An Interval Neural Network Based on Improved Differential Evolution Algorithm
Dapeng Niu, Zicheng Zhao, Mingxing Jia · 2021
To solve the problems of slow convergence speed, low precision and easily trapped into local extremum for traditional interval neural network, the evolutionary operation in JADE algorithm is extended to interval operation, and an improved interval differential evolution algorithm is proposed to express uncertain data in metallurgical process. This algorithm adopts interval midpoint radius interval to perform arithmetic operation instead of traditional interval arithmetic operation to ensure the tracking performance of population individuals. This interval neural network is used to perform nonlinear interval functions fitting. Simulation results indicate that compared with gradient descent algorithm and traditional interval differential evolution algorithm, the interval neural network trained by the improved interval differential evolution algorithm has better performance.