Intelligent Distributed System for Indoor Heat Flow Control
Y. S. Nurakhov, B. Bektugan, K. Nurbergen, T. S. Imankulov and D. Zh. Akhmed-Zaki
Al-Farabi Kazakh National University, Al-Farabi Ave., 71, Almaty, Kazakhstan
Keywords: FPGA, Decision Making, Neural Network, Data Collection Network, Heat Equation, Heat Propagation.
Abstract: In this paper, we consider the software and hardware implementation of an intelligent distributed system for
forecasting and controlling the optimal distribution of heat in the room. Prediction is based on a pre-trained
neural network model. The system uses calculation results of the one-dimensional heat conduction problem
to correct neural model being trained and decides to turn on / off a specific air conditioner depending on the
predicted data.
1 INTRODUCTION
The principle of operation of modern air conditioners
(Dubolazova, 2009) is based on maintaining the
temperature of the room at a given level. The air
conditioner generates a stream of air pre-cooling or
pre-heating it. When the temperature reaches the
desired value, the air conditioner turns off. Air
conditioner sensors continue to record air thermal
values. When the temperature changes to value above
or below the threshold level, the air conditioner
switches on again. Thus, the temperature of the room
is maintained.
Inverter air conditioners have a special approach
to controlling room temperature. Principle of
operation of the inverter air conditioner is that it is
possible to smoothly (multi-stage) adjust the speed of
rotation of the compressor motor, depending on the
heat load in the room. For faster achievement of set
temperature, the inverter controller increases the
speed of rotation of the compressor engine. The air
conditioner starts to work in the forced mode until the
room temperature reaches the set value. Then the
engine speed decreases, but the compressor continues
to operate, maintaining a constant temperature with
minimal deviations (Nagata, 2015).
This approach has a significant drawback. The
temperature sensor located on the air conditioner
itself does not reflect the overall thermal picture of
the room, because of the characteristics of the
premises (external heat sources, batteries, an open
window/door, etc.), the temperature in different areas
may vary very strongly (Svirina, 2016). Therefore,
there is a need to control the air conditioners in such
a way that the system of heat distribution in the room
reacts to sudden temperature changes in certain areas.
Also in the air conditioner does not take into account
the work of other air conditioners. The operation of
each air conditioner is autonomous and controlled
only by reading temperature sensor.
In previously published papers, algorithms for
controlling the temperature in rooms and air
conditioners were analyzed depending on various
criteria. In the article of Tverskoy (Tverskoy, 2012),
the principle of controlling the thermal regime of a
building with radiator and air heating devices in the
heating system is considered. Spitsyn’s work
(Spitsyn, 2012) is devoted to the analysis of indoor
temperature control algorithms. Also, in Nasution’s
paper (Nasution, 2016) the approach for regulation of
air conditioners through controllers with fuzzy logic
to achieve energy saving is considered. The
researchers of the Swiss Federal Institute of
Technology Lausanne present in their works the
simulation of the heat distribution in the construction
of thermo syphon for high heat flux components
(Seuret, 2018).
In this paper, by analyzing surveys in the field of
decision-making systems (Phillips-Wren, 2008;
Rábová, 2005; Averkin, 2011; Vasilescu, 2011), we
propose the implementation of an intelligent system
for effective thermal control. The system uses a
neural network (Kozadaev, 2006; Santhosh Baboo,
2010; Smith, 2006; Smith, 2007) to predict the
distribution of heat based on the history of
temperature changes in the room, received from the
sensors, which is complemented by data from a
numerical calculation of the problem of heat
distribution. Based on forecasts, the optimal
operation mode of air conditioners is chosen to
320
Nurakhov, Y., Bektugan, B., Nurbergen, K., Imankulov, T. and Akhmed-Zaki, D.
Intelligent Distributed System for Indoor Heat Flow Control.
DOI: 10.5220/0007927903200324
In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 320-324
ISBN: 978-989-758-380-3
Copyright
c
 2019 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved

