Abstract: Firstly, the approximation theorem and basic properties of one-dimensional Bernstein polynomial were introduced, and two-dimensional or even n-dimensional Bernstein polynomials were introduced, and then a three-layer forward neural network was constructed by using the difference values of Bernstein polynomials in the adjacent equidistant points as the back connection weights. Secondly, the operation process of the algorithm was given through coding mechanism, simulation selection, genetic duplication, crossover and mutation. Finally, the error function and fitness function were used to iterate and update the front connection weights and threshold to design a genetic algorithm for the network. The experimental result shows that the algorithm is effective.

Key words: Bernstein polynomial, Sigmodial transfer function, forward neural network, fitness function, genetic algorithm