Feasibility-guaranteed machine learning unit commitment: Fuzzy Optimization approaches
Bala Venkatesh,Mohamed Ibrahim Abdelaziz Shekeew,Jessie Ma
DOI: https://doi.org/10.1016/j.apenergy.2024.124923
IF: 11.2
2024-11-28
Applied Energy
Abstract:The unit commitment (UC) problem is solved several times daily in a limited amount of time and is commonly formulated using mixed-integer linear programs (MILP). However, solution time for MILP formulation increases exponentially with the number of binary variables required. To address this, machine learning (ML) models have been attempted with limited success as they cannot be trained for all scenarios, whereby they may contain false predictions leading to infeasibility, hindering their practical applicability. To overcome these issues, we first propose a hybrid deep learning model comprising a convolutional neural network (CNN) and bidirectional long-short-term memory (BiLSTM) to predict the UC decisions. Second, we incorporate these predictions as non-binding fuzzy constraints, enhancing the traditional UC model and creating an ML-fuzzy UC model. Two implementations of non-binding fuzzy constraints are studied. The first develops each ML decision variable as a separate fuzzy set, while the second creates one fuzzy set per hour, considering all decisions within. Introducing ML-UC decisions as non-binding fuzzy constraints ensures the ML-fuzzy UC model has a feasible solution if the basic MILP-UC problem does, while leveraging ML predictions. Moreover, the proposed model benefits from a reduced solution space, leading to substantial reductions in computing time. Results on IEEE 118-bus and Polish 2383-bus systems demonstrate 92 % and 89 % computation time reductions for both systems, respectively and achieve 100 % feasibility for both seen and unseen datasets when the basic MILP-UC problem has a feasible solution.
energy & fuels,engineering, chemical