Method for predicting time required for personnel to return to working face after blasting in underground non-coal mine

Abstract

This invention discloses a method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines. It relates to the field of mine safety prediction technology. First, a negative feedback neural network model is established to correlate the personnel return time with four relevant indicators: roadway, ventilation duct, explosives, and gas. An initial model of the negative feedback neural network is configured. Sample data is collected from multiple channels. The weights and biases of the negative feedback neural network are globally optimized to obtain a prediction model for the personnel return time after blasting mining within the allowable error range. This prediction model is then trained and tested. Decision-makers use the prediction model to predict the return time and guide relevant personnel to return to the working face to continue operations. This invention achieves intelligent and accurate prediction of the personnel return time after blasting mining, effectively improving the accuracy and speed of the prediction results, making the predictions more scientific and effective, and enhancing their reliability.

Description

Technical Field

[0001] This invention relates to the field of mine safety prediction technology, and in particular to a method for predicting the time required for personnel to return to the working face after blasting in underground non-coal mines. Background Technology

[0002] In underground non-coal mine tunneling and mining, the drill-and-blast method is widely used due to its advantages of simple operation and low cost. However, the explosives decompose after detonation, producing large amounts of toxic gases. Inhaling these gases can cause symptoms such as vomiting and fainting, or even death, making it a dangerous source of poisoning and asphyxiation accidents. Statistics show that poisoning and asphyxiation accidents caused by fires and blasting fumes account for over 40% of all underground mine accidents, ranking first in both the number of accidents and fatalities. The frequent occurrence of poisoning and asphyxiation accidents in underground non-coal mines not only seriously affects the normal production of mining enterprises but also poses a severe threat to the lives and health of workers, resulting in extremely negative social impacts.

[0003] To prevent poisoning and asphyxiation accidents, various methods have been established at the national and industry levels for determining the return-to-workface time of personnel after blasting. Currently, the most commonly used method combines worker experience with the results of inspections by testing personnel. This method requires testing personnel to return to the workface three hours after blasting to conduct inspections. After inspection, reporting and notification are also required, and the entire process often takes up an entire work shift or even longer. In long, deep tunnels, this method severely impacts production efficiency. To improve the scientific rigor and accuracy of decision-making, some mines use empirical formulas to predict the return-to-workface time of personnel after blasting. However, this method can only consider a limited number of influencing factors, and the calculation results will deviate significantly if too many indicators are considered. Furthermore, empirical formulas are highly specific, and the calculation results are only applicable to a particular tunnel, lacking universality. With the advent of "smart mines," some mines have begun using gas sensors to monitor the concentration of toxic gases in real time to determine the return-to-workface time of workers. However, because some components of the sensor have high requirements for environmental conditions, the sensitivity, resolution and accuracy of the sensor will be greatly reduced under the corrosive effects of toxic gases and humid environments. In the event of a disaster emergency, it may even lose its monitoring capability. Therefore, staff have to frequently calibrate and replace the expensive sensors. Considering the large workload and high cost, many mines generally do not use this method.

[0004] In view of this, the present invention proposes a method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention discloses a method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines. This method establishes a negative feedback neural network topology with factors such as roadway length, wall roughness, effective range, air supply, ventilation duct location, gas production per kilogram of explosive, explosive usage, diffusion coefficient, and occupational exposure concentration limits as input layers, and personnel return time as the output layer. By integrating an atomic search optimization algorithm with the negative feedback mechanism of the negative feedback neural network, the weights and biases of the negative feedback neural network model are globally optimized, establishing a deep intrinsic connection between the input and output layer indicators. This enables intelligent and accurate prediction of the time required for personnel to return to the working face after blasting mining, improving upon the inaccuracies, deficiencies, and untimely nature of existing prediction methods.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines includes the following steps:

[0008] Step 1: Establish a negative feedback neural network relationship model between personnel return time to the working face and four related indicators: roadway, ventilation duct, explosives, and gas, and configure the initial negative feedback neural network model.