Figure 1: Description of the general system.
achieve the target heat distribution. To predict the
thermal distribution, a neural network has been
developed, which has been trained by the method of
back propagation (Alejo, 2008; Buscema, 1998;
Makin, 2006). The calculation of the numerical
solution for the heat conduction problem is performed
by a computational accelerator based on FPGA
(Field-Programmable Gate Array), thus reducing the
overall load on the system.
2 GENERAL SYSTEM
DESCRIPTION
The system is a software and hardware complex for
optimal adjustment of the temperature regime of the
room. It consists of a module for collecting,
processing and monitoring data on room temperature,
a prediction module based on a neural network and a
decision-making module.
The data-collecting module transmits
temperature data from sensors located along the
premises to the computing module and to the
prediction module. The computational module
performs heat distribution calculations and transmits
the calculated values to the prediction module. The
prediction module, based on a previously trained
neural model, provides predictions of temperature
distribution for various operating conditions of air
conditioners. The decision-making module selects the
optimal operation mode of air conditioners to obtain
the target function of heat distribution in the room and
transmits the corresponding signals to the devices
controlling temperature condition.
3 DATA COLLECTING MODULE
The primary collection and transmission of
temperature data from sensors is performed by a
hardware module based on Arduino UNO. The
network of temperature sensors will cover the
perimeter of the considered area (room). The module
converts the values into the form of floating point
numbers (IEEE 754, 32 bits). Next, the values are
transmitted to the computing device (FPGA) via the
I2C data transfer protocol.
4 COMPUTING MODULE
A hardware prototype of a calculation module based
on the FPGA model Nexys 4 with Artix-7 family
XC7A100T-1CSG324C chip was implemented. The
module uses a microcontroller with Wi-Fi interface
ESP8266 for informational interaction with the
prediction module. The computing module calculates
the problem of heat conduction in the room.
To calculate the heat distribution, heat conduction
equation is used, which looks as follows:






, where ∈0,; ∈

0, 

, (1)
with initial and boundary conditions:


, 0



,


0, 



, 

, 



.
The finite differential scheme form of equation (1)
has the form (Samarskii, 2001):
Intelligent Distributed System for Indoor Heat Flow Control
321

















2






(2)
where τ - grid step along the time coordinate, h - grid
step along the spatial coordinate, 

- coefficient of
thermal conductivity.
Problem (2) is solved by the finite-difference
method using the Jacobi iteration method
(Kuznetsov, 2007), in which the values of points with
indices
i − 1, i, i + 1 from the previous time layer are used
to calculate the value at each point i.
The computing block for each iteration consists of
three sub-blocks. The computational device operates
with real numbers using the Floating-Point Operator
IP Core (Floating-Point Operator v7.1, 2017).
Calculations are made at a frequency of 100 MHz.
Figure 2 shows the logic diagram of the
computational module that computes the values of
each iteration, where result_1 is the result of the first
calculation sub block, result_2 is the result of the
second calculation sub block, result_3 is the result of
the third calculation block.
For the sequential transfer of intermediate data
between computational operations, finite automata
with two states are developed for each of the sub-
blocks:
- STATE_0 - waiting status of incoming
parameters. On receiving input values, the machine
performs the calculation.
- STATE_1 - result output, input parameters reset
and switching machine to the pending state.
The results of the computational block are
recorded in the generated transfer queue. The core of
the standard IP directory FIFO Generator (FIFO
Generator v13.1, 2017) implements the data transfer
unit.
The data transfer unit in turn transmits data to the
Wi-Fi module via the i2c protocol. The data transfer
rate was reduced to 250 bytes/s due to limitations
because of the characteristics of the transmitting
device. Wi-Fi module is programmed using the
Arduino IDE. Wireless network is used to transmit
data to the prediction module.
Figure 2: Diagram of the computing unit.
Figure 3: Input data structure.
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
322

5 INTELLIGENT DECISION
MAKING SYSTEM
The intellectual component of the system consists of
two modules: a prediction module and a decision-
making module.
Prediction module. The heat distribution
prediction module is designed as a neural network
model trained using error back propagation method.
The following parameters are selected as model input
parameters:
- history of changes in the temperature array
obtained from the sensors;
- computed data (based on historical values)
using the heat conduction model;
- status of air conditioners, where 1 - air
conditioner is turned on, 0 air conditioner is turned
off (00 - both off, 01 - first off, second on, 10 - first
on, second off, 11 - both on).
Figure 3 shows the general structure of the input
data.
The output structure corresponds to an array of
temperature values of the expected data.
The artificial neural network has four layers, an
input layer with 12 neurons, 2 intermediate layers of
7 neurons, and an output layer with 10 neurons. The
neural network was trained at 7500 and tested on
1500 data sets. The parameters of speed and moment
of learning are 0.01 and 0.1, respectively. This
network structure is chosen empirically and provides
accuracy with an error of no more than 5%.
The expected data are presented in the form of 317
unique sets of values from sensors obtained from
experiments. The forecasting model assigns a set of
input values to one of such sets, so when building a
forecast, the neural network solves the classification
problem.
Decision making module. The system makes
decisions based on heat distribution predictions for
various combinations of air conditioner operation
modes. The required temperature condition is
represented as a linear target function with a
predetermined temperature. The forecast of heat
distribution for each of the modes of operation are
compared with the target function. The system selects
the mode with the smallest average deviation from the
target function (Figure 4). The parameters of the
selected mode are transmitted to the air conditioners
as a control signal. Air conditioners according to the
command maintain the desired temperature
condition. The whole process is performed in a cyclic
manner with a period of 2 minutes.
Figure 4: Temperature distribution graph for two modes.
6 CONCLUSIONS
The paper considers an approach to the software and
hardware implementation of an intelligent room air
conditioning system. The system has data collection
sensors that transmit in real time information about
the current heat distribution in the room. The data is
saved for later use for learning the neural model. The
neural model is able to predict the further distribution
of heat depending on various parameters and choose
the optimal operation mode of air conditioners.
Implemented a computational accelerator based on
FPGA, the results of which are also used to make
predictions.
Figure 5: The cycle of the system.
Intelligent Distributed System for Indoor Heat Flow Control
323

