② Calculate the fitness value of each individual in the population. The fitness value is directly related to the distance of the optimal solution.

③ The population replicates according to the fitness value.

(4) If the termination conditions are met, stop; Otherwise, go to step (2).

As can be seen from the above steps, PSO and genetic algorithm have many similarities. Both of them randomly initialize the population, both of them use fitness value to evaluate the system, and both of them conduct random search according to fitness value. Neither system can guarantee to find the optimal solution. However, PSO does not have crossover, mutation and other genetic operations, but determines the search according to its own speed. Another important feature of particles is that they have memory. The advantage of evolutionary computing is that it can handle some things that traditional methods can't. Examples are nondifferentiable node transfer functions or no gradient information.

But the disadvantage is that:

1, which is not particularly good on some issues.

2. The coding of network weights and the selection of genetic operators are sometimes troublesome.

Recently, some papers have used particle swarm optimization instead of back propagation algorithm to train neural networks. The research shows that particle swarm optimization is a potential neural network algorithm. PSO is faster and can get better results. And there are no problems encountered by genetic algorithms.