APSO-LSTM model based on improved Tent mapping for natural gas demand forecasting

Abstract

In order to accurately forecast the change of natural gas demand under the condition of multi-dimensional influencing factors, the adaptive inertia weight factor was introduced to improve the Particle Swarm Optimization (PSO) algorithm. Then, by combining the constructed Adaptive Particle Swarm Optimization (APSO) algorithm with the improved Tent mapping, hyper-parameter optimization was performed to the hidden layer nodes, the learning rounds and the initial learning rate in the Long Short-Time Memory Network (LSTM) model, so as to change the deficiency of the traditional LSTM model of setting the hyper-parameters empirically. Moreover, the calculation example was verified based on the subdivided data of 10 highly correlated influencing factors from 1999 to 2020, and the natural gas demand from 2021 to 2030 was forecasted. The results show that the improved Tent-APSO-LSTM model has the best optimization effect on the parameters and could be better applied to the gas demand forecasting in short and medium term.

Keywords

• natural gas demand
• tent chaotic mapping
• apso-lstm
• pso-lstm
• lstm
• bp neural network
• forecasting