A SVM ( Support Vector Machine) model for predicting the limit parameters of coal self-ignition was developed. By comparing with polynomial function and Sigmoid function, radial basic function was selected as the kernel function of SVM. A method for optimizing SVM parameters was proposed. The procedure of this method includes three stages: ① Selecting an area large enough to cover the optimal parameters; ② Searching the large area with search step length changing along with the mean squared error of training set, to quickly locate a small area in which the optimal parameters lie; ③ Searching the small area grid by grid for the optimal parameter, which is obviously more efficient than searching the large area directly with grid search method. Experimental results show that, when used for predicting limit parameters of coal self-ignition, SVM-based model performed significantly better than the neural network-based model on both prediction-precision and modeling speed, especially under the condition of limited training samples. So, SVM is feasible for predicting the limit parameters of coal self-ignition.

Spontaneous Combustion in Coal Seam (SCCS) is seriously threatening coal mine safety. A novel approach to predict SCCS by using Support Vector Machine (SVM) is present. The SVM is based on statistical learing theory with a simple structure and good generation properties. The basic SVM principle was firstly reviewed. Then, the kernel function was choiced, and the model parameters were optimized with cross-validation and grid-research method. Finally,a comparision of the preformance of SVM with radial basis function neural networks (RBF-NN) was carried out.The experimental results show that the highest classification accuracy (100%) is obtained for the SVM model, and the SVM-based model makes predictions much more accurate than RBF-NN model does when the samples are limited. Consequently, a properly trained SVM classification model can be a strong predictor for SCCS prediction procedure.

An analytical solution of one-and three-dimensional diffusion models in an unbounded space about a time-continuous point source and the method of image interference were used to study three-dimensional diffusion in roadway air with a fully reflecting boundary wall.The shortest fully-mixed distance and its stabilizing time were determined for dispersion of the tracing gas in the roadway air.If the hydraulic radius and friction coefficient are kept constant,the shortest fully-mixed distance increases linearly,and the time to reach equilibrium decreases by a negative power,with an increase in airflow velocity.The shortest fully-mixed distance decreases by a negative power,and the time to reach equilibrium follows a power law function,with an increase in the friction factor when the hydraulic radius and airflow velocity are constant.

Four groups of published experimental data about length of countercurrent smoke layer were collected. By analyzing these data, three qualitative descriptions for countercurrent smoke layer length varying with the heat release rate and ventilation velocity were concluded. Different regression models consistent with these descriptions were tried in multiple nonlinear regression analyses. A novel empirical relationship for countercurrent smoke layer was obtained, which coincides completely with the three descriptions, and gives a better correlation of the four groups of data than known relationships do.

Solving d-MC problem is often a tedious process. Three ways are suggested to improve the efficiency of solving d-MC problem. The first way is to make the best use of some special properties of network. A property of Network with Joint Parallel Part, which is more common than series–parallel network, is illustrated. The second way is to reduce the number of d-MC candidates and then to reduce the cost of testing. Two theorems on how to find the d-MCs with only one element unsaturated and on how to set the Lower Capacity Limits (LCLs) of elements to some values higher than zero are proved. These two theorems will be helpful to reduce d-MC candidates without any loss of real d-MC. The third way is to efficiently remove the duplicated d-MCs. A theorem, elucidating which Minimal Cuts (MCs) the duplicated d-MCs will be generated from, is proved. Finally, an algorithm is proposed by adding a pre-numerating step to the algorithm presented in Yeh [A new approach to the d-MC problem. Reliab Eng Syst Safety 2002;77(2):201–6], and two examples are employed to illustrate the proposed algorithm, especially the pre-numerating step.

Application of 1D Transient Flow Model to a network analysis finds that the fluids in two branches (which are connected in parallel and contain no drive source) may no longer flow in a same direction while the network flow state changes from being steady into being unsteady due to a fire in the ventilation network. This phenomenon can be referred to as non-parallel flows in parallel branches. A non-dimensional criterion ξ is deduced theoretically. It is proved that for any two branches which are connected in parallel and contain no drive source, Non-parallel flows can be expected in the branches only if ξ is not equal to unity. If ξ is equal to unity, the non-dimensional flow velocities in the two branches will always be equal to each other. If ξ is not equal to unity, a prediction can be made on which one of the non-dimensional velocities in the two parallel branches will change more rapidly.

Based on theoretical analysis, a dimensionless criterion xi, for determining the sequence of flow reversals in two airways both connected in parallel with an ascensionally ventilated airway on fire is obtained. The further computer-simulation analyses show that this criterion is fairly rational. If xi is equal to unity, flow reversal will occur simultaneously in both of the two airway or will occur in neither of the two airway; If xi is not equal to unity, flow reversal (if any) will occur earlier in one of the two airway than in another one.