TY - CHAP

T1 - Volume of Imbalance Container Prediction using Kalman Filter and Long Short-Term Memory

AU - Kim, Geunjeong

AU - Block, Brit Maren

AU - Mercorelli, Paolo

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

PY - 2024/2/28

Y1 - 2024/2/28

N2 - This paper develops a hybrid prediction model KALSTM that is a combination of a Kalman filter and a well-known neural network-based prediction model, Long Short-Term Memory (LSTM), to generate prediction results from the data, more specifically the weekly number of imbalance 20 general (GE) dry cargo container at Hamburg depots in Germany. The given data has been split into training and testing set. The training set has been used as input for the LSTM model to forecast the future weekly number of imbalance 20 GE containers. Even if the result from the LSTM model normally outperforms other existing forecasting models, it is still an approximation to the future because there are factors such as international political situations, overseas economic trends, and any kind of unexpected cases which increase the uncertainty of the prediction. Therefore, the output of LSTM has been used as container model xk, x^k of Kalman filter which filters the noise out with container measurement zk applying other forecasting models. Thus, a Kalman filter has been applied to improve the output of LSTM by correcting the noise of it. The prediction results of KALSTM demonstrate improved predictive performance based on common metrics such as mean squared error (MSE), root mean squared error (RMSE), and R2. Thus, the results can help decision-makers to improve tracking the volume of imbalance containers and to better prepare for further investigation to deal with the uncertainty of imbalance container volume.

AB - This paper develops a hybrid prediction model KALSTM that is a combination of a Kalman filter and a well-known neural network-based prediction model, Long Short-Term Memory (LSTM), to generate prediction results from the data, more specifically the weekly number of imbalance 20 general (GE) dry cargo container at Hamburg depots in Germany. The given data has been split into training and testing set. The training set has been used as input for the LSTM model to forecast the future weekly number of imbalance 20 GE containers. Even if the result from the LSTM model normally outperforms other existing forecasting models, it is still an approximation to the future because there are factors such as international political situations, overseas economic trends, and any kind of unexpected cases which increase the uncertainty of the prediction. Therefore, the output of LSTM has been used as container model xk, x^k of Kalman filter which filters the noise out with container measurement zk applying other forecasting models. Thus, a Kalman filter has been applied to improve the output of LSTM by correcting the noise of it. The prediction results of KALSTM demonstrate improved predictive performance based on common metrics such as mean squared error (MSE), root mean squared error (RMSE), and R2. Thus, the results can help decision-makers to improve tracking the volume of imbalance containers and to better prepare for further investigation to deal with the uncertainty of imbalance container volume.

KW - Hybrid prediction model

KW - Machine learning

KW - State-observation model

KW - Engineering

UR - http://www.scopus.com/inward/record.url?scp=85187643010&partnerID=8YFLogxK

U2 - 10.1007/978-981-99-8476-3_5

DO - 10.1007/978-981-99-8476-3_5

M3 - Article in conference proceedings

AN - SCOPUS:85187643010

SN - 978-981-99-8475-6

T3 - Lecture Notes in Networks and Systems

SP - 53

EP - 63

BT - Artificial Intelligence

A2 - Sharma, Harish

A2 - Chakravorty, Antorweep

A2 - Hussain, Shahid

A2 - Kumari, Rajani

PB - Springer Science and Business Media Deutschland GmbH

CY - Singapore

T2 - International Conference on Artificial Intelligence - AITA 2023

Y2 - 11 August 2023 through 12 August 2023

ER -