Abstract: Single neuron adaptive PID controller not only has simple structure, but also is able to adapt to environmental changes with stronger robustness. An improved genetic algorithm based on evolutionary mechanism is proposed in the paper and the detail process is given. In order to improve the convergence performance of the algorithm, the replacement operation is done to individuals which have lower affinities before cross and mutation operation. Function optimization experiments verify the effectiveness of the algorithm. Finally, the algorithm is applied to the synaptic weights of the neural network optimization. Through the adjustment of synaptic weights, the proportional, integral and differential coefficients control of the PID can be achieved adaptively. The simulation comparison results show the superiority of the algorithm.