Optimized method for predicting coke quality through universal coal blending

Abstract

The invention relates to an optimized method for predicting coke quality through universal coal blending, which comprises the following steps of: 1) segmenting a vitrinite reflectivity histogram of each single coal, and weighting among partitions to obtain a mixed coal vitrinite reflectivity histogram; 2) establishing an inequality set as a constraint condition by using the distribution characteristics of the mixed coal vitrinite reflectance histogram; and 3) for the blended coal meeting the constraint condition, predicting the coke quality by adopting a prediction model based on an artificial intelligence algorithm. According to the method, the inequality set is established by using the characteristics of the distribution of the mixed coal vitrinite reflectance histogram, the range of coal blending optimization is constrained, the prediction model established by the artificial intelligence algorithm is introduced after the constraint condition is met, and the prediction accuracy and universality are improved while the advantages of the artificial intelligence prediction model are brought into full play.

Description

Technical Field

[0001] The present invention relates to the technical field of coking coal blending, and particularly to an optimized method for predicting coke quality by general coal blending. Background Art

[0002] Coking with blended coal means blending several kinds of single coals with different grades in a certain proportion for coking. The advantages of coking with blended coal are as follows: 1) Saving high-quality coking coal and expanding the coking coal source; 2) Making full use of the coking characteristics of various coals to make up for each other's deficiencies and improving the quality of metallurgical coke; 3) Reasonably utilizing coal resources, increasing the yield of coking chemical products and the generation amount of coking gas on the premise of ensuring coke quality; 4) Making full use of local resources, developing coking enterprises according to local conditions.

[0003] The principles of coking with blended coal are as follows: 1) The coke quality reaches the specified index and meets the usage requirements; 2) It will not generate expansion pressure harmful to the furnace wall and cause difficulties in pushing coke; 3) On the premise of meeting the coke quality, as much gas coal as possible is blended to increase the yield of chemical products, as little high-quality coal as possible is blended, and more inferior coal is blended; 4) The ash content and sulfur content in the blended coal are reduced as much as possible; 5) Making full use of local resources, achieving reasonable transportation, reducing costs, and implementing regional coal blending to the greatest extent; 6) Striving to achieve stable blended coal quality, which is beneficial to production and operation.

[0004] The index range of the blended coal is determined according to the coke index. Generally, the main product of a coking plant is grade-II coke, and the deduced blended coal indexes are as follows: moisture 9% - 11%; ash content ≤ 9.5%; sulfur content ≤ 0.85%; volatile matter 28% - 32%; G value ≥ 68; Y value 13 - 15. The pulverizing fineness (the percentage of the weight of the coal with a particle size below 3mm in the total weight of all coal materials after the coal material is pulverized is called the fineness of the blended coal) is about 90%. Coals of different grades have their own characteristics, and their functions in coal blending are also different. If the coal blending scheme is reasonable, the characteristics of various coals can be fully utilized to improve the coke quality. For example: The coking property of gas coal is worse than that of coking coal and fat coal, but its expansion pressure is small, shrinkage is large, and volatile matter is high. When coking alone, due to large shrinkage, the number of coke cracks increases and the coke lump size decreases. However, in coal blending, it can play the role of reducing expansion pressure, increasing shrinkage to make coke pushing smooth, and increasing chemical products and gas. Another example is that the caking property of lean coal is poor, and the abrasion resistance of coke is poor when coking alone, but its shrinkage cracks are few. Blending lean coal in coal blending can increase the coke lump size. The coking property of coking coal is the best, but most of the ash content and sulfur content of coking coal are relatively high. If some low-ash and low-sulfur coal is blended in coal blending, this shortcoming can be overcome. From the above examples, it can be seen that coking with blended coal can give full play to the advantages of various coals and overcome the disadvantages of various coals, thus producing high-quality coke.

