0.4, and the termination threshold is set to 1e-5. Individuals are crossovered, specifically using a binomial distribution crossover. That is, for each gene locus of an individual, a random number rand between 0 and 1 is generated. If rand < CR, then the gene locus of the mutant individual is adopted; otherwise, the gene locus of the parent individual is retained.

[0015] Compared with the prior art, the beneficial effects of the present invention are: By fitting the B-spline selectivity curve using the differential evolution algorithm, the selectivity characteristics of single coal types and mixed coals can be accurately characterized, overcoming the extensiveness of traditional empirical methods. Based on this, a stochastic optimization model for the mixed coal selection ratio is constructed. This model optimizes the coal blending scheme with the goal of maximizing clean coal yield while ensuring the quality of clean coal products. This allows for the rapid and accurate determination of the most efficient multi-coal blending scheme, significantly improving the scientific nature and efficiency of coal blending decisions. This invention employs an incremental learning Bayesian estimation method, which can learn and update the distribution of quality indicators (such as sulfur and volatile matter) of clean coal products online under unknown conditions, and estimate the pass rate of product quality indicators under different coal blending schemes. This enables the optimization decision-making process to fully consider the volatility and uncertainty of quality indicators in actual production, and ensure that the final scheme meets strict quality constraints, thereby reducing the economic risks caused by substandard product quality. By achieving synergistic optimization of raw coal of different types, this invention can maximize the yield of clean coal while ensuring stable product quality, thereby improving the overall utilization efficiency of coal resources. For coal washing enterprises, this means that under the same washing conditions, they can obtain higher clean coal output and better product quality, which directly improves the economic benefits and market competitiveness of the enterprises. The technical solution provided by this invention effectively solves the problem of multi-coal blending optimization in the prior art, and realizes the determination of a fast, accurate and robust coal blending scheme that maximizes benefits. It has important application value for improving the intelligent level and economic benefits of the coal washing and beneficiation industry. Attached Figure Description

[0016] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, such as Figure 1 As shown in the figure, an embodiment of the present invention discloses a multi-coal blending optimization method. The method achieves the optimal selection of multi-coal blending scheme through five steps: data processing, curve fitting, yield estimation, quality assessment and optimization decision-making, so as to maximize the yield of clean coal while ensuring the quality of clean coal products.

[0019] S1. Data Processing Steps: First, collect float-sink test data for each coal type. The float-sink test data includes ash values ​​at different density levels and the corresponding cumulative yield of floating matter. For example, for coal type A, its ash values ​​at density levels of 1.3 g / cm³, 1.4 g / cm³, 1.5 g / cm³, 1.6 g / cm³, and 1.8 g / cm³ are 5.2%, 7.8%, 12.5%, 18.3%, and 25.6%, respectively, with corresponding cumulative yields of floating matter of 85%, 70%, 45%, 20%, and 5%. Subsequently, an optionality curve between ash content and cumulative yield of floating matter is established using B-spline basis functions. In this embodiment, the order of the B-spline basis function is set to 2, and the domain of the ash node sequence is the minimum ash sample value in the float-sink test data. (e.g., 5.2%) and the maximum ash sample value (e.g., 25.6%). To ensure the physical meaning of the curve, constraints are applied when establishing the selectivity curve to ensure that the curve is monotonically increasing, that is, the ash content increases monotonically with the increase of the cumulative yield of floating matter. Among the constraints, when the function is continuously differentiable, its derivative is required to be greater than or equal to 0. When the function is discrete, the difference between adjacent points is required to be greater than or equal to 0.

