传感技术学报
傳感技術學報
전감기술학보
Journal of Transduction Technology
2015年
1期
49-55
,共7页
磁致伸缩液位传感器%温度补偿%改进型ANFIS%BP算法%神经网络%PSO-LSSVM模型
磁緻伸縮液位傳感器%溫度補償%改進型ANFIS%BP算法%神經網絡%PSO-LSSVM模型
자치신축액위전감기%온도보상%개진형ANFIS%BP산법%신경망락%PSO-LSSVM모형
magnetostrictive liquid level sensor%temperature compensation%improved ANFIS%BP algorithm%Neural network%PSO-LSSVM model
考虑到磁致伸缩液位传感器在温差变化大的环境中温漂现象严重，且产生温漂的多种因素与温漂的程度呈非线性关系，难以用数学模型表达等问题，建立基于改进型ANFIS的温度补偿系统。该系统采用附加动量算法不断修正ANFIS中的前题参数以避免采用梯度下降算法时易陷入局部极小，训练速度较慢等缺点，提高系统的忽略网络中微小变化的能力。为了验证该温度补偿系统的性能，将其与基于PSO-LSSVM模型和基于BP神经网络的温度补偿系统相比较。分析与实验结果表明，改进型ANFIS模型的温度补偿的最大误差为0.88%，平均误差为0.65%，远小于另外两种补偿方法。使用了改进型ANFIS的温度补偿方法具有较强的泛化能力，能够有效消除温度对磁致伸缩液位传感器的影响。
攷慮到磁緻伸縮液位傳感器在溫差變化大的環境中溫漂現象嚴重，且產生溫漂的多種因素與溫漂的程度呈非線性關繫，難以用數學模型錶達等問題，建立基于改進型ANFIS的溫度補償繫統。該繫統採用附加動量算法不斷脩正ANFIS中的前題參數以避免採用梯度下降算法時易陷入跼部極小，訓練速度較慢等缺點，提高繫統的忽略網絡中微小變化的能力。為瞭驗證該溫度補償繫統的性能，將其與基于PSO-LSSVM模型和基于BP神經網絡的溫度補償繫統相比較。分析與實驗結果錶明，改進型ANFIS模型的溫度補償的最大誤差為0.88%，平均誤差為0.65%，遠小于另外兩種補償方法。使用瞭改進型ANFIS的溫度補償方法具有較彊的汎化能力，能夠有效消除溫度對磁緻伸縮液位傳感器的影響。
고필도자치신축액위전감기재온차변화대적배경중온표현상엄중，차산생온표적다충인소여온표적정도정비선성관계，난이용수학모형표체등문제，건립기우개진형ANFIS적온도보상계통。해계통채용부가동량산법불단수정ANFIS중적전제삼수이피면채용제도하강산법시역함입국부겁소，훈련속도교만등결점，제고계통적홀략망락중미소변화적능력。위료험증해온도보상계통적성능，장기여기우PSO-LSSVM모형화기우BP신경망락적온도보상계통상비교。분석여실험결과표명，개진형ANFIS모형적온도보상적최대오차위0.88%，평균오차위0.65%，원소우령외량충보상방법。사용료개진형ANFIS적온도보상방법구유교강적범화능력，능구유효소제온도대자치신축액위전감기적영향。
Taking into account the temperature drift of the magnetostrictive liquid level sensor is serious in the large temperature difference,and it is difficult to use the mathematical model to express the nonlinear relation between temperature drift phenomenon and the variety of factors,establish a temperature compensation system based on im-proved ANFIS. This system uses the Additional momentum method to constantly modify premise parameters in AN-FIS in order to avoid the shortcomings that it easy to fall into local minimum point and slow training speed when u-sing the gradient descent algorithm,and improve the capacity of ignoring tiny changes in the network. In order to verify the performance of the temperature compensation system,it has been compared with other temperature com-pensation system based on PSO-LSSVM and BP neural network. Analysis and experimental results show that the maximum error and its mean error of improved ANFIS model is 0.88% and 0.65%,far less than the other two kinds of compensation methods. This temperature compensation system based on improved ANFIS has strong generalization ability and can effectively eliminate the influence by temperature on the magnetostrictive liquid level sensor.
