Predictive Simulation & Big Data Analytics ISD Analytics

(1)

Predictive Simulation

& Big Data Analytics

(2)

Overview

Simulation can play a vital role in the emerging $billion field of Big Data

analytics to support Government policy and business strategy decisions

Overview

 How simulation plays a key part in the Big Data Predictive Analytics process

 Introduce Simulait simulation-based consumer analytics platform

 Case studies: water, energy, emergence response, retail, transport

 Simulait Online – simulation in the cloud for on-demand access and large scale

(3)

Data Analytics & Decision Process

Past Future Observe Descriptive Analytics Predict Predictive Analytics Influence Prescriptive Analytics Business Questions: What happened? Business Questions:

What is likely to happen?

Business Questions:

What should I do about it?

What happened? Why did it happen? What is happening? Why is it happening?

What is likely to happen? What should I do about it? How do I influence the future? What are the consequences?

Solutions:

Data mining & forensics Real-time analytics & mining Market segmentation

Solutions:

Simulation

Statistics & linear regression Predictive data-mining

Forecasting & trend reporting

Solutions:

Simulation

(4)

Projection vs Prediction

Traditional statistical approaches project future behaviour by extrapolating

past behaviour

 Observe and forecast what people do but not “why” they do it

 Unable to effectively represent complex consumer behavior

 Limited functionality – unable to address a broad range of business problems

 Past demand is not always a good predictor of the future

10 000

1000

100

To

ta

lS

al

es Changing population

& consumer trends

Influence future sales by testing strategies with Simulait

(5)

SimulAIt – An Analogy

SimulAIt is a real life SimCity application where businesses or Government can

predict

and test strategies to

influence

the behaviour of large populations

 Diverse domains: water, energy, emergency response, retail, transport, ...

 Diverse applications: strategy, policy, pricing, demand forecasting, marketing,

community behaviour and social planning, new product uptake, etc....

 Global applicability: Australia, Europe, USA

(6)

Simulait: A Truly Predictive Approach

 Accurate: proven approach, demonstrated over 95% accuracy

 Model not built on past demand data – demand data used to validate the model

 Accuracy due to greater representation of a broad range of consumer factors

(7)

(8)

Case Study 1: Victorian Water Utilities

Objectives

 Isolate and quantify the effectiveness of past water conservation strategies – economic,

regulatory, social (communications) & environmental

 Forecast bounce-back in water demand from easing restrictions & price increases

 Assess impact of product uptake on demand and revenue

 Build a business case to industry regulators – pricing review

 Build demographic demand profiles

 Blind validation: Used 4 yrs of demand data to calibrate outdoor water use and then forecast next 6 years of demand without access to actual demand data

(9)

25 30 35 W at er co ns um pt io n

Average monthly household water consumption

Case Study 1: Victorian Water Utilities

Blind validation results

0 5 10 15 20

Jul-00 Jul-01 Jul-02 Jul-03 Jul-04 Jul-05 Jul-06 Jul-07 Jul-08 Jul-09 Jul-10

W at er co ns um pt io n Simulated Actual-calibration data Actual - blind validation data

(10)

Case Study 1: Victorian Water Utilities

Key outcomes and benefits

 Informed capital expenditure, corporate plans, water restriction schedules

 Rigorous business case to industry regulators to maximise product price and

revenue

 Isolated and quantified the effectiveness of past & future strategies (campaign

analysis)

(11)

Case Study 2: Water in USA & France

Key outcomes and benefits

 Model transferable to different countries

 Better for long term forecasting – tendering, strategic & financial planning, design

future cities, etc...

 Support water conservation, regulation, new water rates, impact of recession, etc...

Calibration

>90% Accuracy

Calibration point

(12)

Case Study 3: Rebates/Retail

Objective

 Identify a mix of products and prices for the water rebates program that maximises

efficiency and keeps within the program budget

 Three projects, and now a 3 year license to 2015

Approach

 Simulated 2 million households, 4.5 million consumers

 Incorporated consumer preference and affordability, and product age, failure and

price

 Simulated product uptake and efficiency with different prices

Key outcomes and benefits

 Accurate predictions of product up-take and budget spend

 Prevented budgets blow-outs

 Cost/benefit (triple bottom line) analysis of different strategies

 Forecast the ROI of different demographics and regions, and to assist with targeted

(13)

Case Study 4: Energy

Customer Personalization

Energy load forecasting accuracy

Total Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

2008 99.0% 99.2% 97.9% 98.8% 98.0% 95.0% 98.5% 99.6% 97.0% 99.6% 98.7% 96.5% 85.0%

2009 99.8% 96.7% 99.3% 99.3% 99.0% 98.9% 98.4% 98.8% 95.1% 97.3% 93.1% 98.6% 98.3%

Using 1% of CRM data in the first 6 months, Simulait was able to accurately

predict what each specific customer will do, and why, for the next 2 years!!!