[0020] S2. Curve fitting step: The B-spline selectivity curve described above is fitted using the differential evolution algorithm to obtain the parameters corresponding to the optimal curve. Specifically, a B-spline curve is generated based on the control points, the sum of squared errors between the estimated and actual float yield values ​​is calculated, and the control points of the B-spline curve are optimized using the differential evolution algorithm to minimize the objective function. The formula for calculating the sum of squared errors is:

[0021] in, Let be the actual cumulative yield of floating objects at the i-th buoyancy test data point. To estimate the cumulative yield of floating objects using B-spline curves, where n is the number of experimental data points, the parameters of the differential evolution algorithm are set as follows: initial population size of 20, differential weight (F) of 0.6, and termination threshold of 1e-5. During algorithm execution, an "exponential" crossover scheme is used when performing crossover operations on individuals in the population: a starting point (k) and a length (L) are randomly selected, and the values ​​of consecutive (L) gene loci from position (k) are assigned to the experimental individual. When sorting individuals by fitness, a feasibility criterion is followed. Individuals that satisfy the monotonically increasing constraint are better than those that do not. If all conditions are met, the one with higher fitness is better; If none of them are satisfied, the one with the lower degree of constraint violation is better; Through iterative optimization, the optimal B-spline curve parameters are finally obtained, minimizing the fitting error between the curve and the experimental data.

[0022] S3. Yield estimation step: Under the given specific value of target clean coal ash content, based on the parameters corresponding to the optimal curve, estimate the clean coal yield of a single coal type and the clean coal yield of a mixed coal type. For a single coal type, directly find the cumulative yield of floating matter corresponding to the target ash content through the ash-floating matter cumulative yield curve, which is the clean coal yield of that coal type. For example, for coal type A, when the target clean coal ash content is 8.0%, the yield of its washed clean coal is found to be 65% through a curve. For blended coal, assuming the blended coal consists of coal type A and coal type B in a certain proportion... and , If the mixture is mixed, the yield of the mixed coal into washed clean coal is calculated by weighted summation: ,in, and The yield of washed clean coal for coal types A and B are respectively.

[0023] S4. Quality assessment steps: Incremental learning Bayesian estimation method is used to learn and update the distribution of clean coal product quality indicators online when the indicator information is unknown, and to estimate the pass rate of clean coal product quality indicators under different coal blending schemes. The specific clean coal product quality indicators include sulfur content, volatile matter, moisture, and caking degree. Taking sulfur content as an example, it is assumed that the prior distribution mean of the (d)th quality indicator (sulfur content) of raw coal C is... The variance is The variance of the detected values ​​is When a new sulfur content (x) value is obtained for raw coal C, the posterior distribution parameter is updated using the following formula:

[0024] Through the above update, the posterior distribution of sulfur content is obtained, and then the pass rate of sulfur content index under different coal blending schemes (such as the probability of sulfur content ≤1.0%) is calculated. Similar processing is performed on other quality indicators (volatile matter, moisture, caking) to obtain the pass rate of clean coal product quality indicators.

[0025] S5. Optimize the decision-making process by constructing a stochastic optimization model for the proportion of mixed coal in the blending process. The objective function is to maximize the yield of clean coal under the given condition of target clean coal ash content. Under the constraint of ensuring the pass rate of clean coal product quality indicators, the optimal selection of the multi-coal blending scheme is obtained. The objective function is:

[0026] Where m represents the quantity of coal types. Let i be the mixing and washing ratio of the i-th type of coal. Let be the yield of the washed clean coal of the i-th type of coal; The constraints include: Furthermore, the pass rate of the refined coal product quality indicators is not lower than a preset threshold (e.g., 95%). The differential evolution algorithm is used to solve this optimization model, with the following parameter settings: initial population size of 20, differential weight (F) set to 0.6, crossover probability (CR) set to 0.4, and termination threshold set to 1e-5. Binomial crossover is used when performing crossover operations on individuals. For each gene locus of an individual (i.e., the mixing ratio of each type of coal), a random number (rand) between 0 and 1 is generated. If (rand < CR), the gene locus of the mutated individual is used; otherwise, the gene locus of the parent individual is retained. Through iterative optimization, the optimal mixed coal selection ratio (p1, p2, ..., p...) is finally obtained. m (i.e., the optimal coal blending scheme). Example

