Adaptive-Sensorless Monitoring of Shipping Containers

TL;DR

This work tackles the variability and connectivity challenges in monitoring shipping containers by introducing adaptive-sensorless monitoring through the residual correction method (RCM). By training an unconditional land-model f on live-data and then applying shipment-specific residual corrections via a weighting function to predict delayed ocean data with a separate model h, the approach outperforms traditional sensorless predictions on a particularly large, geographically diverse dataset. Key findings show average MAE/RMSE improvements for temperature (≈2.241°C/3.188°C vs 2.426°C/3.376°C) and relative humidity (≈5.72%/7.70% vs 7.99%/10.0%), demonstrating the practical viability of adaptive-sensorless monitoring for early risk detection and reduced reliance on constant connectivity. The framework offers scalable, cost-effective cargo monitoring with broader applicability to real-world supply chains.

Abstract

Monitoring the internal temperature and humidity of shipping containers is essential to preventing quality degradation during cargo transportation. Sensorless monitoring -- machine learning models that predict the internal conditions of the containers using exogenous factors -- shows promise as an alternative to monitoring using sensors. However, it does not incorporate telemetry information and correct for systematic errors, causing the predictions to differ significantly from the live data and confusing the users. In this paper, we introduce the residual correction method, a general framework for correcting for systematic biases in sensorless models after observing live telemetry data. We call this class of models ``adaptive-sensorless'' monitoring. We train and evaluate adaptive-sensorless models on the 3.48 million data points -- the largest dataset of container sensor readings ever used in academic research -- and show that they produce consistent improvements over the baseline sensorless models. When evaluated on the holdout set of the simulated data, they achieve average mean absolute errors (MAEs) of 2.24 $\sim$ 2.31$^\circ$C (vs 2.43$^\circ$C by sensorless) for temperature and 5.72 $\sim$ 7.09% for relative humidity (vs 7.99% by sensorless) and average root mean-squared errors (RMSEs) of 3.19 $\sim$ 3.26$^\circ$C for temperature (vs 3.38$^\circ$C by sensorless) and 7.70 $\sim$ 9.12% for relative humidity (vs 10.0% by sensorless). Adaptive-sensorless models enable more accurate cargo monitoring, early risk detection, and less dependence on full connectivity in global shipping.

Figures (5)

• Figure 1: The concept of adaptive-sensorless monitoring for shipping containers. Adaptive-sensorless monitoring has a sensorless model for the ocean route to estimate the temperature/relative humidity inside a shipping container. The sensorless model for the ocean route uses residuals between sensor data and estimation results of the sensorless model for land route, in addition to weather data.
• Figure 2: Geographical coverage of our sensor dataset. Redline shows the path of each shipping route. Our dataset covers almost all common shipping routes in the world.
• Figure 3: An illustrated process to create environment tags from map data. First, we extract an area containing the given GPS coordinate (target location) and parse the geofeatures (e.g., Farmland, Pond) around the coordinate. We then compute each geofeature's relevance to the GPS coordinate using a combination of heuristic rules before using them to characterize the environment around the coordinate. In this diagram, we give the example of a simple case where the environment is defined to be the most relevant geofeature.
• Figure 4: An illustration of different residual forecasting strategies: local versus global.
• Figure A1: Histogram of temperature prediction error for the baseline model ($^\circ$C ) and RCM with local forecasting strategy and linear weights for the subset 202204.