Modeling energy supply unit of ultra-low power devices with indoor photovoltaic harvesting
Loading...
Date
2020
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Challenges in the field of logistics have pushed development and integration of cyber-physical systems in these applications. PhyNode as one of these systems has shown promising results for enabling a transportation box with intelligence. However, engineering shortcomings during its development and implementation have shown potential for further research topics. Among them, balancing the Energy Supply Unit (ESU) to avoid periodic battery recharge is the main motivation of this work addressed by its modeling.
For a systematic analysis of PhyNode's ESU, two types of models are developed for each of its three modules, including: Indoor photovoltaic harvesting (IPV), power management device and the battery. First type of models are computationally lightweight for on-board monitoring implementation. In contrary, system level detailed models are more advanced and computationally intensive. They are used to properly dimension the hardware or optimize the operational process during system design phase.
At first IPV devices are analyzed extensively to highlight their differences from solar applications. In addition to the development of a high precision measurement platform for measurement of IPV behavior, collected data is used for model development. Due to wide range of signals, a normalized space is introduced in addition to guidelines for model's parameters estimation. Moreover, a new evaluation criteria is suggested enabling comparison of model's performance in different environmental situations.
A battery measurement setup is introduced for analyzing battery with ultra-low power loads. In addition to the comparison of different battery identification methods, effect of aging on the battery performance has been analyzed. By measurement of PhyNode's load, both developed models are evaluated showing error less than 0.5% on estimation of the models' output.
Furthermore, internal structure of power management device designed for ultra-low power applications is analyzed. Converter and maximum power point tracking as two main parts of this system are modeled separately. Despite suggestion of a partial model based on physical principles of converter, lack of design information leads to a black-box modeling approach. Therefore, two machine learning based models are developed for these parts. Combined model of them is tested on an evaluation data-set, showing a performance with a RMSE of 1.2%.
Finally, a holistic model including all modules builds the overall structure of PhyNode's ESU. This model is tested with real data from different hardware combinations of PhyNode in action for long time periods showing a MAPE less than 1%. Due to the high accuracy of developed model, it is used for simulation of PhyNode in a real world scenarios. In addition, potentials of holistic model are shown by simulating energy balancing after different changes in either hardware or operational process of PhyNode.
Description
Table of contents
Keywords
Modeling, Energy harvesting, Photovoltaic, Ultra-low power, Battery, Energy supply, Cyber-physical systems