Analysis and Prediction of the Meteorological Characteristics of Dust Concentrations in Open-Pit Mines

Abstract

Based on the dust concentration data and meteorological environment data monitored at the open-pit mine site, the characteristics of dust concentration and the influence of temperature, humidity, wind speed, air pressure and other meteorological conditions on dust concentration were analyzed, and the causes of the change of dust concentration were clarified. Meanwhile, a dust concentration prediction model based on LSTM neural network is established. The results show that the dust concentration of the open-pit mine is high in March, November and the whole winter, and it is low in summer and autumn. The daily variation of humidity and temperature in different seasons showed the trend of “herringbone” and “inverted herringbone”, respectively. In addition, the wind speed was the highest in spring and the air pressure distribution was uniform, which basically maintained at 86–88 kPa. The peak humidity gradually deviates with each month and is obviously affected by seasonality. The higher the humidity, the lower the temperature and the higher the concentration of dust. In different seasons, the wind speed is the highest around 20:00 at night, and the dust is easy to disperse. The R² values of PM2.5, PM10 and TSP concentrations predicted by LSTM model are 0.88, 0.87 and 0.87, respectively, which were smaller than the MAE, MAPE and RMSE values of other prediction models, and the prediction effect was better with lower error. The research results can provide a theoretical basis for dust distribution law, concentration prediction and dust removal measures of main dust sources in open-pit mines.

1. Introduction

Coal is supporting the rapid economic development of China, but coal mining and utilization processes also cause a series of environmental problems [1,2,3], such as greenhouse effects, climate change and dust pollution [4,5,6]. Strong disturbances from mining activities of large open-pit coal mines inevitably lead to the degradation of the fragile ecological environments of mining areas, and dust pollution has always been an associated problem of coal open-pit mining [7,8,9]. The associated changes in dust particle concentrations are closely related to meteorological factors and are affected by humidity, wind speed, air pressure, temperature and other meteorological conditions [10,11,12]. The prediction of dust concentrations in open-pit mines can prevent and control pollutant creation in time and be used to distinguish the diffusion and pollution levels of dust particles, which is a key research approach of efficient dust suppression technology. Many scholars in China and abroad have proposed a variety of dust concentration prediction models, such as the gray prediction model [13,14,15], BP neural network model [16,17], particle sswarm optimization algorithm model [18,19,20,21], random forest model [22,23,24], convolutional neural network and cyclic neural network [25,26] and Markov model [27,28]. Based on the prediction models used, the prediction of air quality concentrations and classification of air quality in relevant cities have been carried out [29,30].

However, the dust concentrations of open-pit mines are associated with few meteorological factors and the change characteristics of dust concentrations in different seasons, and the existing prediction models are still limited in the prediction of dust concentration. It is difficult to analyze the nonlinear relationship between concentrations and various influencing factors using the ordinary linear prediction model. In addition, the traditional neural network model lacks the memory functions of data [31,32] and fails to pay timely attention to information of the previous moment and apply it to the next moment. The long-term memory neural network (LSTM) can solve long-term data dependence problems based on its special cyclic neural network function [33,34,35].

Therefore, the meteorological and dust concentration monitoring data of open-pit coal mines are taken as the dataset by which this study analyzes the distribution characteristics of on-site dust and meteorological parameters, and the Pearson correlation coefficient model is used to analyze the correlation between meteorological factors and pollutant concentrations. Meteorological elements such as wind speed, relative humidity, air pressure and temperature and previous concentrations of pollutants were selected as characteristic factors to build a dust concentration prediction model based on LSTM neural network, which provides important theoretical significance and practical value for the study of dust distribution laws, concentration prediction and dust removal measures of the main dust sources in open pit mines.

2. Research Area and Data Source

2.1. Study Area

The study mining area is located in the northern part of Shanxi Province as shown in Figure 1. It is a loess hilly landform bordering the Loess Plateau and northern soil and rock mountains. It is a semi-arid climate zone with a warm temperate zone and dry climate. The spring, summer, autumn and winter are four distinct seasons; the difference between day and night temperatures is significant, the winter is long and cold, the summer is short and hot, the temperatures are variable, and the spring and winter are characterized by windy conditions and serious sandstorms. Precipitation mainly occurs in July, August and September. The stope form and surrounding environment of open-pit coal mines in the area have typical characteristics of semi-open space; open-pit mines are spatially divided into stopes and inner dumps (or outer dumps). A stope is deep in the middle, and the depth from the bottom to the ground is approximately 260 m. The four sides are composed of a working wall (approximately 12°), a dumping wall (approximately 11°) and two end walls (approximately 32°), and the pit plane size is approximately 3 km*2.6 km.