[0009] Step 2: Collect sample data through multiple channels, and use 80% of the sample data for training the initial model of the negative feedback neural network and 20% of the sample data for testing the initial model of the negative feedback neural network.

[0010] Step 3: Based on the atomic search optimization algorithm, the weights and biases of the negative feedback neural network are globally optimized to obtain the optimal weights and biases within the allowable error range.

[0011] Step 4: Based on the obtained optimal weights and biases, train and test the prediction model for the time of personnel returning to the working face after blasting, and obtain the prediction model for the time of personnel returning to the working face after blasting.

[0012] Step 5: Decision-makers, based on the actual situation on site, provide values ​​for each influencing indicator, use a predictive model for the time it takes for personnel to return to the working face after blasting to predict the results, and guide relevant personnel to return to the working face to continue operations.

[0013] Furthermore, in step one, the output of the prediction model for the time when personnel return to the working face after blasting is the time when personnel return to the working face. The input parameters are four aspects: roadway, ventilation duct, explosives, and gas. The roadway aspect includes roadway length and wall roughness. The ventilation duct aspect includes effective range, air supply volume, and ventilation duct location. The explosives aspect includes gas production per kilogram of explosives and explosive usage. The gas aspect includes diffusion coefficient and occupational exposure concentration limits stipulated by relevant laws and regulations.

[0014] Furthermore, in step three, the optimization methods specifically include:

[0015] (3.1) To avoid the influence of the order of magnitude of the sample data, all parameter data in the training samples are normalized. The normalization process can be expressed as:

[0016] y = 2*(xx) min ) / (x max -x min )-1 (1)

[0017] In the formula, x is the original value of the sample parameter, x max With x min denoted as the maximum and minimum values ​​in the original data, and y represents the normalized result;

[0018] (3.2) Initialize weights and biases using random seeds;

[0019] (3.3) Initialize the parameters of the atomic search optimization algorithm;

[0020] (3.4) The weights and biases in (3.2) are used as the initial solution vectors of the atomic search optimization algorithm, and the optimal solutions of the weights and biases are continuously updated based on the fitness function of the atomic search optimization algorithm.

[0021] Furthermore, in step (3.3), the parameters in the process of initializing the atomic search optimization algorithm include: initial population size, maximum number of iterations, number of independent variables, upper and lower limits of independent variables, depth weight and multiplier weight.

[0022] Furthermore, in step (3.4), the specific process of updating the optimal solution of weights and biases is as follows:

[0023] (3.4.1) Update the acceleration of each atom according to Newton's second law, that is:

[0024]

[0025] In the formula, Let be the total force of the i-th atom in d-dimensional space at the t-th iteration. Let represent the geometric constraints imposed on the i-th atom in d-dimensional space by the globally optimal atom at the t-th iteration. Let be the weight of the i-th atom in d-dimensional space at the t-th iteration;

[0026] (3.4.2) Solve for the total force using the potential energy relationship, i.e.:

[0027]

[0028] In the formula, α is the depth weighting coefficient, and hij (t) represents the distance between the two atoms, rand j It is a random number;

[0029] (3.4.3) Solve for the geometric constraints between the optimal atom and the atom to be solved, i.e.:

[0030]

[0031] In the formula, β is the multiplier weighting coefficient. The optimal population at the t-th iteration in d-dimensional space;

[0032] (3.4.4) Solve for the mass of each atom using the fitness function, i.e.:

[0033]

[0034] In the formula, Fiti(t) is the fitness function;

[0035] (3.4.5) After the acceleration calculation is completed, the velocity of the atom is updated using the following expression:

[0036]

[0037] In the formula, Let be the velocity of the i-th atom in d-dimensional space at the t-th iteration. Let be the velocity of the i-th atom in d-dimensional space at the (t+1)-th iteration;

[0038] (3.4.6) The atomic position is updated based on the atomic velocity, and its expression is:

[0039]

[0040] In the formula, Let be the position of the i-th atom in d-dimensional space at the t-th iteration. Let be the position of the i-th atom in d-dimensional space at the (t+1)-th iteration;

[0041] (3.4.7) Repeat the above steps until the fitness function meets the expectation, then terminate the solution process and derive the optimal weights and biases.

