Residential Load Disaggregation and Demand Management Tool

This dissertation project focuses on developing a software tool for residential load disaggregation and demand management. The objective is to accurately estimate the energy consumption of individual appliances within a household using non-intrusive load monitoring (NILM) techniques. The research explores various methodologies, including regression time and average power algorithms, recurrent neural networks, and disaggregation stage analysis. The project also involves the development of a prototype monitoring system and analysis of customer characteristics. The goal is to improve energy efficiency and provide insights for demand response strategies. The project includes MATLAB coding for recurrent neural networks and time series regression analysis. The project aims to address the limitations of existing methods, such as parameter tuning and generalization challenges, by focusing on iterative disaggregation of appliance consumption patterns. The project also discusses the expected results, limitations, and a detailed project plan.

D I S S E R TAT I O N P R E S E N TAT I O N
A Tool For Residential Load
Disaggregation And Demand
Management.

Contents
Objective
Overview
Research Problem
Desk Analysis
Methodology of Research
Purpose of Research Method
Proposed Research
 Regression Time and Algorithm for Average Power
 Recurrent neural network
 Disaggregation stage analysis
 Prototype for monitoring system in the marketplace
 general characteristic of customers
Expected Result and Limitation
Project Plan
Conclusion
References

Objective
Software tool development for disaggregation in residential aggregate load
and low voltage distribution network load.
Tool invention for automatic estimation of demand response capacity by the
residential loads which is controllable in nature.
Finding approach for iterative disaggregation about the algorithm which is
related to the appliance consumption pattern .
Determining the software tool for the purpose of disaggregation in the field
of residential electric consumption.

Overview
NLM technology, applicable in economics for the purpose to provide the advantage to the
utilities and demand strategies for energy management evolution.
total electrical energy feedback for deriving the energy savings for people by approximately
15%.
approach for contribution in iterative disaggregation about the consumption of appliance
pattern.
The ideology of this research is the combination of Sequence searching for dynamic time
warping which receives the single consumed energy about the power consumption pattern and
the Fuzzy C-means algorithm.
This approach provide the appliance operation’s initial status. [1]

Research Problem
The aim of the research is to making an estimate for each electrical appliances,
which provides the contribution in net energy consumption by a electric meter for
a particular house hold of Non-Intrusive Load Monitoring disaggregation.[2]
Provide the certainty for a particular tool in the field of residential load
disaggregation and demand management at the marketplace.
Due to the excessive energy consumption by houses there is a requirement for the
certainty in the exact measurement of consumed energy and market demand
assess.

Desk Analysis
Energy Disaggregation [5], is an approach to compute the actual predication for individual appliances power
demand in single meter.
There is a requirement for the evaluation of end-users and their call consideration, for the purpose of
aggregation and appropriate measurement.
The existing research method of iterative disaggregation ideology about the appliance consumption pattern,
has some limitations that needs to be overcome in future basis.[3]
Existing techniques of load disaggregation are unsupervised and some of them include the Markov hidden
models. And these methods requires parameter’s manual tuning which is challenging for method
generalization in practical manner.
It used for the observation purpose in multi-label algorithm classification and utilized this algorithm for both
time domain and wavelet domain appliance sets. [3]

Methodology of Research
Research approaches are categorized into two parts [4], and these are:
 Inductive
 Deductive
The Research methodology is the reviewing process of existing approach which offers the procedural and
extensive ideology for the development of residential disaggregation tools in the sequence to come up with
more efficiency.
Demand management elaborates the combination of various research methods.
Another research method is to review the existing approaches in books, articles and use the data in
marketplace at random house for getting the feedback about the tool for residential load disaggregation.
By taking the random interviews for observing the customer behavior in the marketplace and used the data in
the analysis of market demand for electricity.

Purpose of Research Method
The motive of these research methodologies, is to help the
researcher in choosing a way for the research purpose,
which is being described below:
 Description
 Exploration
 Prediction
 Analysis

Proposed Research
Proposed Research describes the various
concerns such as :
 Regression Time and Average Power
Algorithm.
 Recurrent neural network.
 Disaggregation stage analysis.
 Prototype for Monitoring system in the
marketplace.
 Characteristic of customers.

Regression Time and Algorithm for Average
Power
It is the statistical methodology for future prediction, which response depends
upon the response history.
Helps to understand, the prediction behaviour of dynamic system from
observational and experimental data.
Useful in forecasting and modelling of biological, financial and economical
system.
Applying the ordinary least squares to the MLR model, for analysing the time
series through design building matrix (Xt), which includes current and predictors
past observations in order to time (t) can be mathematically derived as:
yt=Xtβ+ut
For getting the estimation of the linear response analysis (yt) with the design
matrix, β shows the estimates of linear parameters which is being computed and
(ut) shows the innovation terms.

Figures for Potential Predictors and response of Time series
Regression

Recurrent Neural Network
Each layer of the recurrent neural network has a recurrent
connection with the tap delay which is connected with it,
unless the recurrent neural network is as same as the feed-
forward network.
This feature allows the network for having the infinite
dynamic responses in time series input data. This network
is also similar with the distributed delay and time delay
network, which is having the capability of finite input
responses.