Abstract: As the core component of the fire water jet system, the self-excited oscillating fire nozzle is suitable for eliminating most solid fires, and has the advantages of simple structure, high efficiency and low cost. The numerical simulation method can be used to calculate the flow field and obtain high precision results, but the calculation time is long and the calculation amount is large. The reduced order prediction model is an effective means to reduce the dimension of the flow field, realize flow field reconstruction and predict the flow field distribution. Therefore, taking the self-oscillating fire nozzles as the research object, a reduced order model based on encoder and proper orthogonal decomposition is proposed, and a long short-term memory networks and a deep neural network are used as reasoners to carry out the pulse mechanism analysis, feature extraction and reconstruction, timing prediction and pulse impact force analysis of the nozzle flow field. The prediction results show that the average relative error of the prediction of nozzle outflow field is 7.02%, that of internal flow field is 7.20%, and that of pulse jet pressure is 3.25%. The experimental results show that the minimum impulse force error between the prediction model and the test is 7.74% and the maximum impulse force error is 5.65%. The minimum pressure error of numerical simulation and test is 3.14%, and the maximum pressure error is 6.72%. The research of this project can provide a way to predict the nozzle flow field.

Key words: reduced order model, long short-term memory networks, self-excited oscillation, fire nozzle, flow field prediction