[0042] Furthermore, in step four, the prediction model for the return time of personnel to the working face after blasting is trained and tested. The calculation process between the input and output of the three-layer negative feedback neural network involved can be represented as follows:

[0043]

[0044] in, It is the predicted output, f输出 with f 隐含 Let be the activation functions of the output layer and the hidden layer, respectively; let I and J be the number of neurons in the input layer and the hidden layer, respectively; and let w be the activation function of the input layer and the hidden layer, respectively. ij It is the weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer, w jk It is the weight between the j-th neuron in the hidden layer and the k-th neuron in the output layer, b j With b k These are the biases of the j-th neuron in the hidden layer and the k-th neuron in the output layer, respectively.

[0045] The beneficial effects of this invention are:

[0046] (1) This invention integrates nine indicators from four aspects: roadway, ventilation duct, explosives, and gas. Compared with empirical formula prediction, it considers more comprehensive factors and has greater flexibility in controlling the number of indicators, which can effectively improve the accuracy and speed of prediction results.

[0047] (2) This invention uses a negative feedback neural network and an atom search optimization algorithm to predict the time for personnel to return to the working face after blasting. Compared with personnel experience, the prediction results are more scientific and effective, and can significantly improve the reliability of the prediction results.

[0048] (3) Based on historical sample data, this invention trains and optimizes the negative feedback neural network, which can realize the purpose of multiple multi-condition use of the prediction model after a single training, solve the problem that decision-makers have no emergency reference when sensor monitoring data is distorted, and enhance safety assurance capabilities while reducing production costs.

[0049] (4) This invention organically combines historical mine data with deep learning algorithms. The trained prediction model can also be applied to different roadway scenarios in multiple mines, thus improving the universality of the prediction model. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the main process of the present invention;

[0051] Figure 2 This is a schematic diagram of the negative feedback neural network topology in an embodiment of the present invention;

[0052] Figure 3 A schematic diagram of the optimization algorithm flow in this embodiment of the invention;

[0053] Figure 4 A schematic diagram of the prediction results in an embodiment of the present invention. Detailed Implementation

[0054] 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.

[0055] This invention discloses a method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines. The main process is as follows: Figure 1 As shown, this method establishes a negative feedback neural network topology that correlates personnel return time to the working face with four relevant indicators: roadway, ventilation duct, explosives, and gas. The collected sample data is then used to train and test the initial model of the negative feedback neural network, ultimately constructing a predictive model for personnel return time after blasting.

[0056] During the model training phase, an atomic search optimization algorithm is employed to prevent model weights and biases from getting trapped in local optima. Using a limited error as a benchmark, the deep intrinsic relationship between input indicators and output results in the negative feedback neural network model is determined. In the testing and usage phase, decision-makers can provide measured values ​​for each indicator based on actual conditions and use the prediction model for the return time of personnel to the working face after blasting to guide relevant personnel to return to the working face and continue operations.

[0057] A method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines includes the following steps:

[0058] (1) Establish a negative feedback neural network relationship model between personnel return time to the working face and four related indicators of roadway, ventilation duct, explosives and gas, and configure the initial model of negative feedback neural network.

[0059] The output of the prediction model for the time when personnel return to the working face after blasting is the time when personnel return to the working face. The input parameters are four aspects: roadway, ventilation duct, explosives, and gas. The roadway aspect includes roadway length and wall roughness. The ventilation duct aspect includes effective range, air supply volume, and ventilation duct location. The explosives aspect includes gas production per kilogram of explosives and explosive usage. The gas aspect includes diffusion coefficient and occupational exposure concentration limits stipulated by relevant laws and regulations.

[0060] The negative feedback neural network topology is a three-layer network structure containing only one hidden layer, such as... Figure 2 As shown, it is recommended that the number of hidden layer neurons be 4-6, and in this embodiment, the value is 4.