The production links of open-pit mines in semi-open spaces generally include perforation, blasting, mining, loading, transportation, discharge and other links, as shown in Figure 2. Dust forms to varying degrees due to the crushing or loading, transportation and unloading of materials. Due to the large number of vehicles and long transportation times involved in the transportation network, the amount and scope of dust on the road surface are significant. In addition, open-pit mines are directly exposed to atmospheric environmental conditions, such as rain, snow and freezing weather, and natural factors such as wind speed, temperature and humidity, open-pit mine production material characteristics, road quality, operating equipment, operation processes, preventive dust control measures and so on affect the composition, content, concentration, structure and size of open-pit dust, diffusion ranges and pollution levels.

2.2. Data Sources

The data for dust concentration were obtained from the surface monitoring system of a solar multipoint network installed in open pit mine. The stope of the open-pit coal mine is exposed to direct contact with the atmosphere for a long time in the open space, and the monitoring period is long. Therefore, the selected ground monitoring equipment should exhibit temperature, humidity and tide resistance. As shown in Figure 3, the front end of the dust monitoring system is a single-point monitoring substation deployed on a fixed point or equipment in the pit, and the back end is a background data management platform deployed in a computer room. Each single-point substation is composed of a data collector mounted on a pole, a monitoring sensor for meteorological and airborne particulate matter, and a wireless transmission system unit powered by solar panels. Information collected by the single-point monitoring substation includes dust concentration (PM2.5, PM10 and TSP), as well as meteorological characteristics factors (ambient temperature, humidity, air pressure and wind speed). Lora wireless networking can be conducted between single sensors. Data can be uploaded through the host’s sensor summary upload platform. The sensor can also include a built-in SIM card to upload data separately. The data management platform is a network platform with internet architecture with monitoring functions of each substation and recording, query, statistical and report output functions of the collected data. The data collection period ran from 1 August 2020 to 31 July 2021. Monitoring data are used to analyze dust concentration distribution characteristics in the open pit mine, providing theoretical support for efficient dust suppression technology used in this area.

2.3. Methodology

Various environmental and meteorological factors affecting dust concentrations were determined by extensive field investigation and analysis, including temperature (T), humidity (H), wind speed (W), air pressure (P), etc. The traditional cyclic neural network algorithm used for the backpropagation of updating parameters and the gradient can easily disappear or explode since the gradients was continuously multiplied according to the time step.

The Long Short-Term Memory neural network (LSTM) based on a special cyclic neural network algorithm is widely used to predict the PM concentration and has the excellent function to study long term dependency problems [32]. The forecasting model is made up of a set of memory blocks and is adopted to predict the dust concentration over the next 1 h according to the particle concentration and meteorological factors of the previous stage. The complete architecture of the model is described in Figure 4. The LSTM network is made up of three parts: input layer, hidden layer and output layer. The original data in the input layer contain seven features (TSP, PM10, PM2.5, W, H, P and T). The hidden layer includes a LSTM unit and 32 neurons, and, through forward operation, information for the previous 1 h is continuously transmitted backward in the form of a memory stream, affecting the processing of each input datum and the output of each stage. The output layer of the model is used to obtain the final forecasting result of TSP, PM10 and PM2.5.

3. Results and Analysis

3.1. Dust Concentration Distribution Characteristics

The characteristics of dust concentration are studied by actual data from the monitoring system of the open-pit mine which were processed into daily mean to meet the national environmental protection department. According to definitions from climatology, seasons are divided (in the Northern Hemisphere, the spring season runs from March to May, the summer season runs from June to August, the autumn season runs from September to November, and the winter season runs from December to February). The daily average concentration of PM2.5, PM10 and TSP in each month is obtained and the distribution characteristics were analyzed as shown in Figure 5. (Due to equipment damage occurring from the end of June to the beginning of July, data were not obtained for this period.).

It can be seen from Figure 5 that the concentrations of PM2.5, PM10 and TSP are highly correlated, and most of the annual monitoring data are lower than the coal industry pollutant emission standard (1000 μg/m³) [36], only in some periods of spring and winter are higher than the emission standard. In the spring, the dust concentration is the highest in March and reached a maximum in the middle of March due to a large number of loose particles being suspended by strong wind-sand weather in the spring. The dust concentration monitored at this time is the result of the combined effects of open-pit mine operations and strong wind-sand weather. The dust concentration gradually decreased in April and May, and low probability outlier events appeared on a few days in May. The distribution of dust concentrations in June, July and August in the summer is stable, and the overall dust concentration is relatively low due to the combined effects of high temperatures and heavy rainfall in the summer. In autumn, the dust concentration is low in September and October and increased sharply in November. As the temperature of the open-pit mining area decreased in November, dust diffusivity declined, resulting in a gradual increase in dust levels. Dust concentrations continued to increase in the winter, and an inversion layer easily formed on the mine surface in the winter, easily leading to the accumulation of dust and exceeding the monthly concentration limit. The concentration is highest in December and gradually decreased in January and February.

