基于遗传-蚁群算法优化BP神经网络的医用输液泵输液误差补偿

Infusion error compensation of medical infusion pumps based on BP neural network optimized by genetic-ant colony algorithm

一次性泵用输液器搭配医用输液泵进行输液时,输液精度受时间、温度和流速影响,可能会出现精度下降等不利现象。为了对输液误差进行补偿,提出一种遗传算法(genetic algorithm, GA)和蚁群算法(ant colony optimization, ACO)相结合以优化BP神经网络的方法。首先分析误差影响因素,通过多控制因素下的输液实验获得补偿前的误差数据。其次,运用GA-ACO优化BP神经网络的初始权值和阈值,建立误差补偿模型。最后将该算法和BP算法以及分别用遗传算法、蚁群算法优化BP神经网络的补偿效果进行对比。仿真结果表明,提出的算法能够有效减小输液误差。模型的平均绝对误差为0.575 7 mL、均方误差为0.533 9、均方根误差为0.730 7、平均绝对百分比误差为0.092 2%。公司产品样机实测结果表明,采用该优化算法能够提高输液精度,最大相对误差低于3.5%,验证了使用遗传-蚁群算法优化BP神经网络补偿泵用输液器输液误差的有效性,具有一定的实际应用价值。

When a disposable pump infusion set are used with a medical infusion pump for infusion,the accuracy of infusion is affected by time,temperature and flow rate,which may result in unfavorable phenomena such as decrease in accuracy.In order to compensate for infusion error,a method of optimizing back propagation(BP)neural network by combining genetic algorithm(GA)and ant colony optimization(ACO)is proposed.Firstly,the influencing factors of error are analyzed and the error data before compensation is obtained by infusion experiments under multiple conditions.Secondly,the initial weights and thresholds of the BP neural network are optimized by using genetic-ant colony algorithm(GA-ACO),and an error compensation model is established.Finally,the compensation effects of BP algorithm and optimizing BP neural network with genetic algorithm and ant colony algorithm respectively are compared with proposed method.Results show that the compensation effect of BP neural network optimized by genetic-ant colony algorithm is good and can effectively reduce the infusion error.The average absolute error of the model is O.5757 mL,the mean square error is 0.5339,the root mean square error is 0.7307,and the average absolute percentage error is 0.0922%.Actual measurement results from company made prototype show that the optimization algorithm can improve the infusion accuracy and the maximum relative error of infusion accuracy is less than 3.5%,which verifies the effectiveness of the compensation model of BP neural network optimized by genetic-ant colony algorithm and has certain practical application value.