[0061] The initial parameters of a negative feedback neural network model are crucial for subsequent training and prediction. These initial parameters include the number of training iterations, learning rate, minimum error, momentum factor, and minimum performance gradient. In this embodiment, the number of training iterations is 1000, the learning rate is 0.01, the minimum error is 0.00001, the momentum factor is 0.01, and the minimum performance gradient is 0.000001. Besides the initial parameters, the selection of the activation function also needs to be considered. In this embodiment, the tansig function is used for the hidden layer, and the purelin function is used for the output layer.

[0062] (2) Collect sample data through multiple channels, and use 80% of the sample data for training the initial model of the negative feedback neural network and 20% of the sample data for testing the initial model of the negative feedback neural network.

[0063] In the initial stage of model use, it is necessary to obtain sample data on tunnel length, wall roughness, effective range, air supply, ventilation duct location, gas production per kilogram of explosive, explosive usage, diffusion coefficient, occupational exposure concentration limit, and return-to-work face time through various methods such as engineering practice, scientific experiments, and numerical simulation.

[0064] 80% of the data was used for model training, and 20% was used for model testing. In this embodiment, a total of 50 sets of sample data were collected, of which 40 sets were used for model training and 10 sets were used to evaluate the model's predictive performance.

[0065] (3) The weights and biases of the negative feedback neural network are globally optimized based on the atomic search optimization algorithm to obtain the optimal weights and biases within the allowable error.

[0066] The optimization method specifically includes the following processes:

[0067] (3.1) To avoid the influence of the order of magnitude of the sample data, all parameter data in the training samples are normalized. The normalization process can be expressed as:

[0068] y = 2*(xx) min ) / (x max -x min )-1 (1)

[0069] In the formula, x is the original value of the sample parameter, x max With x min denoted as the maximum and minimum values ​​in the original data, and y represents the normalized result;

[0070] (3.2) Initialize weights and biases using random seeds;

[0071] (3.3) Initialize the parameters of the atomic search optimization algorithm.

[0072] The parameters used in the initialization of the atomic search optimization algorithm include: initial population size, maximum number of iterations, number of independent variables, upper and lower limits of independent variables, depth weights and multiplier weights, etc.

[0073] In this embodiment, the population size is 10, the maximum number of iterations is 100, the number of independent variables is 45, and the upper and lower limits are 3 and -3, respectively.

[0074] (3.4) The weights and biases in (3.2) are used as the initial solution vectors of the atomic search optimization algorithm, and the optimal solutions of the weights and biases are continuously updated based on the fitness function of the atomic search optimization algorithm.

[0075] The specific process is as follows:

[0076] (3.4.1) Update the acceleration of each atom according to Newton's second law, that is:

[0077]

[0078] In the formula, Let be the total force of the i-th atom in d-dimensional space at the t-th iteration. Let represent the geometric constraints imposed on the i-th atom in d-dimensional space by the globally optimal atom at the t-th iteration. Let be the weight of the i-th atom in d-dimensional space at the t-th iteration.

[0079] (3.4.2) Solve for the total force using the potential energy relationship, i.e.:

[0080]

[0081] In the formula, α is the depth weighting coefficient, and h ij (t) represents the distance between the two atoms, rand j It is a random number.

[0082] (3.4.3) Solve for the geometric constraints between the optimal atom and the atom to be solved, i.e.:

[0083]

[0084] In the formula, β is the multiplier weighting coefficient. Let be the optimal population in the t-th iteration in d-dimensional space.

[0085] (3.4.4) Solve for the mass of each atom using the fitness function, i.e.:

[0086]

[0087] In the formula, Fiti(t) is the fitness function. When an atom has a better fitness, its mass is also greater.

[0088] (3.4.5) After the acceleration calculation is completed, the velocity of the atom is updated using the following expression:

[0089]

[0090] In the formula, Let be the velocity of the i-th atom in d-dimensional space at the t-th iteration. Let be the velocity of the i-th atom in d-dimensional space at the (t+1)-th iteration.