[0027] This embodiment is basically the same as Embodiment 1, except that in the data processing step, the order of the B-spline basis function is set to 3 instead of 2 in Embodiment 1. The parameters and operations of other steps are the same as in Embodiment 1. By adjusting the order of the B-spline, the flexibility of curve fitting can be further optimized to adapt to the selectivity curve characteristics of different coal types. Example

[0028] This embodiment is basically the same as Embodiment 1, except that in the curve fitting step, the termination threshold of the differential evolution algorithm is set to 1e-4 instead of 1e-5 in Embodiment 1. The parameters and operations of other steps are the same as in Embodiment 1. By adjusting the termination threshold, a balance can be achieved between computational efficiency and fitting accuracy, which is suitable for scenarios with high real-time requirements. Example

[0029] This embodiment is basically the same as Embodiment 1, except that in the quality assessment step, the quality indicators of the clean coal product only include sulfur and moisture, instead of sulfur, volatile matter, moisture and caking degree in Embodiment 1. The parameters and operations of other steps are consistent with Embodiment 1. By reducing the number of quality indicators, the calculation process can be simplified, which is suitable for scenarios with relatively simple product quality requirements. Example

[0030] This embodiment is basically the same as Embodiment 1, except that in the optimization decision step, the crossover probability (CR) of the differential evolution algorithm is set to 0.6 instead of 0.4 in Embodiment 1. The parameters and operations of other steps are the same as in Embodiment 1. By adjusting the crossover probability, the global search capability of the algorithm can be enhanced, and it can avoid getting trapped in local optima. Example

[0031] This embodiment is basically the same as Embodiment 1, except that in the yield estimation step, the weighted geometric mean is used to calculate the yield of the mixed coal into the washed clean coal, instead of the weighted arithmetic mean in Embodiment 1. The specific formula is as follows: Where m is the quantity of coal type, p i Let y be the mixing and washing ratio of the i-th type of coal. i Let be the yield of the washed clean coal of the i-th type of coal. The parameters and operations of other steps are the same as in Example 1. By using a weighted geometric average, the nonlinear relationship between the yields of different coal types can be reflected more accurately. Example

[0032] This embodiment is basically the same as Embodiment 1, except that in the quality assessment step, the incremental learning Bayesian estimation method uses Gaussian process regression instead of Gaussian distribution update in Embodiment 1. Specifically, the Gaussian process regression model is used to learn the relationship between quality indicators and ash content and yield, and to predict the distribution of quality indicators under different coal blending schemes. The parameters and operations of other steps are consistent with Embodiment 1. By using Gaussian process regression, nonlinear relationships can be handled more flexibly, and the accuracy of quality assessment can be improved. Example

[0033] This embodiment is basically the same as Embodiment 1, except that in the optimization decision-making step, the objective function is the weighted sum of clean coal yield and quality indicator pass rate, rather than maximizing the single clean coal yield as in Embodiment 1. The specific formula is as follows: ,in, This is a weighting coefficient (e.g., 0.7). R represents the yield of washed clean coal from mixed coal, and R represents the pass rate of clean coal product quality indicators. The parameters and operations of other steps are consistent with those in Example 1. By adopting a weighted sum objective function, a balance can be achieved between clean coal yield and quality, which is suitable for scenarios where both have high requirements. Content not described in detail in this specification belongs to the prior art known to those skilled in the art. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for optimizing coal blending of multiple coal types, characterized in that, The optimization method includes the following steps: S1. Data processing steps: Collect raw coal floating and sinking test data, and use B-spline basis functions to establish an optionality curve between ash content and cumulative yield of floating matter; S2. Curve fitting step: Use the differential evolution algorithm to fit the B-spline selectivity curve to obtain the parameters corresponding to the optimal curve; S3. Yield estimation step: Under the given specific value of target clean coal ash content, estimate the yield of clean coal for single coal type and the yield of clean coal for mixed coal type based on the parameters corresponding to the optimal curve. S4. Quality assessment step: The incremental learning Bayesian estimation method is used to learn and update the distribution of clean coal product quality indicators online when the indicator information is unknown, and to estimate the pass rate of clean coal product quality indicators under different coal blending schemes. The quality assessment step is based on the yield of clean coal from mixed coal. S5. Optimize the decision-making steps by constructing a stochastic optimization model for the proportion of mixed coal to be selected. The objective function is to maximize the yield of clean coal under the given condition of the target clean coal ash content. Under the constraint of ensuring the pass rate of the quality indicators of clean coal products, the optimal selection of the multi-coal blending scheme is obtained. The optimization decision-making steps are based on the pass rate of the quality indicators of the clean coal products.