[0005] After determining the coal blending indexes, it is necessary to formulate a coal blending plan by combining the characteristics of coal types and following the basic principles of coal blending. For indexes with additivity (such as ash content, volatile matter, moisture content, sulfur content, etc.), additivity calculations are also required. According to traditional formulas, methods such as the volatile matter-binder index method, the volatile matter-maximum fluidity method, and the vitrinite reflectance-inert component-inert capacity prediction method are used for coal blending. In recent years, due to the progress of information technology and artificial intelligence technology, artificial intelligence algorithms have begun to be used to analyze the production data of industrial coke ovens in coking plants, and coke quality prediction models and coal blending optimization prediction models have been established.

[0006] For example, the Chinese invention patent with the authorization announcement number CN 112784396 B discloses "a coke quality prediction method, device and system". According to the detection data of the first blended coal and the target mechanism model, the first predicted quality data of the coke is predicted, wherein the first blended coal is formed into the coke after being refined; a coal petrographic image of the first blended coal is obtained, and the coal petrographic image characterizes the content of each component in the first blended coal; according to the first predicted quality data, the above-mentioned coal petrographic image, the detection data of the first blended coal and the coke quality prediction model, the second predicted quality data of the coke is obtained, wherein the coke quality prediction model is a trained artificial intelligence AI model. Using an artificial intelligence model and combining the parameter data of a large number of blended coals to predict the quality data of coke makes the generalization ability of the coke quality prediction model stronger and the prediction result more accurate.

[0007] The Chinese patent application with the application publication number CN 114692986 A discloses "a coking artificial intelligence coal blending system based on a neural network model", and the implementation process is as follows: S1. Model definition, combining the existing coking data, analyzing the corresponding relationship between the coal blending ratio and coke quality, and calculating a coke quality prediction model. The coke quality prediction model includes an input layer, a hidden layer, and an output layer. The input layer includes several input neurons x1…xi…xd, the hidden layer includes several hidden layer neurons b1…bh…bq, and the output layer includes several output neurons y1…yj…yl. The corresponding weight between the input neuron and the hidden layer neuron is vih, and the corresponding weight between the hidden layer neuron and the output neuron is whj; S2. Training; S3. Evaluation and adjustment; S4. The verified model; S5. Application.

[0008] The Chinese invention patent with the authorization announcement number CN 111950854 B discloses "a method for predicting coke quality indicators based on a multi-layer neural network". Using industrial actual production data, first, the data is cleaned, and gradient boosting trees are used to perform a correlation analysis on the factors affecting coke quality indicators, select the variables most relevant to ash content, sulfur content, M10, M40, CRI, CSR, etc., and then construct training samples, establish a multi-layer neural network prediction model to predict coke quality indicators, and use an intelligent optimization algorithm to optimize the variables in the model to give the final prediction result of coke quality indicators.

[0009] (1) However, it is found in actual applications that the coal blending and coke quality prediction model based on artificial intelligence cannot yet characterize the real coal blending and coking mechanism, resulting in the coking quality after coal blending not being well controlled; the current remedial method is to be constrained by coal blending experts based on experience, but since the experience of coal blending experts varies from person to person, it cannot be universal in each coking plant. Summary of the Invention

[0010] The present invention provides an optimized method for predicting coke quality with universal coal blending, uses the characteristics of the vitrinite reflectance histogram distribution of blended coal to establish an inequality set to constrain the range of coal blending optimization, and then introduces a prediction model established by an artificial intelligence algorithm after meeting the constraint conditions, improving the prediction accuracy and universality while giving full play to the advantages of the artificial intelligence prediction model.

[0011] To achieve the above object, the present invention is implemented by adopting the following technical solutions:

[0012] An optimized method for predicting coke quality with universal coal blending includes the following steps:

[0013] 1) Segment and weight the vitrinite reflectance histogram of each single coal by section and interval to obtain the vitrinite reflectance histogram of the blended coal;

[0014] 2) Use the distribution characteristics of the vitrinite reflectance histogram of the blended coal to establish an inequality set as a constraint condition;

[0015] For top-loading coke ovens, establish the inequality set (1) as follows:

[0016]

[0017] In the inequality set (1), M = 40% - 55%;

[0018] For stamping coke ovens, if the weakly caking coal blended is mainly gas coal, establish the inequality set (2) as follows:

[0019]

[0020] In the inequality set (2), M = 35% - 50%;

[0021] For the stamping coke oven, if the weakly caking coal used is mainly lean coal, the inequality set (3) is established as follows:

[0022]

[0023] In the inequality set (3), M = 35% - 50%;

[0024] 3) For the blended coal that meets the constraint conditions, a prediction model based on the artificial intelligence algorithm is used to predict the coke quality.

