Abstract:
Tar yield is the most important coal quality parameter for low temperature carbonization of coal, which determines the clean utilization direction of tar-rich coal. However, due to various reasons, the test of coal tar yield in the stage of coal geological exploration is very limited, which greatly restricts the fine evaluation and efficient utilization of tar-rich coal. In order to improve the scientificity and accuracy of the fine evaluation of tar-rich coal, based on the previous 1073 groups of coal quality data in Jurassic coal field of northern Shaanxi, and screened out 141 groups of data with complete coal rock and coal quality parameters such as maceral, industrial analysis, elemental analysis and grey component analysis. The BP neural network algorithm was used to construct the prediction model of tar yield of 20 coal quality indexes and the prediction model of tar yield based on industrial analysis, and the accuracy and rationality of the prediction model were analyzed and evaluated. The results show that: The mean square error of the final training of the prediction model based on 20 coal-rock and coal-quality indexes is 0.30, and the mean absolute error of the prediction results of the test set data is 0.65. The mean square error of the final training of the prediction model was 1.07, and the mean absolute error of the test set data was 1.35. The average absolute error of the two prediction models is 0.84 and 1.34, respectively, indicating that the prediction model established by 20 coal, rock and coal quality indexes has higher goodness of fit and generalization performance than that established by 4 industrial analysis coal quality indexes. The importance of 20 coal rock and coal quality indexes in the prediction model was further quantitatively analyzed by SHAP algorithm. The results showed that vitrinite, hydrogen, ferric oxide, moisture, volatile matter, carbon, exinite and oxygen content were the positive influencing factors of tar yield, and aluminum oxide, inertinite, fixed carbon, ash and silica content were the negative influencing factors of tar yield. The intrinsic relation between coal quality and tar yield in the model accords well with the basic understanding of geological factors affecting tar yield. The tar yield prediction model can be well applied to the prediction of tar yield in Jurassic coal field in northern Shaanxi, and can provide support for clean and efficient utilization of oil-rich coal in northern Shaanxi.