Optimized Extreme Learning Machine by an Improved Harris Hawks Optimization Algorithm for Mine Fire Flame Recognition

In this paper, in order to solve the problems of low accuracy and slow speed of fire flame recognition, an extreme learning machine (ELM) method based on improved Harris hawks optimization (IHHO) is proposed for fire flame recognition. A novel Harris hawks optimization (HHO) is used to solve the problem of parameter selection of ELM. In order to solve the problem that the original HHO is prone to fall into local optimum, firstly circle mapping is used to initialize the population to solve the problem of uneven distribution and small range of the initial population. Then the formula of escaping energy of HHO is modified to make the population have a large range in the middle and late iterations while gradually decreasing in the whole iteration process. Thus, the exploitation stage in HHO is improved to make the search range smaller near the optimal solution to accelerate the convergence process. Finally, at the end of each iteration, a certain number of individuals are selected to perform Gaussian and Cauchy hybrid mutation to prevent the IHHO from falling into local optimization. Through three groups of experiments, the effectiveness of the proposed IHHO and IHHO-ELM is verified. In experiment 1, the convergence performance of IHHO is significantly better than that of the other remaining algorithms in the test results of six benchmark functions. In experiments 2 and 3 of real fire flame recognition case, IHHO-ELM outperforms other remaining algorithms on the whole and has significant advantages in some indexes.