24-09-2016, 02:21 PM
1456072570-Newselfadaptivebatinspiredalgorithmforunitcommitmentproblem.pdf (Size: 418.69 KB / Downloads: 5)
Abstract: Bat-inspired algorithm (BA) is a new evolutionary meta-heuristics algorithm inspired by a known technique of bats for
finding prey. This study presents a self-adaptive BA to solve the unit commitment (UC) problem. The applied self-adaptive
technique increases the population diversity and improves the exploration power of BA which results in better solutions and
higher speed of convergence in solving the UC problem. This study, also, applies simple methods to handle the minimum
on-/off-time constraint and spinning reserve requirement in generation of all solutions directly and without using any penalty
function. The performance of the proposed method is verified by applying 10 up to 100-unit systems as well as a Taiwan
power (Taipower) 38-unit system in a 24 h scheduling horizon.
Introduction
In the mid-term planning of the power systems, the unit
commitment (UC) problem has a very important role in
decreasing generation cost and economic operation of units
and resources. From an economic viewpoint, the objective
of UC is to find the best combination of committed units to
generate power at each hour of the scheduling horizon,
while satisfying the load demand with the minimum total
operational cost [1]. This problem involves several different
constraints which should be satisfied such as the minimum
and maximum limit of power generation of each unit,
spinning reserve required at each hour, minimum on-/
off-time and ramp up/down rates of units [2].
Considering the minimum on-/off-time and ramp rate
constraints, the operation status and generation level of
units at each hour depend on the previous hours which
make this problem as one of the most complex problems in
electric power systems [3, 4]. In general, the UC problem
can be defined as a non-linear, non-convex, mixed integer
combinatorial optimisation problem.
Numerical calculations methods are capable of providing
the exact optimal solution of the UC problem. However,
this method is not suitable for a large size power system
because of the excessive execution time requirement. Hence
up to now, other methods have been widely used to solve
this problem such as priority list [5], dynamic programming
(DP) [6, 7], branch-bound [8], mixed integer programming
[9] and Lagrangian relaxation (LR) [10]. Most of these
methods are simple and provide fast solutions. The ‘curse
of dimensionality’ is the main challenge for these methods
when the committed units increase [11]. Mixed integer
linear programming (MILP) [12] has also been proposed as
an efficient method in solving large UC problems.
However, there are certain constraints such as valve points or forbidden zones which may be difficult to capture with
MILP and other similar methods.
In addition to the numerical-based optimisation method,
meta-heuristic algorithms have been used to solve the UC
problem. These techniques can easily handle different types
of system and operational constraints [13]. Genetic
algorithm (GA) is a well-known method of this branch [11,
14–16]. It can find near optimal solutions, but in most
cases, the search process fails to converge to the optimal
solution. Requiring a long computational time in some
problems is another important drawback of GA [17].
Simulated annealing (SA) is a different meta-heuristic
algorithm used to solve the UC problem [18]. SA can easily
accommodate various constraints of the UC problem and
also be applied to large-scale power systems. This method
provides feasible and near optimal solutions, however, a
long time is often required for convergence. In comparison
with other evolutionary algorithms, it is proven that particle
swarm optimisation (PSO) is able to obtain more efficient
solutions in terms of convergence speed, accuracy and
robustness [19–22].
Bat-inspired algorithm (BA) is a new evolutionary
algorithm used in this paper to solve the UC problem. This
optimisation approach is inspired by biological evolution of
bats in finding prey and has the advantages of simplicity,
easy implementation and providing faster solutions
compared with other evolutionary algorithms [23]. In
practice, BA utilises some of the PSO and SA
characteristics simultaneously and so takes advantage of
both algorithms. To enhance the exploration ability of the
proposed algorithm, a self-adaptive reformation mechanism
with some mutant rules is implemented in this study. Using
a probabilistic selection procedure, one of these rules is
selected to apply to each particle at each epoch of the
algorithm. As will be shown, the reinforced BA provides
faster and more robust solutions for the UC problem in
comparison with other performed algorithms.
Unlike most of the evolutionary optimisation methods
which applied penalty functions to consider the effect of
violations on the minimum on-/off-time and spinning
reserve constraints, this paper utilises simple procedures
based on the duration of operation in each state of the
unit (ON or OFF) to handle these constraints after
generation of each solution [16]. Therefore there is no
need to apply penalty functions for the mentioned
constraints. In addition, the number of feasible solutions
and subsequently the speed of convergence will be
increased in the optimisation process