What effect will the choice of center point have on the performance of bp neural network?

Because of the randomness and uncertainty of prediction, traditional regression analysis, mathematical statistics and other methods are often difficult to achieve the ideal prediction effect. Back propagation network (BP) is one of the most widely used neural network models in artificial neural network (ANN), which has the characteristics of strong nonlinear mapping ability, robustness, fault tolerance, adaptability, self-organization and self-learning, and is widely used in hydrological forecasting.

Disadvantages of 1.2 BP neural network

However, in practical application, the selection of initial connection weights and thresholds of BP neural network has a key impact on the performance of BP neural network. If the initial connection weights and thresholds are not properly selected, it will easily lead to the inherent defects of BP neural network-slow convergence speed and easy to fall into local minima.

Optimization of 1.3 BP neural network

Genetic algorithm, particle swarm optimization and their improved algorithms are often used to optimize the initial connection weights and thresholds of BP neural networks. In addition, some bionic swarm intelligence algorithms are used to optimize the initial connection weights and thresholds of BP neural network, such as artificial fish swarm algorithm (AFSA), cuckoo search algorithm (CS), artificial bee colony algorithm (ABC), firefly swarm optimization algorithm (GSO) and differential evolution algorithm (DE), which have achieved certain results in improving the prediction or classification performance of BP neural network.

However, due to the performance of network prediction or classification, some achievements have been made. However, because the initial connection weight and threshold dimension of the BP neural network to be optimized are often high, it is difficult for traditional intelligent algorithms such as GA to obtain ideal optimization results. Wolves Algorithm (WPA) is a new bionic swarm intelligence algorithm, which simulates the cooperative hunting behavior and prey distribution of wolves. The algorithm has good robustness and global search ability. Compared with particle swarm optimization, AFSA algorithm and genetic algorithm, wavelet packet analysis algorithm shows great performance advantages, especially for complex functions with high dimensions and multiple peaks.