FIELD OF INVENTION
The invention in general relates to a system to determine day-ahead
loading pattern forecast of heavy machineries in industries and proactive
control of peak load overshoot by curtailing pre-designated individual
heavy machineries. This method advocates continuous monitoring of
electrical load, futuristic prediction of load trends based on historical
data and managing heavy machineries electrical loads to avoid exceeding
maximum demand limit.
BACKGROUND OF THE INVENTION
Electrical power needs of industries are increasing day-by-day at a
tremendous pace. It has become necessary to implement possible control
and management schemes to utilize the available electrical energy in an
efficient manner. If power management of heavy machineries in an
industry can be decentralized to the level of the end user and carried out
locally, then it is expected to improve the power of the country to some
extent. The industries cannot draw power more than the permissible
limit specified by the utility companies. This permissible limit is
designated as ‘maximum demand’. The utility companies impose penalty
on industries for exceeding the maximum demand. Industries become
tapped out when a sudden unprecedented load overdraws power and
causes maximum demand overshoot. The power management system
requires complex control circuitry and real time data from heavy
machineries for control of maximum demand and data acquisition. Such
complex systems are to custom built involving highly sophisticated
processors and huge capital investments. Even with this high cost set-up
proactive control to curtail maximum demand overshoot it requires

manual intervention.
Therefore, there is need for a cost effective load predicting system that
curtails maximum demand overshoot proactively to benefit the Heavy
Engineering Industries.
US patent no.US8660706 B2, Frank Szemkus details a decentralized
control and data recording unit for controlling the decentralized energy
resources, a database for storing operating data and/or operating
parameters, and also a network communications interface for exchanging
data and/or control commands with external units via an external
network. However, the system lack proactive power management
controls.
US patent no. US8600559 B2, Wojciech Grohman et. al provides an
HVAC data processing and communication network. This network
includes a sensor and a local controller. The sensor is configured to
detect a fault condition associated with operation of a demand unit. The
local controller, associated with the demand unit, is configured to receive
sensor data from the sensor and to communicate the sensor data over
the network. A network controller is configured to receive the sensor data
via the communication network and to generate an alert in the event that
the sensor data indicates the fault. Such a system senses the real time
load and sends signal for turn off the load points.
However, this system is not a proactive management system and requires
real time data always.
OBJECT OF THE INVENTION
An object of the invention is to reduce the maximum demand by
proactive power management and control to avoid the penalty levied by

power utility companies.
Another object of the invention is to use artificial neural network for
smart scheduling schemes of switching ON/OFF heavy machineries, less
priority-high load consuming electrical equipment and distribution
panels supplying to fan/lighting zones.
Yet another object is to develop switching assembly for managing
remotely located devices either through local area network for devices
with Ethernet accessibility or through wireless trans-receiver switching
system.
SUMMARY OF INVENTION
The invention relates to the design, development and implementation of
an intelligent Predictive Power Management system. The system uses
artificial intelligence technique known as neural networks to predict the
load on a day-ahead basis by learning from the historical load and
weather data. Inferring anticipated peak load times from the forecasted
data, a switching circuit sends appropriate transmission data packets to
switch ON/OFF individual loads either through wired, local area network
or wireless communication modes depending on operation and control
requirements. The switching system comprises of a microcontroller,
Ethernet controller, and wireless transmitter on the control side and
receivers, signal conditioners and relays on load side. Forecasted data
obtained from the neural network are fed to the microcontroller for
devising control signals at definite time. The invention aids in load
scheduling on a proactive basis by curtailing peaking of load before the
contractual demand limit is exceeded and thereby avoiding penalties to
power utility companies and in turn substantial savings to the
organization.