[0025] In the said step 1), the range of the average maximum reflectance of the vitrinite of the coking coal is 0.5% - 2.5%; among them, the long-flame coal corresponds to the range of the average maximum reflectance of the vitrinite being 0.5% - 0.65%, the gas coal corresponds to the range of the average maximum reflectance of the vitrinite being 0.65% - 0.80%, the 1 / 3 coking coal or gas-fat coal corresponds to the range of the average maximum reflectance of the vitrinite being 0.8% - 0.9%, the fat coal corresponds to the range of the average maximum reflectance of the vitrinite being 0.9% - 1.2%, the coking coal corresponds to the range of the average maximum reflectance of the vitrinite being 1.2% - 1.5%, the lean coal corresponds to the range of the average maximum reflectance of the vitrinite being 1.5% - 1.7%, the lean-lean coal corresponds to the range of the average maximum reflectance of the vitrinite being 1.7% - 1.9%, and the lean coal corresponds to the range of the average maximum reflectance of the vitrinite being 1.9% - 2.5%.

[0026] Compared with the prior art, the beneficial effects of the present invention are:

[0027] By using the characteristics of the histogram distribution of the vitrinite reflectance of the blended coal to establish an inequality set to constrain the range of coal blending optimization, and then introducing the prediction model established by the artificial intelligence algorithm after meeting the constraint conditions, while giving full play to the advantages of the artificial intelligence prediction model, the accuracy and universality of the prediction are improved. Specific embodiments

[0028] The method for predicting the coke quality by optimizing the general coal blending of the present invention includes the following steps:

[0029] 1) Segment and weight the histogram of the vitrinite reflectance of each single coal by section and interval to obtain the histogram of the vitrinite reflectance of the blended coal;

[0030] 2) Use the distribution characteristics of the histogram of the vitrinite reflectance of the blended coal to establish an inequality set as a constraint condition;

[0031] For the top-charging coke oven, the inequality set (1) is established as follows:

[0032]

[0033] In the inequality set (1), M = 40% - 55%;

[0034] For a stamping coke oven, if the weakly caking coal used is mainly gas coal, the following inequality set (2) is established:

[0035]

[0036] In the inequality set (2), M = 35% - 50%;

[0037] For a stamping coke oven, if the weakly caking coal used is mainly meager-lean coal, the following inequality set (3) is established:

[0038]

[0039] In the inequality set (3), M = 35% - 50%;

[0040] 3) For the blended coal that meets the constraint conditions, a prediction model based on an artificial intelligence algorithm is used to predict the coke quality.

[0041] In the said step 1), the range of the average maximum reflectance of the vitrinite group of the coking coal is 0.5% - 2.5%; among them, the long-flame coal corresponds to the range of the average maximum reflectance of the vitrinite group of 0.5% - 0.65%, the gas coal corresponds to the range of the average maximum reflectance of the vitrinite group of 0.65% - 0.80%, the 1 / 3 coking coal or gas-fat coal corresponds to the range of the average maximum reflectance of the vitrinite group of 0.8% - 0.9%, the fat coal corresponds to the range of the average maximum reflectance of the vitrinite group of 0.9% - 1.2%, the coking coal corresponds to the range of the average maximum reflectance of the vitrinite group of 1.2% - 1.5%, the lean coal corresponds to the range of the average maximum reflectance of the vitrinite group of 1.5% - 1.7%, the meager-lean coal corresponds to the range of the average maximum reflectance of the vitrinite group of 1.7% - 1.9%, and the anthracite corresponds to the range of the average maximum reflectance of the vitrinite group of 1.9% - 2.5%.

