Research On Time Series Prediction Method For LNG Storage Tank Safety Monitoring

Liquefied natural gas LNG,as a clean and green energy,occupies an important position in the domestic and international energy market.With the continuous expansion of trade scale and the continuous growth of transportation volume,it is of great significance to ensure the safety of LNG in the transportation process and storage process.Based on the analysis of the development status of ensuring the safety of LNG transportation at home and abroad and the research status of time series prediction methods,this paper designs and develops a system of LNG storage tank terminal and real-time monitoring of LNG storage tank in order to improve the safety of LNG storage tanks in LNG multimodal transportation.In order to improve the intelligence level of the system,On the one hand,an adaptive time series prediction method is proposed to perform single-step prediction for multiple sensor data to realize the sensor early warning function.On the other hand,the long short-term memory neural network LSTM and grey model GM algorithm are combined to realize the prediction of the safe pressure storage time of LNG storage tank in a multi-step prediction way.The main contents of this paper are as follows:(1)Focusing on the monitoring problem of LNG storage tank,the overall design scheme of LNG storage tank terminal and monitoring system is proposed,and a set of LNG storage tank terminal and monitoring system is developed,which realizes the data acquisition of sensors such as pressure,liquid level and temperature inside the LNG storage tank,and external temperature,humidity,GPS and methane concentration outside the LNG storage tank.Data transmission,data saving,data prediction,data display and other functions.(2)In view of the lack of real-time early warning in the LNG storage tank monitoring system,this paper proposes a single-step adaptive time series prediction algorithm TAEF combining data trend and prediction error,and applies it to the LNG storage tank monitoring system.By predicting real-time sensor data,it can determine in advance whether there is a possibility of exceeding the threshold,so as to realize sensor early warning.The experimental results show that under different thresholds,the average success rate of methane concentration early warning is 78%,the average success rate of voltage early warning is 100%,the average success rate of internal temperature early warning is 100%,and the average success rate of liquid level early warning is 86.2%,which can basically meet the use needs.(3)In order to realize the safety guarantee of LNG storage tank transportation and improve economic efficiency,the problem that LSTM fixed step output cannot directly predict safe storage time and the problem that GM has poor prediction effect on safe storage time is solved.In this paper,combining long short-term memory neural network LSTM and grey model GM prediction algorithm,the LSTM-GM algorithm is proposed to predict the safe storage time.Firstly,the LSTM fixed step prediction results were put into the GM algorithm as input,and then the number of prediction steps was increased until the pressure exceeded the limit,and then the safe pressure storage prediction time was calculated according to the number of prediction steps and the sampling frequency of sensor data.The LSTM-GM algorithm solved the problem that the LSTM direct multi-output method could not calculate the safe pressure storage time.At the same time,compared with GM direct prediction of safe storage time,this algorithm improves the prediction accuracy.