=Paper=
{{Paper
|id=Vol-2267/318-322-paper-60
|storemode=property
|title=Smart buildings energy savings with gradient boosting algorithm
|pdfUrl=https://ceur-ws.org/Vol-2267/318-322-paper-60.pdf
|volume=Vol-2267
|authors=Eugene Yu. Shchetinin,Evgenia A. Popova
}}
==Smart buildings energy savings with gradient boosting algorithm==
https://ceur-ws.org/Vol-2267/318-322-paper-60.pdf
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
SMART BUILDINGS ENERGY SAVINGS WITH GRADIENT
BOOSTING ALGORITHM
Shchetinin E.Yu.a, Popova E. A. b
Financial University under the Government of the Russian Federation
E-mail: a riviera-molto@mail.ru, b evgenia.popova1397@gmail.com
In this article, authors pay attention to global tendencies in energy consumption and production
specifically shifting to a new energy pattern which is aimed at resource-saving and resource
efficiency. This new energy pattern is connected with the industrial internet of things. Specifically, in
commercial buildings, the advanced measurement infrastructure is commonly used. It has resulted in
the availability of high-frequency interval data. These data can be used in a number of energy
efficiency tasks, including the query response, the definition and diagnosis of malfunctions, the
optimization of heating, ventilation and air-cooling systems. The arrays of these data enable the use of
advanced statistical training models and, therefore, lead to accurate estimates of energy
conservation.We underlined the importance of a movement to smart buildings equipped with smart
meters, Big Data technologies, special analyzing mechanism – the Gradient Boosting machine. The
algorithm of gradient boosting is a powerful tool that has been applied in the fields with intensive data
application, including ecology, computer vision, biology. In this article, a method for modeling the
baseline energy consumption based on the gradient boosting algorithm is proposed.
Keywords: smart meters, smart buildings, energy production, consumption, GBM, Big Data,
renewables, BEMS, BMS, optimization, efficiency, energy saving.
© 2018 Eugene Yu. Shchetinin, Evgenia A. Popova
318
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
1. Introduction
Nowadays the world is becoming digital, the transition between physical reality and virtual is
disappearing. Enterprises are transformed in accordance with changes in the behavior of consumers
who always use different gadgets. Artificial intelligence processes large data in giant super-powerful
data centers, spawning new classes of software products - voice assistants, "smart" wearable devices,
robot-guards. All these possibilities are accompanied by one side effect - humanity consumes more
and more electricity.
2. Global tendencies
According to the forecast of International Energy Agency global electric power generation
will increase by 63% in the nearest 30 years (picture 1). This trend is caused by rapid economic
development of such countries as China, India, Southeast Asia, Middle East etc. Global energy
demand will grow by 70% from 2015 to 2040. That is why, there is a new global energy model which
implies 4 main tendencies: shifting to renewables sources of energy as additional and more clean types
of energy, decentralization of electricity generation, focusing on energy efficiency of buildings and
urban infrastructure, mastering the latest technologies like an internet of things or more precisely
industrial internet of things.
Based on these tendencies, the world industry of production and distribution of energy is
restructuring. In 2016, according to the analytical company IHS Markit, the volume of electricity
supplied by photovoltaics rose by 34% in comparison with a year earlier growth which was 32%. The
Bloomberg New Energy Finance report, published in 2017, says that the cost of generating electricity
using solar cells in Germany, Australia, the United States, Spain, Italy has already been equal to the
cost of energy production by burning coal. The rapid growth of a renewables consumption leads to
decentralization of industry. Previously, the main energy generators were large corporations and state-
owned entities. Now ordinary citizens and private companies install solar panels and wind generators
on the roofs of residential buildings and office buildings. In order to implement energy efficiency
processes it is important to understand the structure of primary energy consumption and choose the
sphere of such innovations.
In accordance with Annual Energy Outlook 2018, the biggest part of global electricity
consumption is used for residential needs and slightly less amount of electricity is used for industrial
purposes and commercial needs (picture 3). In 2045 there will be quite familiar situation despite some
peculiarities: the share of industrial sector is going to be a little bit higher than the share of residential
sector; total quantum of consumed electricity is going to be higher in all sectors [1]. The U.S.
Department of Energy and UNEP estimates that buildings heating and cooling systems accounts for
nearly 18-24% of all energy usage. Besides, they produce 40% of total carbon footprint. At the same
time, buildings certified as "green" throw 34% less carbon dioxide, consume 25% less energy, 11%
less water [2]. Owners save lots of money and above this receive benefits from the government - many
states are struggling for the reduction of greenhouse gas emissions. So, now such technologies are
becoming more and more popular.