The results show that dust concentrations are high in March, November and December, and dust—mainly pit production link and surrounding dust—easily accumulated in the winter. Dust blown by strong sand weather in March generated high concentrations in a short period of time, so the mine must take corresponding measures to ensure the normal operation of the open pit mine.

3.2. Analysis of the Characteristics of Meteorological Factor Changes

According to previous research results and investigations of mine sites, meteorological factors affecting dust concentrations mainly include four indicators: humidity, temperature, wind speed and air pressure, and the distribution characteristics are obtained as shown in Figure 6, Figure 7, Figure 8 and Figure 9.

As seen from the humidity and temperature distribution diagram, the daily humidity distribution in different seasons has a consistent change pattern and follows a “herringbone” trend of rising, falling and then rising. Humidity levels are high in July, August and September, and the lowest humidity levels in the winter. The time corresponding to the peak humidity is at approximately 12:00 most of the time and the minimum humidity is at approximately 20:00. The humidity extremum gradually deviates with each month, and humidity is clearly affected by seasonality. In addition, the distribution characteristics of daily temperature in different seasons roughly follow an “inverted herringbone” trend of falling, rising and then falling. The temperatures gradually decrease from 0:00 to 12:00, then gradually increase from 12:00 to 20:00 and gradually decreases again after 20:00. The average daily temperature in different spring months is above 0 °C. In the summer, the temperature is gradually rising, and the minimum temperature is at approximately 10:00. Temperatures gradually decrease in the autumn and winter after September and then increase after February. The daily temperature in the winter is similar to that in the spring and summer. The analysis results show that temperature and humidity in different seasons are inversely proportional, and the higher the humidity level becomes, the lower the temperature becomes, and all have a positive effect on the concentration of dust particles, resulting in an increase in the concentration of dust particles.

It can be seen from Figure 8 that wind speeds in the open-pit mine are highest in the spring and lowest in the autumn. The daily wind speed in different seasons slightly decreases from 0:00 to 12:00 with small change in wind speed. The wind speed gradually increases after 12:00, reaching a peak at approximately 20:00 and then gradually decreasing. High wind speeds are observed in the open pit from 12:00 to 20:00. The results show that wind speeds are highest at approximately 20:00 in different seasons and have a positive effect to promote dust diffusion. As seen from Figure 9, the air pressure distribution in the open-pit mine is relatively uniform, and air pressure in different seasons is basically maintained at 86–88 kPa, which is relatively stable. Air pressure is mainly affected by altitude.

3.3. Correlation Analysis of Meteorological Factors

The above analysis results show that meteorological factors vary greatly in open-pit mines across the different seasons. To further study the influence of different factors on the variation in dust concentrations, Pearson’s correlation analysis is commonly used in general mathematical statistical analysis. The coefficient is also known as the Pearson product moment correlation coefficient and is used to describe and measure the correlation between two variables x and y. The value of the correlation coefficient r_pcc is between −1 and 1. A value close to −1 represents a stronger negative correlation between variables, and a value close to 0 represents basically no correlation between variables. Finally, a value close to 1 represents a stronger positive correlation between variables. The Pearson correlation coefficient is expressed by the quotient of covariance and the standard deviation between two variables. The calculation formula can be expressed as follows:

$$r_{pcc}=\frac{{\displaystyle \sum _{i=1}^n(x_i-\overline{x})(y_i-\overline{y})}}{\sqrt{{\displaystyle \sum _{i=1}^n{(x_i-\overline{x})}^2}}\sqrt{{\displaystyle \sum _{i=1}^n{(y_i-\overline{y})}^2}}}$$

(1)

where i represents the sample number of the x and y variables, and x and y represent, respectively, the sample mean of the x and y variables.

The analysis results for the correlation coefficient between the concentration of dust particles and meteorological factors are shown in

Table 1. Correlation coefficient between PMs and different meteorological factors.