BRIEF DESCRIPTION OF ACCOMPANYING DRAWING
The invention is described herein by the following description of the
accompanied drawings.
Figure 1 shows typical load curves and control relay actions on exceeding
the approved maximum demand values.
Figure 2 shows the local area network topology for relay control.
Figure 3 shows the relay control mechanism using local area network.
Figure 4 shows the radio-frequency wireless network topology for relay
control.
Figure 5 shows the relay control mechanism using radio-frequency
wireless system.
Figure 6 shows the control logic for training the neural network with
historical power and weather data.
Figure 7 shows the control logic for day-ahead forecasting of load.
DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT
Figure 1 shows a typical load curve for a particular day and their
corresponding forecasted load values. The relay control that operates
selectively to prevent exceeding of maximum demand are also shown. X-
axis values represent the time of the day in hours. Y-axis represents the
apparent electrical power values in kVA. Maximum Demand (MD) is the
approved contractual limit of the industry. 97.5% of the MD is
considered for activating the control action by the Predictive Power
Management system. Thick curve lines indicate the actual load power
obtained from an online energy integrator (4) (Fig.2). Dotted lines indicate
the day-ahead load forecasted for the same time by the neural network
using historical power and weather data. When actual load power
reaches point A, forecasted power is anticipated to cross the threshold

limit (97.5% of MD). This control system activates and cuts off
appropriate loads to keep the actual consumption under the MD values.
Point B indicates the probable end time of anticipated MD overshoot from
the forecasted power.
Figure - 2 shows the control mechanism using local area network
connectivity. Actual power consumption is read from the 3-phase
incoming power supply [1] by integrating the voltage, obtained from a
potential transformer [2], with the current, obtained from a current
transformer [3] in an energy integrator [4]. The temperature and energy
consumption data are fed to the processor [5] and logged periodically.
Ambient temperature for the day is also measured by a weather
monitoring station [6].
Prerecorded power and temperature data are used as inputs to the
neural network program inside the processor. This program learns from
the input and forecasts the power consumption pattern for the next day.
The forecasted load pattern is utilized for devising a switching scheme for
the relay control mechanisms. The processor (5) continuously monitors
the current consumption trend, logs the historical power and weather
data for automatically training the neural network on a daily basis and
makes decisive control actions on running devices for regulating the peak
demand either proactively based on neural network’s forecasted schedule
or reactively by reading the present conditions.
Optimal combination of switching schemes will be carried out at different
time slots (working/non-working hours) and a look up table will be
created. Non-conventional bio-inspired optimization techniques like
genetic algorithms, ant colony algorithms, artificial bee colony algorithms
etc., may be implemented
and the best switching scheme will be stored in the memory. The daily 24

hour load pattern and weather parameters will be added up in the
training patterns and training is carried out for learning of the neural
network. Finally, the forecasting of the day-ahead load pattern is carried
out by giving the previous days load and weather parameters as test
inputs. The output of the neural network will be the day-ahead predicted
load and will act as the decisive input for the automatic switching
scheme. Whenever the load reaches threshold of the permissible contract
limit, based on the time interval between A and B (Figure – 1), a suitable
switching scheme will be selected from the look up table and necessary
control signals are sent to the relay units from the processing station. To
take immediate and accurate decisions based on load forecasting, hybrid
neural networks have been used for forecasting.
In algorithms that has a continuous and differentiable activation
function and which is significantly more powerful than the one based on
first generation neurons can solve complex pattern recognition problems.
When a suitable switching scheme is selected, then the transmitter
station will aggregate control command data packets. The control
command for each load is sent separately. Command packets consist of 4
bytes. Two identification tags, ID1 and ID2 will specify the location of the
load to be controlled (all controlled loads are given a unique identifier).
The COMMAND packet will be ‘1’ if the load is to be cut or ‘0’ if the load
is to be released from control. The transmitter station will calculate a
checksum by performing a single byte XOR operation with ID1, ID2 and
COMMAND. The result, which is a single byte, is stored in CHECKSUM.
The checksum will determine if the data packet that is received at the
relay side is the same as the one sent from the transmitter station. This
can be verified by performing a single byte XOR operation on ID1, ID2,
COMMAND and CHECKSUM. If the result is zero, then the data packet is
valid. The control signals can be sent to the individual loads by utilizing

