Unit IV: Modelling and Prediction of

performance

Dr. Krishna K. Warhade

Ph.D. ( IIT Bombay)
Dean Research and Head Ph.D. Research Center
Dr. Vishwanath Karad MIT World Peace University,
Pune
Mathematical modelling
 Representation of real world problems into
mathematical form with some simplified assumptions
which helps to understand fundamental and
quantitative way
 Have widespread application in Engineering and
Technology, Medicine, Biology, Nuclear science,
defense etc.
 It is used to perform experiments and solve real
world problems which may be risky, expensive and
time consuming
Mathematical modelling
 Emerged as powerful tool for studying variety of
research problems in scientific research, product,
process development and manufacturing.
 It improves the quality of work, and reduce changes
errors and rework
 Solve the real world problems and has widely used
due to increasing computation power and computing
methods
 Used to handle large scale and complicated
problems
Mathematical modelling
Mathematical models are highly used in following
areas
 Climate modelling, Aerospace, Space technologies,

 Manufacturing and design, Seismology, Environment,

 Economics, Material research, drug design,

population dynamics
 Medicine, Biology, Atomic research
Mathematical modelling
 Types of mathematical models
 Empirical models
(i) Experiments
(ii) Observations
❑ Theoretical models
(i) Statistical
(ii) Mathematical
(iii) Computational
Mathematical modelling:
Mathematical process
 Real world problems: Simplify
 Working model: Represent
 Mathematical model: Translate
 Computational model: Simulate
 Results and conclusion : Interpret
Mathematical modelling: Types of
Models
 Static or dynamic
 Discrete or continuous
 Deterministic or probabilistic
 Linear or nonlinear
 Explicit or Implicit
 Qualitative or Quantitative
Mathematical modelling: Types of
Models
Static or dynamic
 Static model : A static or steady state model
calculates the system in equilibrium and thus is time
invariant. A static model can not be changed.
 Dynamic model: Dynamic model accounts for time
dependent changes in the state of the system.
Dynamic models are represented by differential
Equations.
Mathematical modelling: Types of
Models
Discrete model or continuous model
 Discrete model : It treats objects as discrete such
as the particles in molecular models
 Continuous model: It represents the objects in
continuous manners such as the velocity field of fluid
in pipe or channels, temperatures and electric field
Mathematical modelling: Types of
Models
Deterministic model or Probabilistic model
 Deterministic model : A deterministic model is one
in which every set of variable state is uniquely
determine by parameters in the model and by sets
of previous state of these variables. It describe
behavior on the basic of some physical law
 Probabilistic model: is one where exact prediction is
not possible, randomness is present and variable
sets are not represented by unique values, but
rather by probabilistic distribution.
Mathematical modelling: Types of
Models
Linear model or nonlinear model
 Linear model : If all the operations in
mathematical model exhibit linearity, the resulting
mathematical model is defines as linear. A linear
model uses parameters that are constant and do not
vary throughout simulations.
 Nonlinear model: Nonlinear model introduces
dependent parameters that are allowed to vary
throughout the course of simulation run, and its use
become necessary where interdependencies
between parameters can not be considered.
Mathematical modelling: Types of
Models
Explicit model or Implicit model
 Explicit model : Calculate the state of the system
using past time from the state of the system using
current time.
 Implicit model: Solution is obtained by solving an
equations involving both the current state of the
system and the later one which requires extra
commutations and could be harder to solve.
Mathematical modelling: Types of
Models
Qualitative model or Quantitative model
 Qualitative model : It is basically a conceptual
model that display visually important components of
an eco system and linkages between them. It is a
simplification of complex system
 Quantitative model: These models are
mathematically focused and many times are based
on complex formulas. In addition quantitative are
generally represent through input-output matrix. It
gives precise numerical answers.
Concept of Mathematical modelling
Real time function and methodology converted into
the language of mathematics is called models.
Execution stages of research methodology
Concept of Mathematical modelling
System Engineering cycle
Concept of Mathematical modelling
Stages of modeling
Stages of Modeling
 Building stage: It comprises of system analysis on
the basis of assumptions and flow of the model.
Choosing a suitable mathematical expression with
the analogy of physical and data exploration is
also covered in the building stage
 Studying model is responsible for realizing system
behavior mostly qualitative behavior.
Transformation into dimensionless quantities,
analyzing asymptotic behavior and sensitivity are
some of the stages of studying model.
Stages of Modeling
Verification and validation of system are covered by
the testing model. Testing of the system depends on
 Assumption linearly and deterministic behavior

 Model structure

 Prediction and previously unused data

 Estimating model parameters

 Using model will facilitate user to predict data and

outcomes
Finally, the mathematical model will be described
and tested for functional behavior of the system
Process of modelling
Process of modeling
 Functional requirements will be studies to
understand observations required for system
feasible observations then forwarded to informal
model
 Informal model is set at possible parameters and
methods that can explain the selected observations.
These methods and parameters are tested for given
system to satisfy required accuracy and precision.
Process of modeling
 Formal model converts assumptions in the informal
model to mathematical form. The formal model
translate text based informal model in the
mathematical model
 A mathematical function decides exact methodology
such as logic algebra, geometry, probability theory
and computer-aided tools.
Advantages of mathematical modelling

 It decides various required characteristics of the

system and defining non-feasible characteristics
also
 It helps in designing hypothesis of the system and
unknown and important parameters are also
derived from it
 The mathematical model optimizes the system to fill
the gap between knowledge and requirements with
less complexity.
Advantages of mathematical modelling