PM	Meteorological Factors	Spring	Summer	Autumn	Winter
PM2.5	T	−0.8532	−0.1115	−0.8170	−0.7130
	H	0.6529	0.1896	0.5693	0.8427
	W	−0.6074	−0.4209	0.2515	−0.4674
	P	0.8643	0.4341	0.6769	0.8539
PM10	T	−0.8365	−0.1079	−0.8165	−0.7129
	H	0.6370	0.1853	0.5696	0.8421
	W	−0.6112	−0.4186	0.2505	−0.4675
	P	0.7942	0.4314	0.6762	0.7702
TSP	T	−0.8028	−0.1076	−0.8167	−0.7131
	H	0.6103	0.1853	0.5696	0.8423
	W	−0.5972	−0.4183	0.2510	−0.4674
	P	0.6783	0.4314	0.6764	0.6472

and Figure 10.

According to the results shown in Table 1, the influence of different meteorological factors varies in different seasons. It is clear that the smaller the particle size of the PM, the higher the correlation coefficient. The humidity and pressure are positively correlated with PM concentration, and the temperature and wind speed are negatively correlated with PM concentration. Results indicate that there is a higher correlation coefficient of humidity and has obvious effect on the PM concentration in winter. In addition, there is a lower correlation coefficient of temperature in summer.

According to the above results, the distribution rules of PM2.5, PM10 and TSP particles for different seasons are roughly the same. Taking TSP as the research object, the daily variation in TSP in different seasons is analyzed as shown in Figure 11. As seen from the figure, the concentration of TSP was high in March, November and December, and corresponding typical climate characteristics include high dust levels in March; low temperature and high humidity levels in November and December; high TSP concentrations in May (high wind speeds), June and July (high temperature). Field monitoring results show that dust particles in open-pit coal mines are affected by temperature, humidity, time and other factors simultaneously, due to multiple factors. The results can be clearly and concretely used to analyze the change rule of dust concentrations in open-pit mines and, on this basis, to reduce dust in open-pit mines scientifically, reasonably and economically.

4. Discussion

4.1. Dataset and Evaluation Index

In this study, average daily hour data of each month were collected, and the time interval of each point is 1 h, accounting for 288 h in a year. Some data for October were missing, and a total of 265 h of valid data were collected. The statistics of the monitoring data are shown in

Table 2. Statistics description of monitoring results.

Statistics	TSP (μg/m3)	PM10 (μg/m3)	PM2.5 (μg/m3)	T (°C)	RH (%)	V (m/s)	P (KPa)
Number	265	265	265	265	265	265	265
Mean	328.67	118.39	94.27	8.32	53.44	0.93	87.34
Minimum	201.58	45.25	36	−14.52	23.73	0.00	86.65
Maximum	641.20	293.84	228.67	29.37	97.90	2.59	88.19

.

The LSTM algorithm is used to train the TSP, PM10 and PM2.5 of monitoring stations to achieve the best simulation effect. Generally, the first seventy percent of the dataset will be utilized to train the prediction model and the rest thirty percent will be applied to be the validation dataset of the prediction model.

Four widely used indices are proposed to fully evaluate and compare the prediction result in this paper. The evaluation indices of the model include the root mean square error (RMSE), mean absolute error (MAE), mean absolute percentage error (MAPE) and determination coefficient (R²). The calculation formula of each evaluation index can be expressed as follows:

$$RMSE=\sqrt{\frac{1}{n}{\displaystyle \sum _{i=1}^n{\left(y_p-y\right)}^2}}$$

(2)

$$MAE=\frac{1}{n}{\displaystyle \sum _{i=1}^n}|y_p-y|$$

(3)

$$MAPE=\frac{100\%}{n}{\displaystyle \sum _{i=1}^n}|\frac{y_p-y}{y}|$$

(4)

$$R^2=1-\frac{{\displaystyle \sum _{i=1}^n{\left({y-y}_p\right)}^2}}{{\displaystyle \sum _{i=1}^n{\left(y-\bar{y}\right)}^2}}$$

(5)

where n is the amount of data; y_p and y represent, respectively, the predicted and actual value of dust concentrations (μg/m³); $\bar{y}$ represents the average of the actual value y (μg/m³); and the MAE and RMSE are the evaluation indices of the absolute error where smaller values are better. The prediction model has better prediction performance when it has a smaller RMSE and larger R².

The prediction results took into account the impact of surrounding environmental conditions and pollution sources in open-pit mines, the analysis of the prediction results of the model is shown in

Table 3. Evaluation results of forecasting models.

Model	Input Data	Indicator	RMSE (μg/m3)	MAE (μg/m3)	MAPE (%)	R2
LSTM	RH, P, T and V	PM2.5	11.09	9.37	9.74	0.88
		PM10	14.98	12.97	10.79	0.87
		TSP	26.16	21.61	6.47	0.87

.