[0042] For the method for predicting coke quality by optimizing general coal blending according to the present invention, an inequality set is established by using the characteristics of the histogram distribution of the vitrinite reflectance of the blended coal to constrain the range of coal blending optimization. On this premise, a prediction model established by using an artificial intelligence algorithm is used to predict the coke quality, so as to achieve the purpose of truly representing the mechanism of coal blending for coking and accurately predicting the coke quality.

[0043] The histogram of the vitrinite reflectance of the blended coal should have the following characteristics;

[0044] 1. If there are obvious gaps in the histogram of the vitrinite reflectance of blended coal, it will lead to the discontinuous plastic state of the blended coal, thus unable to ensure good interfacial reactions between coal particles. Through production practice, it is found that the higher the overlap degree of the vitrinite reflectance of each single coal in the blended coal, the closer the distribution diagram of the vitrinite reflectance of the blended coal approaches a normal distribution, the better the compatibility between each single coal, and the better the coal blending effect.

[0045] 2. For top-loading coke ovens, the average maximum vitrinite reflectance of the blended coal is about 1.3%, and the vitrinite reflectance distributions on the left and right sides should approach a normal distribution as much as possible.

[0046] 3. For stamp-charging coke ovens, in order to reduce production costs, the stamp-charging coke oven needs to blend more low-caking coals with low prices compared to the top-loading coke oven. Generally, in order to ensure the quality of coke, more fat coal with strong caking property is appropriately blended during coal blending, which results in the left shift of the average maximum vitrinite reflectance of the blended coal to about 1.1%. If the low-caking coal blended is gas coal, a second peak will form in the range of 0.65% - 0.8% of the vitrinite reflectance of the blended coal. If the low-caking coal blended is lean coal, a second peak will form in the range of 1.7% - 1.9% of the vitrinite reflectance of the blended coal.

[0047] The present invention uses the distribution characteristics of the histogram of the vitrinite reflectance of blended coal to establish an inequality set as a constraint condition. For the blended coal that meets the constraint conditions, a prediction model based on an artificial intelligence algorithm is then used to predict the quality of coke.

[0048] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes, should be covered by the protection scope of the present invention.

Claims

1. A method for predicting coke quality using optimized universal coal blending, characterized in that: The steps include: 1) Divide the reflectance histogram of each single coal vitrinite into sections and weight the sections to obtain the reflectance histogram of the matching coal vitrinite; 2) Using the distribution characteristics of the reflectance histogram of the blended coal vitrinite group, an inequality set is established as a constraint condition; For the top-charged coke oven, the inequality set (1) is established as follows: In inequality set (1), M = 40% to 55%; For the ramming coke oven, if the weakly caking coal used is mainly gas coal, the inequality set (2) is established as follows: In inequality set (2), M = 35% to 50%; For the ramming coke oven, if the weakly caking coal used is mainly lean coal, the inequality set (3) is established as follows: In inequality set (3), M = 35% to 50%; 3) For blended coal that meets the constraints, a prediction model based on artificial intelligence algorithm is used to predict the coke quality.

2. The method for predicting coke quality by using an optimized universal coal blending method according to claim 1, characterized in that: In the step 1), the average maximum reflectivity range of the vitrinite group of the coking coal is 0.5% to 2.5%; wherein, the average maximum reflectivity range of the vitrinite group is 0.5% to 0.65% for long flame coal, the average maximum reflectivity range of the vitrinite group is 0.65% to 0.80% for gas coal, the average maximum reflectivity range of the vitrinite group is 0.8% to 0.9% for 1 / 3 coking coal or gas fat coal, the average maximum reflectivity range of the vitrinite group is 0.9% to 1.2% for fat coal, the average maximum reflectivity range of the vitrinite group is 1.2% to 1.5% for coking coal, the average maximum reflectivity range of the vitrinite group is 1.5% to 1.7% for lean coal, the average maximum reflectivity range of the vitrinite group is 1.7% to 1.9% for lean lean coal, and the average maximum reflectivity range of the vitrinite group is 1.9% to 2.5% for lean coal.

Images