Prediction of temperature and CO concentration fields based on BPNN in low-temperature coal oxidation

Highlights

• Constructed a system to implement a process of low temperature oxidation of coal.

• Selected back propagation neural network and Bayesian Regularization algorithm.

• Predicted the temperature and CO concentration fields.

Abstract

To prevent coal spontaneous combustion, it is critical to accurately simulate and predict the low-temperature oxidation process of coal. In this study, an experiment system was constructed to investigate the temperature and gas concentration of two typical coals during low temperature oxidation. The back propagation neural network (BPNN) was proposed to simulate this process under six factors, including activation energy, void fraction, moisture content, air flow rate, stacking time and location of measuring point. The average relative error of oxygen concentration, temperature and CO concentration are 1.34 %, 1.09 % and 1.15 %, respectively. Besides, the statistical performance indicators are precise. Most importantly, the predicted temperature and gas concentration fields can be utilized to analyze coal spontaneous combustion process which allows us to track down the crucial factors on spontaneous combustion. By applying our BPNN modeling method, the odds of coal spontaneous combustion can be lowered.

Introduction

Coal spontaneous combustion has tremendously affected the safety production and utilization of coal [1,2]. The resulting combustion led to severe environmental pollution, heavy economic losses and terrible casualties [3]. Coals with long storage time are particularly prone to spontaneous combustion, especially such as residual coals left in the gob [4,5], a large amount of standby coals that are stored in coal mines [6]and long-distance transferring of coals by ship, etc [7]. The abovementioned situations have brought great challenges to the safe storage and transportation of coal.

To prevent and control coal spontaneous combustion, it is important to understand the coal spontaneous combustion mechanism [8]. There are many factors affecting the coal spontaneous combustion [[9], [10], [11], [12], [13]], including internal factors such as degree of coalification, coal rock composition, coal porosity, moisture content and particle size of coal, and external factors such as environmental temperature and air flow infiltration. The oxidation process is attributed to a combination of various factors, and it has highly complex non-linear characteristics. It is also very difficult to investigate how these factors affect coal spontaneous combustion directly. Surprisingly, the continuous variation of temperature and the gas concentration can be used to describe the process of coal [14,15]. When the coal pile starts to combust spontaneously, the coal bodies consume oxygen and release reaction heat and a large amount of gas. Owing to the different oxidation heat release intensity and heat storage conditions at different points in the coal pile, the heating rate, oxygen consumption rate and gas release rate of coal body are different, which changes temperature field and gas concentration field in the coal pile. Therefore, analyzing the changes of temperature and gas concentration in the oxidation process allows us to have better understanding of the occurrence and development of coal spontaneous combustion [[16], [17], [18]].

Numerous methods have been proposed to monitor and predict the temperature and gas indexes [19]. Back propagation neural network (BPNN) was used for coal spontaneous combustion modeling, which gives accurate predictions. Xie et al. established BPNN on the basis of the gas monitoring system with beam tube [20]. The temperature of gob can be forecast by using CO and CO₂ concentration values, which significantly improves the prediction accuracy of spontaneous combustion in gob. Also, a random forest (RF) method, a model ensemble method that combines several decision trees was proposed [21]. The RF model accurately predicted the temperature of coal spontaneous combustion when it was applied to the in-situ data. Moreover, Artificial Bee Colony (ABC) optimization algorithm model is highly important and useful for intelligent coal management systems which can improve decision making and reduce risk, especially for short-term forecasts [22]. Furthermore, a new method, DSC Inflection Point (DSCIP), was proposed to determine the coal auto-ignition temperature [23]. The results demonstrated that the heat flux curve of coal spontaneous combustion can be well fitted using Gaussian mixture model. In addition, R/S analysis method was employed to analyze the chaotic characteristics of coal gas indexes from gas drainage pipeline, this kind of prediction method can reduce the misinformation of the coexisting disaster [24]. Many advanced methods can precisely predict the temperature and gas indexes to prevent coal spontaneous combustion [[25], [26], [27]]. However, these researches just monitored and forecast different points inside the coal pile, they did not analyze the whole coal pile.

In this study, we constructed an experiment system to implement a process of low temperature oxidation of coal and collect data points, with controlling six factors as variables, including activation energy, void fraction, moisture content, air flow rate, stacking time and location of measuring point. The BPNN was established based on 72 sets of data acquired from the experimental system and B-R algorithm was selected appropriately. Moreover, the temperature and gas concentration fields were predicted by BPNN to analyze how various factors affect coal spontaneous combustion and how to prevent coal spontaneous combustion by controlling these factors.