According to Navigant Research, the global smart buildings market will triple in the next few
years. Now it is estimated at a level of 3.6 billion dollars. It is assumed that in eight years the market
will reach the volume of $ 10 billion. According to Research & Markets, now the market of building
management systems around the world is estimated at $ 6.65 billion. And every year it is growing by
17%. So, in 2023 it will be 19 billion dollars [1]. Changings in the whole energy model and creation of
“green” buildings require new intelligent distribution networks which are called “smart grids” that
should be bidirectional. Intelligent electrical networks allow users to be disconnected in those
moments when they can provide themselves with energy, take the excess energy that they generated,
as a payment used before, redistribute it, and so on. This cannot be done with a help of "analog"
networks, only on digital ones, equipped with sensors, switches, remote monitoring and automatic
intelligent control. The course on energy efficiency in recent years has been taken by many companies
around the world. They digitized the infrastructure and converted buildings into "smart" ones. The
building energy management systems (BEMS) are in great demand. According to Navigant Research,
319
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
the market for such analytical software solutions will reach $ 11 billion by 2024, annually growing by
18% since 2015 [3].
3. Some examples of smart buildings
Already there are many examples of the usage of smart technologies. For example, SIBUR is
working on the implementation of IoT and a number of other technologies as part of a long-term full-
scale digitalization strategy that covers all stages of the chemical company's value chain. The aim of
this work is to increase productivity, create the basis for rapid and scalable changes, and provide
additional opportunities for the professional development of employees. What is more, "AUCHAN
Retail Russia" has launched a project in order to reduce electricity costs by 20% due to the transition
to intelligent equipment and new approaches in infrastructure management.
Even larger and more complex example is the construction of a private medical complex
Grand Medica in Novokuznetsk. The building is designed from the very beginning so as not to allow
wasteful consumption of electricity. The system is really smart and keeps track of all actions in real
time. Let's say someone mistakenly turned on the heating more strongly, instead of weakening the air
conditioning. This is not such a rare mistake, and as a result, heating and air conditioning work against
each other, spending the day-gi companies in vain. The EcoStruxure platform will notify dispatchers
of such waste or automatically correct it (depending on which scenario is provided by the user of the
platform).
4. Machine learning models applications
4.1. Gradient boosting machine
In modern buildings lots of things are connected to the internet. sensors are installed on all
sections of the smart energy infrastructure: on consuming devices (office equipment, kitchen and
climate equipment, uninterruptible power supplies, etc.), switchboards, switches, emergency
generators and solar batteries, lines power transmission inside the building and beyond. From them,
readings can be taken in real time, with a certain periodicity or only when it is necessary.
Anyway, this information is accumulated in the storage centers in the form of "Big Data"
These huge arrays of megabytes contain a lot of hidden valuable information, which can be obtained
with the help of special analytical tools. If properly dispose of, the energy system of the building will
be more economical, manageable, reliable and safe. Speaking about the concrete mathematical
mechanism it may be the gradient boosting machine. The gradient boosting machine (GBM) is a
powerful machine learning algorithm that is gaining considerable traction in a wide range of data
driven applications, such as ecology, computer vision, and biology [3]. The results of implementation
of such algorithm show that using the gradient boosting machine model can improve the R‐squared
prediction accuracy and the CV(RMSE) in more than 80 percent of the cases, when compared to an
industry best practice model that is based on piecewise linear regression, and to a random forest
algorithm [4], [5], [14]. Let’s describe this algorithm more specifically.
The main principle is that several simple models, called "weakly learning models," should be
combined into one iterative scheme for selecting parameters in order to obtain the so-called "strong
learner", i.e. models with improved prediction accuracy. The GBM method can be considered as an
algorithm for numerical optimization, the purpose of which is to find an additive model that minimizes
the loss function. Thus, the GBM algorithm incrementally adds at each step a new decision tree that
effectively reduces the loss function. Namely, in the regression model, the algorithm begins with its
initialization, which is usually a decision tree minimizing the loss function (RMSE), and then at each
step a new decision tree is adjusted to the current residue and added to the previous model to update
the residuals. The algorithm continues to work until the maximum number of iterations is reached or
the specified accuracy is achieved. Thus, the model can be improved precisely in those parts where it
still poorly estimates the residuals [6].
The GBM algorithm will be more efficient if at each iteration the contribution of the added
decision tree is considered with the help of some hyper-parameter that can intuitively characterize the
320
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
learning speed. The idea of the procedure for selecting a hyper-parameter is that a greater number of
small steps provides higher accuracy than a smaller number of large steps [11]. The parameter can
take a value from 0 to 1, and the smaller it is, the more accurate will be the model. However, the
choice of a stronger shrinkage (less) implies a larger number of iterations to achieve convergence,
since the value is inversely proportional to the number of iterations.