It can be seen from Table 3, the RMSE of the LSTM model for the three different indicators are 11.09, 14.98 and 26.16; the MAE are 9.37, 12.97 and 21.61; and the MAPE are 9.74%, 10.79% and 6.47%. The results show that the LSTM model has a relatively good prediction effect on the test data; R² values reached 0.88, 0.87 and 0.87; the prediction result is close to the real value and good prediction result is obtained.

In addition, it can be seen from Figure 12 that the predicted and actual monitoring values of TSP, PM10 and PM2.5 particulate matter concentrations predicted by the LSTM model roughly follow a positive proportional distribution pattern, and particulate matter concentrations are distributed in a concentrated and uniform distribution in the low concentration range. The actual monitoring values of the model are consistent with the predicted values, which are consistent with the observed values, indicating good prediction accuracy. Meanwhile, the errors of the prediction results are within 25 μg/m³, 35 μg/m³ and 35 μg/m³. The LSTM model achieved better results and had accurate prediction results.

4.2. Model Comparison

In order to verify the effective prediction performance of LSTM model, four machine learning models were established and compared by MATLAB software. These alternative models include random forest regression (RFR), supported vector machine (SVR) and bounded batch normalization (BBN). The comparison models used the same training and test data as those of the LSTM model. TSP monitoring data were used as analysis variables to calculate the evaluation indicators of the predicted values of the above four models, and the results are shown in Figure 13 and

Table 4. Evaluation result of prediction results of different models.

Indicator	True Value	Model
		RFR	SVR	BBN	LSTM
Mean	353.99	333.17	316.89	326.99	344.76
25% quantile (Q1)	308.67	293.08	278.87	285.28	288.84
50% quantile (Q2)	337.71	303.86	296.26	308.01	323.55
75% quantile (Q3)	384.22	350.83	349.45	364.96	396.66
SD	72.80	63.36	57.63	71.43	81.64
RMSE		34.748	45.63	28.55	26.16
MAE		26.53	41.85	28.27	21.61
MAPE/%		7.88	12.94	8.85	6.47
R2		0.77	0.59	0.84	0.87

. The indices of RMSE, MAE and MAPE of the LSTM model were 26.16, 21.61 and 6.47%, respectively, representing decreases of 24.7%, 18.5% and 17.9% relative to the RFR model values. Compared to the SVR model values, the indices of RMSE, MAE and MAPE were 42.7%, 48.4% and 50% lower, respectively. Compared to the BNN model values, the indices of RMSE, MAE and MAPE of the LSTM model were 8.4%, 23.6% and 26.9% lower, respectively. At the same time, compared with other algorithms, the prediction result of LSTM model is closer to the real value at the extreme value, which indicates that the prediction effect of LSTM model is better overall.

As shown in Figure 14, the R² of the predicted and observed values of the RFR, SVR, BNN and LSTM are 0.77, 0.59, 0.84 and 0.87, respectively, and the MAE values are 26.53, 41.85, 28.27 and 21.61. The LSTM model achieved better results than other predictive models and has quite ideal prediction accuracy.

The LSTM prediction model proposed in this paper shows a good effect on the prediction of dust concentrations in open-pit mines, but some deficiencies and uncertainties remain. First, dust from open-pit mines is composed of mixed particulate matter with a complex composition, so it is necessary to analyze not only the particle size but also the composition of particulate matter. Adding composition data of particulate matter to the dataset built by the model may improve prediction accuracy. Second, the studied open-pit mine forms a semi-open space, and the concentration of dust particles is affected by surrounding pollution sources, vegetation coverage and terrain conditions. In future research, more influencing factors should be considered in the prediction model to enhance the generalizability of the model. Finally, as open-pit dust is affected by seasonality, it is necessary to establish different prediction models by season to further improve the prediction accuracy of the models.

5. Conclusions

(1)

Dust concentrations in the open pit are higher in spring and winter, while the overall concentrations are lower in summer and autumn. The daily variations of humidity and temperature follow the “herringbone” and “inverted herringbone” trends in the four seasons, respectively. The wind speed is highest in spring, and the air pressure is more evenly distributed in all seasons, basically remaining at 86–88 kPa;

(2)

Temperature and wind speed are negatively correlated with dust concentration, while humidity and air pressure are positively correlated. The higher the humidity and the lower the temperature, the higher the dust concentration;

(3)

The LSTM model for dust concentration in open pit coal mines is proposed in this paper. Comparison with other models shows that the MAE, MAPE and RMSE values of the LSTM prediction model are smaller using the model to predict PM2.5, PM10 and TSP concentrations with R² values of 0.88, 0.87 and 0.87, respectively, indicating that the model has a good prediction effect.