Bp Neural Network Optimization And Its Application In The Measurement Of The Fluid

The liquid-droplet size distribution is one of the important characteristic parameters of gas-liquid two-phase flow. Measurement of the liquid-droplet size distribution can provide important data for researching the hydrodynamic and mass transfer characteristics of the gas-liquid contacting equipment, and provide basis for improving its engineering design and operation.In this dissertation, the future research of double electronic probes method using BP network is introduced. At the same time, the optimization of the structure and algorithm of BP network is researched deeply. Based on the experimental result, a series of method to optimize the network are summarized. The optimized network is also used to calculate the over heat steam flow, and its effect is obvious.Compared with former research, we arrive at these conclusions:(1) In network with two hidden layers, the training effect is better when the numbers of the neurons in both hidden layers are near.(2) In network with one hidden layer, the generalization of network is improved when the numbers of the neurons in the hidden layer are close to the numbers of the neurons in the input layer.(3) Using all samples to adjust weight can improve the convergence rate and the generalization.Furthermore, in this dissertation, the convergence process can be accelerated by adjusting the value of T in the activation function and the value of (threshold), and can be improved by using momentum. Strategy of limited training time and learning again prevent the network from over-training, and from being lost in local minimum.