CBR-based energy management framework for hybrid UAVs based on state extraction
An apparatus for managing energy in an unmanned aerial vehicle (UAV) including a hybrid powertrain that is configured to provide electric power to onboard power-consuming units coupled to a DC bus and includes a fuel cell coupled through a boost converter, a battery coupled through a bidirectional converter, and a supercapacitor directly coupled to the DC bus. The apparatus includes processing circuitry configured to optimize weighting factors of an equivalent consumption minimization (ECM) model for multiple load profile scenarios, extract a state variable dataset including state variables and fuel cell power reference values, and generate a representative dataset through data reduction. Based on real-time state measurements, the processing circuitry identifies historical cases from the representative dataset and allocates power among the fuel cell, battery, and supercapacitor.
1 . A method for managing energy in an unmanned aerial vehicle (UAV), the UAV including a hybrid powertrain configured to provide electric power to onboard power-consuming units coupled to a DC bus, the hybrid powertrain comprising a fuel cell coupled to the DC bus through a boost converter, a battery coupled to the DC bus through a bidirectional converter, and a supercapacitor directly coupled to the DC bus, the method comprising:
for each of a plurality of load profile scenarios, optimizing a weighting factor of an equivalent consumption minimization (ECM) model;
extracting a state variable dataset based on the plurality of load profile scenarios and the optimized weighting factors, each entry of the state variable dataset corresponding to a time instance and including a plurality of state variables and an associated fuel cell power reference value;
generating a representative dataset of cases by applying data reduction to the state variable dataset; and
performing real-time energy management of the hybrid powertrain by identifying, based on real-time state measurements of the UAV, a plurality of historical cases from the representative dataset and allocating power among the fuel cell, the battery, and the supercapacitor based on the identified plurality of historical cases.
2 . The method of claim 1 , wherein the plurality of load profile scenarios are generated by:
acquiring a load profile from an experiment conducted with respect to the UAV,
adjusting the acquired load profile to obtain a scaled load profile, and
applying payload modulation (PLM), fast-dynamics modulation (FDM), and slow-dynamics modulation (SDM) to the scaled load profile to synthesize the plurality of load profile scenarios.
3 . The method of claim 2 , wherein each of the plurality of load profile scenarios is synthesized to simulate a corresponding degree of load fluctuations of the onboard power-consuming units.
4 . The method of claim 1 , wherein the ECM model includes a first penalty function responsive to a state-of-charge (SOC) deviation of the battery and a second penalty function responsive to a power deviation of the fuel cell, and
the optimizing step further comprises, for each of the plurality of load profile scenarios, optimizing a first weighting factor corresponding to the first penalty function and a second weighting factor corresponding to the second penalty function, so as to minimize an equivalent power consumption of the UAV.
5 . The method of claim 4 , wherein the optimizing step further comprises, for each of the plurality of load profile scenarios, identifying a pair of optimized first and second weighting factors based on a non-dominated sorting genetic algorithm, so as to achieve a minimized hydrogen consumption and a minimized SOC deviation.
6 . The method of claim 1 , wherein the plurality of state variables includes:
a state-of-charge (SOC) deviation of the battery,
a filtered load power,
a deviation between a load power and the filtered load power,
a deviation between a fuel cell power and the filtered load power, and
a deviation between a battery power and a reference charging power.
7 . The method of claim 1 , wherein the generating step further comprises performing k-means clustering to partition the state variable dataset into a number of clusters, so as to identify the representative dataset, the representative dataset having a size substantially smaller than a size of the state variable dataset.
8 . The method of claim 1 , wherein the plurality of historical cases are identified based on a similarity metric representing a degree of similarity between each case in the representative dataset and a current case represented by the real-time state measurements of the UAV.
9 . The method of claim 8 , wherein the similarity metric is a weighted Euclidean distance calculated across the plurality of state variables,
a k-nearest neighbors algorithm is applied to select a plurality of historical cases having smallest weighted Euclidean distances, as the identified plurality of historical cases, and
weights used in the weighted Euclidean distance represents relative importances of each of the plurality of state variables.
