| In recent years, artificial intelligence, more and more attention by people, especially some intelligent algorithms has been widely used in engineering. However, the traditional intelligence algorithms are not suitable to solve some new problems, and we must adjust the traditional algorithm by the characteristics of the problem, so that it can achieve the satisfactory results.There are three parts in this paper.First, the major drawbacks of the BP algorithm are the problems of local minima and slow convergence. In order to overcome local minima and speed up the convergence of BP, a novel wavelet-based DPFNN (double parallel feed forward neural network) is proposed in this paper. We use the traditional BP, the DPFNN and the wavelet-based DPFNN to identification the mixed gas. Experimental results show that, compared to the traditional BP and DPFNN Neural Network, wavelet-based DPFNN neural network can speed up the convergence rate of 2～3 times, and the concentration estimation accuracy of gas mixture can be improved significantlySecondly, we improve the traditional genetic algorithm by the characteristics of the IP mapping problem in NoC, we change the method of generating offspring with the matrix transformation, while we use the layered initialization to get the individuals with higher fitness value, We do the software simulation with MATLAB, simulation results show that,compared with the best result from the 30 randomly generated, the result from the genetic algorithm has reduced the communication power consumption by 28%.Finally, we build a 4×4 NoC power model with SystemC which can do the right communicate between two IP cores, we put the best mapping result form 30 randomly generated and the mapping result by the improved genetic algorithm into the NoC power model, simulation results show that, compared with the best result from the 30 randomly generated results, the result from the genetic algorithm has reduced the communication power consumption by 23.6%. |
PDF Full Text Request