 The mathematical model thus can give a realistic

representation of system using statistic and
probable outcomes
 It is systemic, iterative and can be redefined
 It predicts behavior at a system and can help to
validate methodology in a physical process.
Disadvantages of mathematical
modelling
 It generates multiple outcomes of system so the
selection of right outcomes is important.
 As it is iterative, consumes more time than traditional
methods.
Example of mathematics modelling
Integrated solar water heater
 Specifications of the system:

Data logger for Analog to Digital converter with 16

bit 0.1 degree Celsius resolution
Thermal flask
Theromometer and thermocouples
Software for simulation purpose, Dynamometer to
measure horizontal global solar radiations
Pressure sensors
Modeling solar water heater using
macro model- Thermal model
 Prediction of performance is a challenging task
 Define a model able to predict water temperature
depending on orientation and location
 Select analytical model to predict a performance
using thermal network analysis
Macro model
Macro model defines following parameters
 Conversion from air cavity and water cavity

 Converted from cover to environment and box to

environment
 Radiations from place to cover and cover to sky

 Conduction through the box

 Head capacitance of the system

Computing model to predict
performance of system
 Computing model of performance prediction is
called performance model of the system.
 The performance modelling is a method of
understanding performance measurements and
analysis by converting characteristic of the system
into mathematical form.
Factors included in performance model

 System size, system architecture

 Processor speed. Multilevel cache latency and BW

 Interprocessor network latency and BW

 System software efficiency

 Types of applications

 Algorithms and paging language

 Problem size

 No. of I/O’s

Performance model can be defined to achieve

accuracy, speed, latency and efficiency.
Performance measurement analysis and
performance curve

 A model designed for runtime performance

prediction with less computations of hardware
efficient system is called performance model.
 It includes steps
(1) Assumptions
(2) Parameters
(3) Equations
Performance measurement analysis
and performance curve
 Example : Data-intensive applications designing
Assumptions considered here are
Computation time is negligible as compared to I/O
speed
For workload optimization pipeline structure allows
instructions to run parallel
Input data is evenly distributed
Deriving the model from I/O operation
I/O operations are mapped with equations that
depends on available data.
Performance measurement and
analysis
 Performance measurement is an analysis of overall
system including prevention and detection.
Performance measurement process flow is as follows
Performance curve
 Various research domains concerned with
classifications task uses receivers' operation
characteristics curve to assess the quality of model
or to compare two or more models with respect to
the operating point.
 Researchers also provide some statistics from ROC
such as break even points or equal error rate.
Instead of ROC curve, performance curve is
preferable in machine learning context.
Performance parameters
 Performance parameters normally observed for
classification and some of the application specific
parameters are
 False acceptance rate (FAR) = FP/(FN+TP)
 False rejection rate (FRR) = FN/(FN+TP)
 Half total error rate (HTER) = FAR/2 + FRR/2
 Precision ( P)=TP/(TP+FP)
 Recall ( R )= TP/(TP+FN)
 F1 score= 2*P*R/(P+R)
Multi-scale modeling
 It is the field of solving problems which have important
features at multiple scales of time or space
 Multiscale modeling predicts system behavior with different
models such as quantum mechanical models, molecular dynamic
model, mesoscales, continuous model and devise model
 It allows predicting system behavior based on atomistic
structure and property of elementary processes in material
Engg.
Multi-scale modeling
Approaches to analyze multiscale structures
 The physical analysis used for spatio-temporal structures with
averaging methods, discrete method, and multiscale method.
 Averaging method uses lumped parameters for certain spatio-
temporal volume by assuming the system to be uniform all over
the volume
 Discrete methods most commonly used in computational
mechanics and chemistry.
 Multiscale method consider the disparity and interactions on
different scales
 It identifies the prevailing behavior of complex system.
Multiscale modeling
 There are different multiscale methodologies such as
descriptive, correlative and analytical
 The descriptive multiscale describes the appearance of various
structures on different scales without analyzing their
performance
 The correlative multiscale methodology explains the behavior
on higher scales by analyzing the mechanism on the next lower
scale
 The analytical multiscale methodology reveals the relationship
between scales by performing stability criterion dominant
mechanism and compromise between sub mechanisms.
Sensitivity analysis
 The basic objective of sensitivity analysis is to measure the
effects of parameter variations on output response.
 The local sensitivity analysis is related to analysis of output
response with respect to selected parameters, point of analysis
and state variables.
 Global sensitivity analysis will analyze critical points in the
mutual phase space formed by the parameters, state variables
and adjoint variables and subsequently analyze these critical
points by local sensitivity analysis.
 The extent of sensitivity analysis is to analyze sensitivities of
the models based on estimated parameters around their
normal values efficiently
Sensitivity analysis
 Sensitivity analysis is involved in identifying ranges of input
data for which output of linear model to remain unchanged
 In statistics sensitivity is one of tool to measure the accuracy of
an instrument for classification. Sensitivity is defined as the
probability of correctly identifying some condition.
 There are two procedures adopted by Statistics
Forward sensitivity analysis procedure (FSAP)
Adjoint sensitivity analysis procedure (ASAP)
Sensitivity analysis
 Forward sensitivity analysis procedure is a direct method of
analysis specifically used where the number of response of
interest exceeds the number of system parameters variations to
be considered
 Adjoint sensitivity analysis procedure is an advanced method
of analysis applicable in the problem where practical interest
comprises a large number of parameters and comparatively
few responses
 Thus sensitivity can be considered as formal method for
evaluating data and models because they are associated with
the computations of specific parameters.