10 . The method of claim 8 , wherein a weighted average estimation is applied, based on weights determined based on inverse of distances between the current case and each of the identified plurality of historical cases, to generate a real time power reference value of the fuel cell.
11 . An apparatus for managing energy in an unmanned aerial vehicle (UAV), the UAV including a hybrid powertrain configured to provide electric power to onboard power-consuming units coupled to a DC bus, the hybrid powertrain comprising a fuel cell coupled to the DC bus through a boost converter, a battery coupled to the DC bus through a bidirectional converter, and a supercapacitor directly coupled to the DC bus, the apparatus comprising:
processing circuitry configured to
for each of a plurality of load profile scenarios, optimize a weighting factor of an equivalent consumption minimization (ECM) model,
extract a state variable dataset based on the plurality of load profile scenarios and the optimized weighting factors, each entry of the state variable dataset corresponding to a time instance and including a plurality of state variables and an associated fuel cell power reference value,
generate a representative dataset of cases by applying data reduction to the state variable dataset, and
perform real-time energy management of the hybrid powertrain by identifying, based on real-time state measurements of the UAV, a plurality of historical cases from the representative dataset and allocating power among the fuel cell, the battery, and the supercapacitor based on the identified plurality of historical cases.
12 . The apparatus of claim 11 , wherein the plurality of load profile scenarios are generated by:
acquiring a load profile from an experiment conducted with respect to the UAV,
adjusting the acquired load profile to obtain a scaled load profile, and
applying payload modulation (PLM), fast-dynamics modulation (FDM), and slow-dynamics modulation (SDM) to the scaled load profile to synthesize the plurality of load profile scenarios.
13 . The apparatus of claim 12 , wherein each of the plurality of load profile scenarios is synthesized to simulate a corresponding degree of load fluctuations of the onboard power-consuming units.
14 . The apparatus of claim 11 , wherein the ECM model includes a first penalty function responsive to a state-of-charge (SOC) deviation of the battery and a second penalty function responsive to a power deviation of the fuel cell, and
the processing circuitry is further configured to, for each of the plurality of load profile scenarios, optimize a first weighting factor corresponding to the first penalty function and a second weighting factor corresponding to the second penalty function, so as to minimize an equivalent power consumption of the UAV.
15 . The apparatus of claim 14 , wherein the processing circuitry is further configured to, for each of the plurality of load profile scenarios, identify a pair of optimized first and second weighting factors based on a non-dominated sorting genetic algorithm, so as to achieve a minimized hydrogen consumption and a minimized SOC deviation.
16 . The apparatus of claim 11 , wherein the plurality of state variables includes:
a state-of-charge (SOC) deviation of the battery,
a filtered load power,
a deviation between a load power and the filtered load power,
a deviation between a fuel cell power and the filtered load power, and
a deviation between a battery power and a reference charging power.
17 . The apparatus of claim 11 , wherein the processing circuitry is further configured to perform k-means clustering to partition the state variable dataset into a number of clusters, so as to identify the representative dataset, the representative dataset having a size substantially smaller than a size of the state variable dataset.
18 . The apparatus of claim 11 , wherein the plurality of historical cases are identified based on a similarity metric representing a degree of similarity between each case in the representative dataset and a current case represented by the real-time state measurements of the UAV.
19 . The apparatus of claim 18 , wherein the similarity metric is a weighted Euclidean distance calculated across the plurality of state variables,
a k-nearest neighbors algorithm is applied to select a plurality of historical cases having smallest weighted Euclidean distances, as the identified plurality of historical cases, and
weights used in the weighted Euclidean distance represents relative importances of each of the plurality of state variables.
20 . The apparatus of claim 18 , wherein a weighted average estimation is applied, based on weights determined based on inverse of distances between the current case and each of the identified plurality of historical cases, to generate a real time power reference value of the fuel cell.