the local area network [7] inside the industry or a wireless system. From
the local area network [7], an Ethernet switch [8] is provided near the
individual electrical load [9]. The relay control box [10] is connected to
the Ethernet switch [8] via cable [11] for receiving control command
packets.
Figure – 3 shows the single line diagram of the relay control box
operating through local area network connectivity [7]. The Ethernet
controller forms the LAN interface [12]. The controller is compatible with
IEEE 802.3 standards. The controller is interfaced with an 8-bit
microcontroller [14] through the Serial Peripheral Interface bus [15]. The
microcontroller sends ON/OFF signals to a signal conditioning circuit
[16] which then controls a NC relay [10] with contact rating of at least
240V and 30A. Based on the activation of the NC relay [10], the electrical
load [9] receives power from the 1-phase supply or gets cut-off.
Figure – 4 shows the transmission of control signals from the processor
via wireless RF for remote locations with no LAN accessibility. The
processor station [5] communicates with the wireless RF transmitter [17]
by means of RS232 serial interface [18]. The data packets are encrypted
for protection from interference. A header consisting of repeating ‘1’s and
‘0’s is attached to the beginning of the data packet and transmitted using
an encoding scheme. The header ensures that the wireless receiver locks
onto the start of the data packet signal. The encoding scheme
incorporated ensures that the data stream will have a constant average
power level, keeping the automatic gain control on the receiver side at a
constant setting. The data transfer is through the wireless transmitter
antenna [19]. The packets are received by the relay control box [20] for
controlling the electrical load [9] through the receiver antenna [21].

Figure – 5 shows the block diagram of the relay control mechanism using
wireless RF system. The RF receiver [22] sends signals captured from the
antenna [21] to a microcontroller [14]. The communication interface is a
standard general purpose IO peripheral on the ATMEGA because the
decoding of the signal is carried out in the stored firmware, unlike in
RS232, which is a hardware protocol. The ON/OFF signals are sent from
the microcontroller [14] to a signal conditioning circuit [16], which then
controls a NC relay [10] with contact rating of at least 230 V and 30 A.
Based on the activation of the NC relay [10], the electrical load [9]
receives power from the 1-phase supply or gets cut-off.
Figure – 6 shows the flow control logic for training the neural network
using the historical power consumption and corresponding hourly
weather data. The abbreviations used in the flow diagram are expanded
as below:
IH – summed input to hidden layer
V – Weights between input and hidden layer
BV – bias given to hidden layer nodes
OH – output from hidden layer nodes
FV – activation function for the hidden layer nodes
IO – summed input to output layer nodes
W – Weights between hidden and output layer
BW – bias given to output layer nodes
OO – output from the neural network
FW– activation function for output layer nodes
Figure – 7 shows the flow control diagram of one day-ahead forecasting
operation performed by the neural network using today’s power and
weather data. The abbreviations used in the flow diagram are expanded
as below:

IH – summed input to hidden layer
V – Weights between input and hidden layer
BV – bias given to hidden layer nodes
OH – output from hidden layer nodes
FV() – activation function for the hidden layer nodes
IO – summed input to output layer nodes
W – Weights between hidden and output layer
BW – bias given to output layer nodes
OO – output from the neural network
FW – activation function for output layer nodes

We Claim:-
1. A system for analyzing industrial load pattern and taking proactive
measure to avoid maximum demand overshoot the system comprising:
- a reading unit to measure the definite consumption of power in a day;
- a transmitter station comprising local area network (LAN) (7) and
radiofrequency (RF) wireless connectivity configured to receive
prerecorded power consumption, wherein a processor (5) of the said
reading unit monitors the power parameter and input the values to a
neural network, wherein the output of the neural network is a day ahead
presumed load parameter to direct an automatic suitable switching
system to manage power consumption suitably;
- a relay control box (10) to receive control signals either on LAN(7) or an
unlicensed wireless band based on selectively operatable said switching
systems.
2. The system for analyzing industrial load pattern as claimed in claim 1,
wherein the reading unit comprises of a power supply (1), potential
transformer (2), current transformer (3), energy integrator (4), processor
(5) and weather monitoring stator (6).
3. The system for analyzing industrial load pattern as claimed in claim 1
and 2, wherein the processor monitors power parameters which includes
current consumption, prerecorded power and weather data.
4. The system for analyzing industrial load pattern as claimed in claim 1,
as illustrated in the accompanying drawings.
5. The system for analyzing industrial load pattern as claimed in claim 1,
wherein the suitable load switching scheme is selected depending on the
output of the neural network.

6. The system for analyzing industrial load pattern as claimed in claim 1,
wherein the suitable load switching scheme is auto-selected by means of
comparisons of prerecorded date and current consumption.