2009 99.8% 96.7% 99.3% 99.3% 99.0% 98.9% 98.4% 98.8% 95.1% 97.3% 93.1% 98.6% 98.3%

2010 98.3% 91.9% 97.9% 97.1% 97.6% 98.6% 98.1% 99.1% 97.1% 87.8%

250 300 350 400

Actual Forecast

Calibration Prediction

(14)

Case Study 5: Energy - EV Uptake & Transport

Objective

 Predict the uptake of Electric Vehicles over time to 2040

 Predict usage and charging behaviour of electric vehicles

 Impact on the electricity network (extra peak load) to support reliability and quality

(15)

Case Study 5: Energy - EV Uptake & Transport

Approach



EV Uptake consumer decision model

 Simulated the new and used vehicle market across Australia

 Considers many dynamic factors: consumer type, petrol and elec price, car

range, charge times, charge infrastructure, upfront price, ongoing costs, dwelling suitability, battery replacement, depreciation, market penetration, etc...



EV usage: transport/activity model

 Model each consumer’s daily activities and transport/vehicle use

 Factors include: consumer type (e.g. occupation, family structure), day of week,

number of vehicles in the household, activity types (work, school, shopping, entertainment, family/social visits, etc...)

 Other factors: passenger trips, infant trips to carers if both parents working,

separate household activities for independents, vacation from work (e.g. for parents during school holidays), etc...

(16)

Case Study 6: Emergency response - bushfire

Following the 2009 bushfires that claimed 173 lives, the Victorian Royal Commission identified that

“...strategies must reflect how people actually behave... Timely and accurate warnings can provide triggers, but the content and delivery must be carefully developed to elicit the right response”

(17)

Case Study 6: Emergency response - bushfire

Objective



Model community behaviour to bushfires and warnings to support bushfire

strategy and policy, and ultimately save lives



The model predicts:

 What people do and when: Stay, leave or “wait and see”

(18)

Case Study 6: Emergency response - bushfire

Approach



Based on a

health model

of behaviour change – individual’s life is at risk

 Potential to be applied to support health policy and manage the unsustainably

increasing health costs



Given where people are, who they are, what they are observing, the

warnings they are receiving (and which mediums, e.g. radio, text, etc.), and

the progression of the bushfire, we determine the level of threat,

vulnerability and uncertainty for each individual/family, and thus response

Wait Perceived Threat Level of motivation to act Wait Wait Wait Leave Stay Decision

(19)

Case Study 6: Emergency response - bushfire

Validation & outcomes



Applied the model to two bushfires in Victoria and South Australia and

demonstrated

>90% accuracy



Currently assessing hypothetical bushfire scenarios to support bushfire

policy and strategy



Can be applied in emergency response situations beyond bushfires...

(20)

SimulAIt Online (SOL)



Access SimulAIt via a web browser



SimulAIt Online allows:

 Access validated models online

 Add many users

 Create multiple scenarios – test assumptions and what-if analysis

 Share scenarios (models), results, notes and descriptions

 Refresh data and configure assumptions, parameters, etc...

 Run simulations

 Download results – disaggregated via region and time or other factor

 Benefits

 On-demand access to models, for technical and non-technical users

 Control, visibility, ease of use

 Facilitates collaboration and consistency: share scenarios and results

 Maximise ROI: execute many scenarios when required

(21)

Case Study 7: Vic ESC & Retailers

Challenge



Limited availability of suitable data and forecasting models presents a

challenge for regional water retailers to produce accurate forecasts for their

pricing review

Approach



Team members collaboratively used SOL to create validated models with

minimal data

Key outcomes and benefits

(22)

Summary



Simulation can add significant value to support strategic decision making

and policy for Government and business



Unique and important role to play in “Big Data”



Provide the “right” information to make better decisions:

predict

and

how to

influence



Simulait is a practical approach for problems involving consumers and

populations: i.e. human behaviour



High level of accuracy and functionality



Demonstrated in various domains and countries with minimal

configuration



Simulait Online web/cloud based solution provides on demand access for

users globally



Collaborative tool: access, share, run scenarios, and download results



Access to “limitless” computing power to run large scale scenarios

(23)

ISD Analytics

Questions?

ISD Analytics

27 Chesser Street,

Adelaide, South Australia, 5000

Phone: +61 8 7200 3589

[email protected]

(24)

SOL Technical Overview

SimulAIt Online (SOL)

Web User Interface

Users

•Configure scenarios

•View/compare results

SimulAIt Hosting Centre

Scenarios & Results

CPU On Demand

Internet SOL Server Application Dynamic Multi-Dimensional

Database _{Micro-Simulation}SimulAIt _Models

Engine Population Dynamics Models Models Domain Specific Models (water, energy, retail, finance, ...)

SimulAIt Platform and Models (SPM)

Census Data

New/Updated Models

Data utilised:

•Market research &

social data

•Econometric &

statistical data

•Engineering and

environmental data

•Customer data

& Results

Rules, behaviors, logic, reasoning, ...

(25)

Main SOL screen

Scenario groups

Model type User &_logout

Scenario menu items Admin menu items

Session message Session messages

groups Scenarios

Working pane

message log

Active scenario

(26)

Scenario: Configuration

Parameters tree:

hierarchical Config input type

Time associated with parameter values

hierarchical to reduce complexity

Time explicit parameter values (cells) Slide to increase working pane

(27)

Run scenario – SimulAIt!

Start simulating the scenario Set the scenario start and end times

(28)

Scenario: Results

Range of results to download:

Water, energy, carbon, revenue, etc.

Monthly, yearly

Disaggregated into different regions, appliances

(29)

(30)

(31)