A new evolutionary system for evolving artificial neural networks
IEEE Transactions on Neural Networks
| Hi-index | 0.01 |
Generally, industrial measuring and control system works in surroundings with rapid temperature change, and most of thermal conductivity sensor is sensitive to temperature to a certain extent. This will lead to change of the sensor's sensitivity and zero performance, so the sensor's signal output varies with the change of temperature. Besides that, the measurement error is caused by pressure and flowing of gas. Therefore, error compensation is always the key problem of industrial measuring and control system. Using neural networks to deal with this error caused by environmental factors is a low-cost and reliable method. The traditional back-propagation algorithm is simple and easy to realize. However, it also has two insurmountable defects: the most fatal one is that back-propagation algorithm is very likely to run into local minimum. The other one is its slow convergence. Thermal conductivity gas analyzer often works in adverse and dangerous surroundings, which requires a short measurement period. So, the system needs a strong anti-shock learning network. Under this condition, fuzzy adaptive regulation is adopted in learning algorithm of back-propagation, and the learning parameters are corrected automatically by expert system. A fuzzy adaptive network model with strong contraction is constructed by training of large data samples, and rapid, high-precision compensation of sensor's error is realized by the improved algorithm.