IEEE Systems, Man and Cybernetics Magazine - April 2021 - 23

social force model to conduct simulations for occupant traffic. An improved cuckoo search (ICS) algorithm is adopted
to find the threshold conditions of powering on/off classrooms, by which the schedules for classrooms can be found.
During the scheduling process, an excessive number of
classrooms may be scheduled to be powered on due to the
dynamic nature of student arrival and departure. Thus, a long
short-term memory-recurrent neural network (LSTM-RNN) is
applied to predict the net incoming student occupants, which
can assist in making exact decisions on whether to power on
an additional classroom or not. Experiments have been conducted in different settings, including various classroom sizes
and occupant traffic. Scheduling strategies are proposed to
greatly reduce the number of power-on classrooms and the
total power-on time of classrooms. On the premise of ensuring that most students have sufficient space in a classroom,
the purpose of further saving electricity is achieved. Experimental simulations are conducted to validate the effectiveness of our proposed ICS approach.
Background
In recent years, the crisis caused by climate change has
prompted people to pay more attention to energy conservation and the reduction of carbon emissions [1]. A large portion of the high energy consumption in nonindustrial energy
use in a country comes from educational buildings [2], [3].
Since every country has many educational buildings, energy
use accounts for a large part of school operating costs,
which is the second-largest category of school expenditures
[4]-[6]. This is definitely a good opportunity to save energy
and funds. For example, in mainland China, the average
energy consumption intensity of schools is 121.81 kWh/m2
annually [7]. In Bordeaux, France, the energy and water consumption of a large campus is similar to that of a mediumsized city [8]. British higher education institutions consume
5.2 billion kWh of energy annually [9].
Reasonable suggestions are put forward to realize the
energy-saving potential of educational buildings. Without
major building renovations, nine measures can be taken
(such as the use of reflective coatings) that can save 18.9% of
the total energy consumption and reduce CO2 emissions by
up to 769.78 tons/year [10]. In educational buildings, 85% of
the total energy is consumed by air conditioning and
mechanical ventilation systems, lighting loads, and plug
loads [11]. The energy saving of university buildings can be
achieved by updating the heating, ventilation, and air-conditioning systems and other equipment or strengthening the
insulation to build an enclosure structure [12], [13].
The utilization rate also has a significant impact on the
energy efficiency of educational buildings [14]. Due to the
specific occupancy and schedule of educational buildings,
they need to be more effectively managed. Appropriately
increasing the campus' control and guidance on the existing
resources of educational buildings can also save energy and
reduce costs. The potential of energy saving during occupied
hours for plug and lighting load energy savings are 8.9 and

65.1%, respectively [15]. In a study of the control method of
the central air-conditioning system in educational buildings, the energy-saving rate of the new scheme is 15.27%
compared with the conventional one [16]. The study in [17]
indicates that the best timetable can save 4% of energy during the cooling and heating seasons compared with the
existing timetable.
Students are the main users of instructional buildings,
and they spend about a quarter of their time in school [18].
For students with a heavy curriculum load and scientific
research tasks, they usually stay on campus longer, especially those who live there. The behavior of occupants in a
building has been identified as a key factor affecting the
energy use inside a building [19].
On campus, when classrooms are not occupied for classes, this article reasonably assumes that they are open to college students for studying. As students enter a classroom,
they turn on the lights, ceiling fans, and air conditioners.
These shared classrooms, including lecture halls, large classrooms, and small classrooms, are called classrooms hereafter (when no confusion is caused by this term). This
convenient occupation of classrooms for students is called
turn-on-as-you-go, which usually wastes a large amount of
electricity. Certainly, if only a few classrooms are scheduled
to be powered on for students to enter freely-or an additional classroom is scheduled to be powered on in case the
number of occupants in one classroom reaches its maximum seating capacity-this can save a lot of energy.
But this is not desirable because the main function of
instructional buildings is to provide a conducive environment for students [20]. The occupant density in almost all
buildings is subject to change [21], and the same is true for
instructional buildings. It is necessary to put forward a
dynamic classroom scheduling model with more consideration of social psychological factors and students' preferences for seating. To our best knowledge, no studies have done
so yet. The times when students enter and leave a classroom
are not fixed; thus, the time span to share classrooms during
the day is large. An occupant's study time in a classroom is
relatively short in the evening, e.g., 1-5 h. The flow of students is concentrated in a fixed period of time that is prone
to large short-term coming and leaving flows.
It is particularly important to propose an efficient method for scheduling classrooms to save electricity. The evolutionary algorithm has been widely used in the scheduling
optimization of crude oil, workshops, water conservancy,
and so on [22]-[25], and it has had a significant effect. A
demand-side management model of home appliances is proposed to utilize three evolutionary algorithms to optimize
the scheduling of traditional homes and reduce the peak
power consumption [26]. The particle swarm optimization
(PSO), artificial bee colony algorithm (ABC), and cuckoo
search (CS) have simple principles and parallel searches,
with global optimization ability [27]-[32].
The CS algorithm has fewer parameters and better performance. An ICS algorithm is proposed to use dynamic
Ap ri l 2021

IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE

IEEE Systems, Man and Cybernetics Magazine - April 2021

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - April 2021