[0091] (3.4.6) The atomic position is updated based on the atomic velocity, and its expression is:

[0092]

[0093] In the formula, Let be the position of the i-th atom in d-dimensional space at the t-th iteration. Let be the position of the i-th atom in d-dimensional space at the (t+1)-th iteration.

[0094] (3.4.7) Repeat the above steps until the fitness function meets the expectation, then terminate the solution process and derive the optimal weights and biases. The specific flowchart is as follows: Figure 3 As shown.

[0095] (4) Based on the obtained optimal weights and biases, the prediction model for the time of personnel returning to the working face after blasting is trained and tested, and the prediction model for the time of personnel returning to the working face after blasting is obtained.

[0096] The final prediction result of this embodiment is as follows: Figure 4 As shown, the computational process between the input and output of the three-layer negative feedback neural network can be represented as follows:

[0097]

[0098] in, It is the predicted output, f 输出 with f 隐含 Let be the activation functions of the output layer and the hidden layer, respectively; let I and J be the number of neurons in the input layer and the hidden layer, respectively; and let w be the activation function of the input layer and the hidden layer, respectively. ij It is the weight between the i-th neuron in the input layer and the j-th neuron in the hidden layer, w jk It is the weight between the j-th neuron in the hidden layer and the k-th neuron in the output layer, b j With b k These are the biases of the j-th neuron in the hidden layer and the k-th neuron in the output layer, respectively.

[0099] (5) Decision-makers give values ​​for each influencing indicator based on the actual situation on site, use the prediction model of the time for personnel to return to the working face after blasting to predict the results and guide relevant personnel to return to the working face to continue the work.

[0100] To address the issues of low universality and narrow applicability of empirical formulas, this invention comprehensively considers multiple influencing indicators and, based on a negative feedback neural network deep learning algorithm, automatically completes the self-learning task of input and output parameters. Therefore, the prediction model for the return time of personnel after blasting mining established by this invention can be applied to various working conditions in multiple mines, and the number of indicators can be adjusted according to actual on-site conditions, greatly improving the versatility of the prediction model for the return time of personnel after blasting mining.

[0101] Given the problems of distorted feedback data from gas sensors, the heavy workload of laying and maintaining wired sensors, and the need for regular battery replacements and calibration of wireless sensors, this invention utilizes a negative feedback neural network model to perform deep learning on the nonlinear relationships between various influencing indicators and output results in historical sample data, eliminating the need to consider the impact of the humid and high-temperature underground environment. Therefore, mines equipped with this predictive model can reduce the number of gas sensors deployed and the workload of calibration, achieving dual safety assurance of "monitoring + prediction" while lowering production costs.

[0102] To address the current reliance on experience by workers to determine the return-to-workface time after blasting, and to resolve issues such as incomplete consideration of influencing factors and low prediction accuracy in related empirical formulas, this invention establishes a negative feedback neural network model that integrates multi-dimensional indicators with the return-to-workface time after blasting. The model is trained using sample data collected from multiple channels and optimized using an atomic search optimization method. Therefore, this invention considers more comprehensive influencing factors, significantly improving prediction accuracy, speed, and reliability, thus providing technical support for scientific and precise decision-making.

