【摘要】 BP神经网络和遗传神经网络是混合气体识别中常用的方法,但在实际应用仍然存在一些缺陷与不足。针对存在的问题,提出了1种改进自适应遗传算法,该算法根据进化过程种群中未产生更优解的代数,自适应调整变异率和变异量。利用该改进自适应遗传算法优化BP神经网络的连接权和阈值,构成改进自适应遗传神经网络,并应用于混合气体的识别中。实验结果表明:改进自适应遗传神经网络收敛成功率由40%提高到80%,平均识别误差H₂S由4.66 mL/m³降为3.69 mL/m³,CH₄由17.14 mL/m³降为15.77 mL/m³,CO由4.38 mL/m³降为4.19 mL/m³。更多还原

【Abstract】 BP neural network and GA neural network were the common methods applied in multi-gas recognition.But the methods were found having some defects and shortages in practical use.According to the shortages,an improved adaptive genetic algorithm which can adaptively adjust mutation probabilities and volumes based on generations having no better solution in evolution is proposed.This algorithm is used to optimize connection weights and thresholds of BP neural network and applied in multi-gas recognition.The experimental results show that the convergence probability of this improved adaptive genetic neural network is increased from 40%to 80%,and the average recognition errors, H₂S is reduced from 4.66 mL/m³ to 3.69 mL/m³,CH₄ is reduced from 17.14 mL/m³ to 15.77 mL/m³,CO is reduced from 4.38 mL/m³ to 4.19 mL/m³.更多还原