| dc.description.abstract | The need to constantly ameliorate Cold Chain Logistics (CCL) can no longer be ignored. This
industry has grown in size with significant positive impact on the GDP of economies globally.
Amidst great opportunities vested in this industry, it stills struggles with ills such as poorly
managed service level agreement, no or inadequate chain visibility and cargo losses. Such
losses are mainly caused by lack of real time information about the current status of cargo as
well as lacking insight into the possible impact of supply chain incidents on cargo quality.
This work presents an empirical temperature modelling for fresh produce logistics during transit
in southern Africa. It describes the characterization of cold chain processes and the development
of predictive neural network models based on data that were collected using off-the-shelves
temperature sensors.
Extensive literature studies were conducted on: containers, RFID, cold chain and logistics
operations, the needs of the industry in southern Africa, state-of-the technology in the industry,
the intelligence and communication capabilities for an improved cold chain monitoring system,
temperature modelling and neural networks,
Cross-border field tests were conducted during normally cold chain logistics operations from
Johannesburg (South Africa) to Lusaka (Zambia)and sufficient experimental data were gathered.
These results were analysed using MS Excel and Matlab and numerous visualization
explanations were generated for various temperature profiles behaviours experienced in reefer
containers during transit.
Artificial neural network models were developed by first training using the Levenberg-Marquardt
backpropagation, Bayesian regularization backpropagation and Scaled conjugate gradient
backpropagation with number of neurons based on the rule of thumbs in other to select the best
and fastest achieving function. Predictions, multi-step predictions and step-ahead prediction
beyond targets were generated and visualised for delays events, offloading events and the
complete events during a fresh produce cold chain logistics operation at set points of 2°,5° and
both. The ANN step-ahead prediction beyond targets predicted five cargo temperatures from a
minimum number of five sensors in the trailer (inputs). The prediction horizon was (5
timestamps), (20, 50, 100 timestamps) and (300 timestamps) all at 5min intervals for offloading
events, delays during transits and complete trips events respectively. The models performances were evaluated using the correlation between the target and the
predicted values also known as regression (R) and the model prediction error (MSE). Both
showed values close to 1 and 0 respectively indication of good model results.
Deployable components of the models were built as DLL files to be deployed and incorporated
on the cold chain management software tool created that runs on Microsoft visual studio platform.
A cost benefit analysis model was generated as published in appendix H comparing the average
value of cargo lost per trip, total value of cargo lost per trip, annual turnover per truck and total
annual turnover relating to 2%, 5%, 15%, 35%, 40% and 50% fractions of fruits and vegetables
impacted by cold chain losses | en_US |
