CoBotAGV_UA - Automated Guided Vehicles integrated with Collaborative Robots - energy
consumption models for logistics tasks planning
Scheme: Support for Ukrainian researchers under Bilateral Fund of ‘Applied Research’ Programme”
Number of the Action contract: FWD/I/105/CoBotAGV_UA/2022
Start date: 01.10.2022
End date: 30.09.2023
Researcher: PhD Olena Pavliuk
The CoBotAGV_UA focus on the prediction of AGV (Autonomous Guided Vehicle ) energy consumption which is one of enabling methodology for AGVs’ task scheduling in real scenarios (unknown and uncertain environments) including: (i) the developed program module for pre-processing and supplementing partially lost data; (ii) forecast of the AGV battery discharging via the machine learning methods; (iii) Training a DNN neural network system for segmental forecasting of AGV battery cell voltage; (iv) RNN for short- and medium-term battery cell voltage forecasting.
Component (i) includes: (a)analysis of the available dataset obtained from AGV using the UPC UA protocol; (ii) development of methods for detecting and supplementing partially lost data based on analysis of variance; (iii) application of correlation analysis to identify dependencies between parameters that affect battery cell voltage.
Component (ii) includes: (a) the method of setting up the experiment for collecting the historical data for an AGV; (b) developed of algorithm, which includes padding any the suppression spontaneous peaks, the recovery of any lost data and data normalization; (c) the collected data for AGV were analyzed using correlation analysis methods (Pearson, Spearman, and Kendall correlations); (d) a battery discharge prediction approach that is based on the quasi-stochastic signal's probabilistic characteristics.
Component (iii) includes: (a) a multiparameter ANN model using a time window developed; (b) intelligent analysis of battery cell voltage data and its correlated parameters; (c) predictive model consisting of a system of neural networks with deep learning for segmental forecasting of battery cell voltage.
Component (iv) includes (a) analysis of methods for smoothing noise and random outliers in signals; (b) statistical analysis and verification of signals for stationarity and white noise; (c) bringing signals to a stationary form using: differentiation, logarithmization and various smoothing methods; (d) training of a recurrent neural network for short- and medium-term forecasting of battery cell voltage based on an additive model of stationary signals obtained using various signal smoothing methods.