64 / 2023-03-31 15:28:46

Retrieving Probability Integration Model parameters using improved Sparrow Search Algorithm

Probability Integration Model, mining subsidence predicting model, Sparrow Search Algorithm, swarm intelligent optimization algorithm

Abstract: Using swarm intelligent optimization algorithm to retrieve parameters of mining subsidence prediction model has obvious advantages over traditional methods. The Sparrow Search Algorithm (SSA) was proposed in 2020, with fast convergence speed, good robustness, and global optimization ability. It has been widely used in practical problems, such as optimal path planning, multi-objective optimization, water supply network hydraulic model verification, and job shop scheduling. This paper introduced the SSA into the inversion of Probability Integration Model (PIM) parameters.

Firstly, the shortcomings of the SSA were improved, mainly including: (1) The SSA used random numbers to generate the initial sparrow swarm location, sparrow locations could not be uniformly distributed throughout the search space and unpredictable, which would affect the search speed. For the problem, the Halton sequence was proposed to generate the initial location of the sparrow swarm, which would make the individual sparrow locations be uniformly distributed throughout the search space. (2) The location updating trend of the discoverer sparrows always tended to zero, for solving practical problems where the optimal solution was not at zero, it tended to fall into a local optimum. A cosine function location updating method was proposed to improve the global search ability. (3) All dimensions of a follower sparrow location updated with the same step size, which would result in some dimensions easily out of the bounds. And the dimension which was out of the bound would be assigned the bounder value, which would reduce search speed. For these problems, a new bound control method was proposed, the out of bound dimension be assigned a new random value in the half part of the bound in the crossing boundary direction. (4) In addition, we proposed two new optimization mechanisms: the swarm location updating direction self-evaluation and self-correction, and using historical optimal location to update the current optimal location dimension by dimension. Test functions for optimization experiments showed that the improved SSA achieved the same accuracy results with only half iterating times of the SSA.

Then, using [q,tanβ,θ0,s1,s2,s3,s4] as the location search space, and the RMSE of the fit value as the fitness function, an inversion model for PIM parameters was constructed.

Finally, taking the 85201 working face as an example, the improved SSA was applied to retrieve the parameters of PIM. At the same time, genetic algorithm, simulated annealing algorithm, and particle swarm optimization algorithm were also used for the same experiment. Experiments results showed that the parameters obtained by the improved SSA make the PIM curve fitted the field measurement subsidence curve well, and the RMSE was ± 52 mm and the relative RMSE was less than 2.0%. The RMSEs of the three other methods were more than ± 57mm, and the relative RMSEs were more than 2.2%. The accuracy of the improved SSA was better than the three other methods.