[0103] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines, characterized in that, Specifically, the following steps are included: Step 1: Establish a negative feedback neural network relationship model between personnel return time to the working face and four related indicators: roadway, ventilation duct, explosives, and gas, and configure the initial negative feedback neural network model. Step 2: Collect sample data through multiple channels, and use 80% of the sample data for training the initial model of the negative feedback neural network and 20% of the sample data for testing the initial model of the negative feedback neural network. Step 3: Based on the atomic search optimization algorithm, the weights and biases of the negative feedback neural network are globally optimized to obtain the optimal weights and biases within the allowable error range. Step 4: Based on the obtained optimal weights and biases, train and test the prediction model for the time of personnel returning to the working face after blasting, and obtain the prediction model for the time of personnel returning to the working face after blasting. Step 5: Decision-makers, based on the actual situation on site, provide values ​​for each influencing indicator, use a predictive model for the time it takes for personnel to return to the working face after blasting to predict the results, and guide relevant personnel to return to the working face to continue operations. In step one, the output of the prediction model for the time of personnel returning to the working face after blasting is the time of personnel returning to the working face. The input parameters are four aspects: roadway, ventilation duct, explosives, and gas. The roadway aspect includes roadway length and wall roughness. The ventilation duct aspect includes effective range, air supply and ventilation duct location. The explosives aspect includes gas production per kilogram of explosives and explosive usage. The gas aspect includes diffusion coefficient and occupational exposure concentration limits stipulated by relevant laws and regulations. In step three, the optimization methods specifically include: (3.1) To avoid the influence of the order of magnitude of the sample data, all parameter data in the training samples are normalized. The normalization process can be expressed as: ； In the formula, These are the original values ​​of the sample parameters. and denoted as the maximum and minimum values ​​in the original data, and y represents the normalized result; (3.2) Initialize weights and biases using random seeds; (3.3) Initialize the parameters of the atomic search optimization algorithm; (3.4) The weights and biases in (3.2) are used as the initial solution vector of the atomic search optimization algorithm, and the optimal solution of the weights and biases is continuously updated based on the fitness function of the atomic search optimization algorithm. In step (3.3), the parameters in the initialization process of the atomic search optimization algorithm include: initial population size, maximum number of iterations, number of independent variables, upper and lower limits of independent variables, depth weight and multiplier weight.

2. The method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines as described in claim 1, characterized in that, In step (3.4), the specific process of updating the optimal solution for weights and biases is as follows: (3.4.1) Update the acceleration of each atom according to Newton's second law, that is: ； In the formula, Let be the total force of the i-th atom in d-dimensional space at the t-th iteration. Let represent the geometric constraints imposed on the i-th atom in d-dimensional space by the globally optimal atom at the t-th iteration. Let be the weight of the i-th atom in d-dimensional space at the t-th iteration; (3.4.2) Solve for the total force based on the potential energy relationship, that is: ； In the formula, For depth-weighted coefficients, The distance between two atoms. It is a random number; (3.4.3) Solve for the geometric constraints between the optimal atom and the atom to be solved, i.e.: ； In the formula, Here, is the multiplier weighting coefficient, and e is the base of the natural logarithm. The optimal population at the t-th iteration in d-dimensional space; (3.4.4) Solve for the mass of each atom using the fitness function, i.e.: ； In the formula, , For the adaptive function; (3.4.5) After the acceleration calculation is completed, the velocity of the atom is updated, and its expression is: ； In the formula, Let be the velocity of the i-th atom in d-dimensional space at the t-th iteration. Let be the velocity of the i-th atom in d-dimensional space at the (t+1)-th iteration; (3.4.6) The atomic position is updated based on the atomic velocity, and its expression is: ； In the formula, Let be the position of the i-th atom in d-dimensional space at the t-th iteration. Let be the position of the i-th atom in d-dimensional space at the (t+1)-th iteration; (3.4.7) Repeat the above steps until the fitness function meets the expectation, then terminate the solution process and derive the optimal weights and biases.

3. The method for predicting the time required for personnel to return to the working face after blasting mining in underground non-coal mines as described in claim 1, characterized in that, In step four, the prediction model for the return time of personnel to the working face after blasting is trained and tested. The calculation process between the input and output of the three-layer negative feedback neural network involved can be represented as follows: ； in, It is the predicted output. and Let be the activation functions of the output layer and the hidden layer, respectively, and let I and J be the number of neurons in the input layer and the hidden layer, respectively. It represents the weights between the i-th neuron in the input layer and the j-th neuron in the hidden layer. It represents the weights between the j-th neuron in the hidden layer and the k-th neuron in the output layer. and These are the biases of the j-th neuron in the hidden layer and the k-th neuron in the output layer, respectively.

Images