Article Preview
TopHowever, there are deficiencies within the FWA approach. If the optimal value of the function is situated at the origin, the FWA converges to the origin very well. When the optimal value is not at the origin, it is difficult for the FWA to get the correct solution. In addition, the time cost in the selection phase of the FWA is long compared with other optimization algorithms. Therefore, the enhanced fireworks algorithm (EFWA) was introduced to enhance FWA (Zheng, 2013). The explosion size of the best firework tends to zero in EFWA, thus a new minimal amplitude check is employed. The magnitude is calculated based on the maximum number of evaluations, which does not take into account the adaptability and dynamics around the core firework. Thus, the dynFWA introduced a dynamic amplitude that will allow quick convergence or will decrease to narrow the local search (Zheng, 2014).
A new cooperative framework for fireworks algorithm (Zheng, 2015) uses the independent choice operator to improve the exploitation capacity of non-core fireworks and adopts a cooperative strategy of crowdedness-avoiding to increase the exploration capacity among fireworks. A new guiding spark in FWA (GFWA) was proposed to enhance the information utilization (Li, 2017). GFWA uses information of objective functions obtained from sparks to build a guiding vector, and to produce an elite solution. To further improve the performance of GFWA, two strategies are proposed, such as weight-based guiding strategy and quantitative increase strategy (Li, 2019).
The amplification coefficient 
and reduction coefficient 
in dynFWA are recommended empirically to be a value of 1.2 and 0.9, but our thorough experiments show the values of 
and 
should vary depending on the actual problems. With the goal of achieving global optimization and further boosting performance of dynFWA, adaptive parameters are added and dynFWAAP is proposed.