| Gas-liquid two phase flow widely exists in human life and industrial production, such as in the chemical industry,petroleum,energy sources,power, pipeline transportation, refrigeration and other industrial applications. To study gas-liquid two phase flow,we need to identify the flow pattern firstly.Correct identification of flow pattern depends on the proper signal feature extraction method and the appropriate neural network identification model,the study of which is not only beneficial to improve the effect of flow pattern identification but also has important significance in industrial production.The flow pattern signal is obtained on the experimental platform of gas-liquid two phase flow,then extraction method of signal feature is used to the signal which has been done denoising process,finally different neural network identification models are adopted for flow pattern identification.Wavelet transform method is used to the obtained signal of flow pattern to do denoising process,the denoising effect of four different threshold rules are compared.After that,wavelet packet and empirical mode decomposition method are adopted to extract the wavelet packet energy feature and intrinsic mode function (IMF) energy characteristics which are later trained and identified in BP neural network and RBF neural network. The result shows that the maximum correct recognition rate is90.0%which appears when the RBF neural network identification model works with empirical mode decomposition method.For the same kind of neural network,recognition rate is higher based on empirical mode decomposition method.For RBF neural network, this paper also adopts genetic algorithm to optimize its structure parameters,it shows the recognition effect after optimization is better than that before optimization.The highest recognition accuracy is94.4%which appears when empirical mode decomposition method works with optimized RBF neural network identification model. |
PDF Full Text Request