It is worth to mention that the system, before
reaching the target function, may have time to
perform several control cycles in different regimes.
That in turn will provide a smoother change in
temperature of the room.
Thus, we constructed a prototype of a self-
sufficient intellectual distributed system, which,
depending on a given objective function, can
systematically regulate the temperature mode of a
room.
In the future, it is possible to improve the system
by adding operating modes with different powers for
air conditioners and adapting the system for non-
linear objective functions.
REFERENCES
Alejo R., Sotoca J. M., De La Rosa M. G. Error analysis in
artificial neural networks: the imbalanced distribution
case. - 2008.
Averkin A.N., Belenky A.G. Hierarchical intellectual
decision support systems in complexly structured areas
using expert information. - 2011.
Buscema M., Dr. Back Propagation Neural Networks.
Substance Use & Misuse, 33 (2), 233–270, 1998.
Dubolazova L.V., Air conditioning systems: history, types,
work principle. Scientific works of Dalrybvtuz, 2009.
FIFO Generator v13.1 LogiCORE IP Product Guide
(PG057). Xilinx 2017.
Floating-Point Operator v7.1 LogiCORE IP Product Guide
(PG060). Xilinx 2017.
Kozadaev A. S., Arzamastsev A. A. Forecasting time series
using the apparatus of artificial neural networks. Short-
term analysis of air temperature. TSU Bulletin, Vol.11,
Issue 3, 2006.
Kuznetsov G. V., Sheremet M. A. Difference methods for
solving heat conduction problems: a tutorial. / G.V.
Kuznetsov, M.A. Sheremet. - Tomsk: TPU publishing
house, 2007. - 172 s.
Makin J. G. Back propagation. http:// www.cs.cornell.edu,
February 15, 2006.
Nagata T., Yasin J., Kishi, Wada A. Development of the
Inverter Air Conditioner for Southeast Asia in FY2014.
Hitachi Review Vol. 64 (2015), No. 2.
Nasution H., Dahlan A. A., Nasib A. M., Aziz A. A. Indoor
Temperature Control Split Unit Air Conditioning
System Using Fuzzy Logic Controller. IAENG
International Journal of Computer Science, 43: 4,
IJCS_43_4_01. 2016.
Rabova I., Konecny V., Matiasova A. Decision making
with support of artificial intelligence. Agric. econ. -
Czech, 51, 2005 (9): 385–388.
Phillips-Wren G., Nikhil Ichalkaranje, Lakhmi Jain.
Intelligent Decision Making: An AI-Based Approach. -
2008.
Samarskii Alexander A. The Theory of Difference
Schemes. CRC Press, 786 Pages, 2001.
Santhosh Baboo S., Kadar Shereef I. An Efficient Weather
Forecasting System using the Artificial Neural
Network. International Journal of Environmental
Science and Development, Vol. 1, No. 4, October 2010
ISSN: 2010-0264.
Seuret A., Iranfar A., Zapater Sancho M., Thome J.R.,
Atienza Alonso D. Thermosyphon Cooling System for
High-Performance Computing Data Centers. The
Intersociety Conference on Thermal and
Thermomechanical Phenomena in Electronic Systems
(ITHERM), Marina San Diego, CA, USA, May 29 -
June 1, 2018.
Smith Brian A. Air temperature prediction using artificial
neural networks. - 2006.
Smith Brian A, McClendon Ronald W., and Hoogenboom
Gerrit. Improving Air Temperature Prediction with
Artificial Neural Networks. International Academy of
Computer Science and Technology Vol: 1, No: 10, -
2007.
Spitsyn V. S., Spitsyn V. V. Algorithms for temperature
control in the premises. UDC 621.3.068. - 2012
.
Svirina I. S., Research of temperature fields of educational.
Bulletin of Cherepovets State University, 2016.
Tverskoy M. M., Rumyantsev D. V. Statement of the
problem of optimal control of the thermal regime of a
building with a combined heating system. UDC
681.513.5, - 2012.
Vasilescu C. Effective Strategic Decision Making. - 2011.
ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics
324