Another way to improve the accuracy of predicting the GBM algorithm is to add
randomization to the estimation process. At each iteration, instead of using a complete set of data, a
subset is used to evaluate the decision tree (usually without replacement). However, in order to assess
the effect of reducing the number of data points on the quality of model fitting, several subsamples of
different dimensions need to be checked. In the GB model, there are four hyper-parameters that are
needed to be configured: (1) d the depth of decision trees, which also determines the maximum order
of the model; (2) K is the number of iterations, which also corresponds to the number of decision
trees; (3) the learning rate, which is usually a small positive value between 0 and 1, the decrease of
which leads to a slowing of the evaluation procedure, thus requiring the user to increase K; (4) a piece
of data that is used at each iteration step [6], [7]. The most popular method for choosing hyper-
parameters is search grid method. This concept is to define a grid of combinations of hyper-parameter
values, build a model for each combination, and select the optimal combination using metrics that
quantify the performance of the model in terms of predictive accuracy. Obviously, it is not
recommended to use the same observations that were used as training data for estimating models for
comparing predictive indicators. Therefore, it is necessary to evaluate the accuracy on an independent
set of data points. Ideally, the available data should be divided into two samples: the training sample
and the test sample. [14]
4.2. What is necessary to create a smart building
The first step is to organize energy consumption accounting. Smart meters will show at what
time which parts consume too much and why. What equipment is installed, and which gadgets
consume too much and why. Already at this stage, there can be a positive economic effect thanks to
identifying the cause of energy over expenditure. Perhaps the printer has broken down and therefore
consumes more than it should or is simply needs to be replaced with a more energy efficient one.
Perhaps the behavior of employees should be adjusted - they print unnecessary documents, or, for
example, someone turns on the heater, and another employee opens the window, because he is hot.
Perhaps the heating system spends a lot of electricity, because the doors are open too often and are not
separated from the room by a thermal buffer.
Cohesion of the infrastructure gives one more important effect: devices can exchange data and
signals, which makes it possible to realize the idea of an "intelligent enterprise". For example, data
from meters, both for predictive repairs, and for monitoring the supply chain. Information from them
can be received in the ERP (Enterprise Resource Planning) accounting system, which in real time will
monitor the speed of business processes and data in other integrated business systems, including
external ones [11], [12]. Despite ERP system there is BMS system. The intelligent BMS system
independently maintain the air in the building at a certain level of purity and humidity, regulate the
temperature in the room, spending a minimum of resources. Monitoring of gas and water pipes are
carried out automatically. If there is a gas leak or a water leak, the notification system will
immediately trigger, and an appropriate message will be displayed on the dispatcher's monitor. The
supply of water and gas can be automatically terminated. The building with the BMS system is more
reliable from all points of view, because it reduces the risk of unexpected breakdowns and allows to
react promptly to any emergency situation.
5. Conclusion
So, in conclusion we can say that in order to create a better world, it is important to change
ideas about energy, its generation, consumption, distribution. We see the creation of a new energy
model, which implies efficiency, environmental friendliness, and security. Moreover, there is a
requirement in a more intelligent equipment to monitor the reliability of energy supply and to optimize
production and consumption. Besides, we need systems that meet the challenges of sustainable
321
Proceedings of the VIII International Conference "Distributed Computing and Grid-technologies in Science and
Education" (GRID 2018), Dubna, Moscow region, Russia, September 10 - 14, 2018
development and comply with new standards, rules and codes. Also, there is a necessity in an
infrastructure that makes it easy to introduce electrical components and systems, protects people and
property from cyberattacks. We need a new, holistic approach to managing electricity. GBM
machinery algorithm allows to realize smart buildings mechanism via building a decision tree with
high accuracy. Already there are lots of successful examples which were mentioned in this paper. So,
smart buildings are the step into the future.
References
[1] EIA Conti J. J. et al. Annual energy outlook 2018 //US Energy Information Administration. –
2018.
[2] IEA Statistics I. E. A. Electricity information 2012 //International Energy Agency. – 2012.
C.~W. Gelling, The Smart Grid: Enabling energy efficiency and demand response. The Fairmont Press
Inc., 2009.
[3] Hong W.C., Intelligent Energy Demand Forecasting, Springer Verlag, London, 2013.
[4] Breiman L., Bagging predictors, Mach. Learn. 24 (2), 123–140, 1996.
[5] Friedman J.H., Stochastic gradient boosting, Comp. Stat. Data Anal. 38 (4), 367–378, 2002.
[6] Kuhn M., Johnson K., Applied Predictive Modeling, Springer, New York, 2013.
[7] Shchetinin E.Yu., Lyubin P., Fast two-dimensional smoothing with discrete cosine transform,
Springer Communications in Computer and Information Science (CCIS), 678, 646-656. Springer,
Berlin, 2016.
[9] Geurts P., Ernst D., Wehenkel L., Extremely randomized trees, Mach. Learn. 63(1), 3–42, 2006.
[10] Srivastav A., Tewari A., Dong B., Baseline building energy modeling and localized uncertainty
quantification using Gaussian mixture models, Energy Build. 65, 438–447, 2013.
[11] Zhao H.X., Magoulès F., A review on the prediction of building energy consumption, Renew.
Sustain. Energy Rev. 16(6),3586–3592, 2012.
[12] Heo Y., Zavala V.M., Gaussian process modeling for measurement and verification of building
energy savings, Energy Build. 53, 7–18, 2012.
[13] Price S., Mahone A., Schlag N., Suyeyasu D., Time dependent valuation of energy for developing
building efficiency standards, in: Report Prepared for the California Energy Commission, 2011.
[14] Shchetinin E.Yu., Cluster-based energy consumption forecasting in smart grids, Springer
Communications in Computer and Information Science (CCIS), 919, 446-456. Springer, Berlin, 2018.
322