Detection of long-term effect in forecasting municipal solid waste using a long short-term memory neural network

Abstract

Abstract Researchers have been successfully applied artificial neural networks (ANNs) in the time-series analysis and forecasting of municipal solid waste (MSW). Despite the reported high accuracy ANNs have achieved in many cases, they are limited by the requirement for consistent input data formats and the high correlation between input and output. This study adopted a deep learning approach—long short-term memory (LSTM)—to solve this issue. Aiming at the temporal variation of MSW generation, an LSTM neural network consisting of LSTM layers and a dropout layer was established and optimized for forecasting MSW generation. To better illustrate the accuracy and reliability of the LSTM neural network, MSW forecasting was also conducted using an autoregressive integrated moving average (ARIMA) model and conventional ANN. The accuracy of LSTM neural network, the ARIMA model and traditional ANN reaches 0.92/935.08/114.36, October 0, 2116.7/264.5 and 0.74/547.14/50.41 in R2, RMSE and MAPE, respectively, which proves LSTM neural network’s excellence in MSW forecasting. The comparisons of LSTM, ARIMA, and conventional ANN also implicated the existence of long-term effects in the temporal variations in MSW generation, which could only be involved in LSTM neural network. This study not only promotes ANN-based MSW prediction by removing its restriction in application, but also broadens the scope for future related research by taking an insight into the effect of temporal variation in MSW prediction. This study provides a novel approach for forecasting municipal solid waste, LSTM neural network, which could consider both static and dynamic characteristics in temporal variation of MSW. Besides, long-term effect in MSW forecasting was detected in this study, which would be of great significance in studies relating the regularity of MSW and MSW forecasting methods.