Theoretical basis of BP neural network optimization algorithm based on genetic algorithm
Overview of BP neural network
It is a type of multi-layer feedforward neural network
80%~90% of neural network models use BP network or its deformation
Disadvantages: The learning convergence speed is too slow, convergence to the global minimum point cannot be guaranteed, and the network structure is not suitable to be determined.
The selection of network structure, initial connection weights and thresholds has a great impact on network training, but they cannot be obtained accurately. According to these characteristics, genetic algorithms can be used to optimize the neural network.
Basic elements of genetic algorithm
chromosomal coding method
Binary method: encoding the individual into a binary string
Real number method: encode the individual into a real number
Fitness function: A function written according to evolutionary goals to calculate individual fitness. The fitness value of each individual is calculated through the fitness function and provided to the selection operator for selection.
genetic manipulation
Select action
crossover operation
mutation operation
Operating parameters: parameters determined by the genetic algorithm during initialization