基于L-BFGS算法的BP神经网络在温度补偿中的应用

Application of L-BFGS Algorithm-based BP Neural Network in Temperature Compensation

为了解决氨气传感器温漂问题,设计了一种基于微处理器MK60DN512VLQ10电化学氨气变送器,采用BP神经网络算法进行软件补偿。针对BP网络收敛速度慢、精度不高、易产生局部最小值的缺陷,在BP网络算法迭代接近最优时引入L-BFGS算法加快收敛。同时,引入动态变量更新权值及权值导数,改进拟Hesse逆矩阵的求解,解决了溢出问题,实现了算法的全局收敛。实验证明:该氨气变送器有效克服了温漂问题,并具有响应时间短、测量精度高、稳定性好等优点。

In this paper,the electrochemical ammonia transmitter was designed based on microprocessor MK60DN512VLQ10 and BP neural network was selected to proceed software compensation. For BP network defects of slow convergence speed,low precision,and easy to produce a local minima,when the iteration of the BP network algorithm is close to the optimal,L- BFGS algorithm was introduced to speed up the convergence. Meanwhile,this paper solved the overflow problem by selecting suitable weight revision method to improve the solver of quasi Hesse inverse matrix,and achieved the global convergence algorithm. Experiments show that,the ammonia transmitter solves temperature drift problem,and has many advantages of short response time,high measuring precision,and good stability.