Optimization of Vacuum Pyrolysis Process Using Generalized Regression Neural Network

A new empirical technique to construct predictive models of vacuum pyrolysis process is presented in this study. Pyrolysis of biomass for preparing bio-oil was studied on a vacuum pyrolysis system, where rape straw was chosen as the raw material. The experiments ran based on orthogonal experimental design method. The operation factors of the system including pyrolysis temperature, system pressure, heating rate and holding time were chosen as input variables, while bio-oil yield and energy transformation ratio were selected as output to establish the prediction models based on Generalized Regression Neural Network (GRNN). The operation factors of the system were optimized for maximizing bio-oil yield and energy transformation ratio, and the optimization result was confirmed by experiments. The results of research showed that the predicted values are fit well with the experimental values, which verifies the effectiveness of the prediction models. Optimal conditions are obtained at pyrolysis temperature of 486.8°C, heating rate of 18.1°C/min, reactor pressure of 5.0kPa and holding time of 55.0min. Confirmation runs give 41.9%, 42.5% and 42.1% of bio-oil yield and 34.3%, 34.0% and 34.9% of energy transformation ratio compared to 43.6% and 35.5% of predicted value. Therefore, the forecasting model based on the GRNN is able to result in good prediction and has research value to the reality.