ML Approach to Improve the Costs and Reliability of a Wireless Sensor Network

Temperature-controlled closed-loop systems are vital to the transportation of produce. By maintaining specific transportation temperatures and adjusting to environmental factors, these systems delay decomposition. Wireless sensor networks (WSN) can be used to monitor the temperature levels at different locations within these transportation containers and provide feedback to these systems. However, there are a range of unique challenges in WSN implementations, such as the cost of the hardware, implementation difficulties, and the general ruggedness of the environment. This paper presents the novel results of a real-life application, where a sensor network was implemented to monitor the environmental temperatures at different locations inside commercial temperature-controlled shipping containers. The possibility of predicting one or more locations inside the container in the absence or breakdown of a logger placed in that location is explored using combinatorial input–output settings. A total of 1016 machine learning (ML) models are exhaustively trained, tested, and validated in search of the best model and the best combinations to produce a higher prediction result. The statistical correlations between different loggers and logger combinations are studied to identify a systematic approach to finding the optimal setting and placement of loggers under a cost constraint. Our findings suggest that even under different and incrementally higher cost constraints, one can use empirical approaches such as neural networks to predict temperature variations in a location with an absent or failed logger, within a margin of error comparable to the manufacturer-specified sensor accuracy. In fact, the median test accuracy is 1.02 degrees Fahrenheit when using only a single sensor to predict the remaining locations under the assumptions of critical system failure, and drops to as little as 0.8 and 0.65 degrees Fahrenheit when using one or three more sensors in the prediction algorithm. We also demonstrate that, by using correlation coefficients and time series similarity measurements, one can identify the optimal input–output pairs for the prediction algorithm reliably under most instances. For example, discrete time warping can be used to select the best location to place the sensors with a 92% match between the lowest prediction error and the highest similarity sensor with the rest of the group. The findings of this research can be used for power management in sensor batteries, especially for long transportation routes, by alternating standby modes where the temperature data for the OFF sensors are predicted by the ON sensors.