2. The method for optimizing multi-coal blending according to claim 1, characterized in that: The domain of the gray node sequence of the B-spline basis function is: ,in This represents the minimum ash content sample value in the buoyancy and sinking experiment data. For the maximum gray sample value, the B-spline order is set to 2.

3. The method for optimizing multi-coal blending according to claim 1, characterized in that: When establishing the optionality curve, constraints need to be applied to ensure that the curve is monotonically increasing. Among the constraints, when the function is continuously differentiable, its derivative is required to be greater than or equal to 0; when the function is discrete, the difference between adjacent points is required to be greater than or equal to 0.

4. The method for optimizing multi-coal blending according to claim 1, characterized in that: In the S2 curve fitting step, the process of fitting the B-spline selectivity curve using the differential evolution algorithm is as follows: generate the B-spline curve based on the control points, calculate the sum of squared errors between the estimated and actual values ​​of the floating matter yield, and use the differential evolution algorithm to optimize the control points of the B-spline curve to minimize the objective function.

5. The method for optimizing multi-coal blending according to claim 1, characterized in that: During the execution of the algorithm, S2 uses an exponential crossover scheme when performing crossover operations on individuals in the population. That is, it randomly selects a starting point k and a length L, and assigns the values ​​of L consecutive gene positions starting from position k to the experimental individual.

6. The method for optimizing multi-coal blending according to claim 1, characterized in that: During curve fitting, individuals need to be sorted by fitness, following the feasibility criterion: Individuals that satisfy the monotonically increasing constraint are better than those that do not. If all conditions are met, the one with higher fitness is better; If none of the conditions are met, the one with the lower degree of constraint violation is better.

7. The method for optimizing multi-coal blending according to claim 1, characterized in that: The yield of washed clean coal from mixed coal is calculated by weighting the yield of washed clean coal from each type of coal with its corresponding mixed washing ratio.

8. The method for optimizing multi-coal blending according to claim 1, characterized in that: The specific quality indicators for refined coal products include sulfur content, volatile matter, moisture content, and caking degree.

9. The method for optimizing multi-coal blending according to claim 1, characterized in that: The distribution of online learning updates is performed when the indicator information is unknown. Specifically, when a new detection value of the d-th quality indicator of raw coal C is obtained, the posterior distribution parameters are updated using the following formula: and .

10. The method for optimizing multi-coal blending according to claim 1, characterized in that: In the S2 curve fitting step, the initial population size of the differential evolution algorithm is set to 20, the differential weight is set to 0.6, and the termination threshold is set to 1e-5. In the S5 optimization decision step, the initial population size of the differential evolution algorithm is set to 20, the differential weight is set to 0.6, the crossover probability CR is set to 0.4, and the termination threshold is set to 1e-5. Individuals are crossovered, specifically using a binomial distribution crossover. That is, for each gene locus of an individual, a random number rand between 0 and 1 is generated. If rand < CR, then the gene locus of the mutant individual is adopted; otherwise, the gene locus of the parent individual is retained.

Images