SlideShare a Scribd company logo
1 of 20
Download to read offline
Probabilistic Forecasting:
How and Why?
By Kostas Hatalis
hatalis@gmail.com
Dept. of Electrical & Computer Engineering
Lehigh University, Bethlehem, PA
2019
Kostas Hatalis Probabilistic Forecasting 2019 1 / 20
Motivation - Application Standpoint
Demand for power, and usage of renewables is both increasing rapidly.
Source: EIA Annual Energy Outlook 2018
Kostas Hatalis Probabilistic Forecasting 2019 2 / 20
Motivation - Application Standpoint
Several challenges can be identified with a higher penetration of
renewables energy into the smart grid:
1) managing variability and uncertainty
2) balancing supply and demand
3) security
All will require the use of forecasting. Applied motivation of my
research.
Kostas Hatalis Probabilistic Forecasting 2019 3 / 20
Motivation - Methodology Standpoint
Goal:develop novel forecasting frameworks that can answer most of
these questions.
1 How do we make robust nonparametric probabilistic forecasts?
2 Can a forecasting model be domain independent?
3 How can nonlinearity and nonstationary be captured in data?
4 Can forecasts be made multi-step?
5 Can the model scale to big data sets?
6 Can we autonomously engineer features?
7 Can we conduct easy model selection (have few parameters)?
Kostas Hatalis Probabilistic Forecasting 2019 4 / 20
Motivation - Methodology Standpoint
Kostas Hatalis Probabilistic Forecasting 2019 5 / 20
Forecasting Setup
Forecasting involves taking models fit on historical data and using them to
predict future observations.
ˆyi
output prediction
=
forecast model
f (xi ) where xi = xi,1, xi,2, ..., xi,k
input features
A forecast model is fit on a set of training samples in the form of (xi , yi )
where xi is the predictor variable, yi is the target variable, and i = 1...N.
Kostas Hatalis Probabilistic Forecasting 2019 6 / 20
Forecasting Questions
When forecasting, it is important to understand your goal:
1 How much data do we have available and are we able to gather it all
together?
2 What is the time horizon of predictions that is required? Short,
medium, or long term?
3 Can forecasts be updated frequently over time or must they be made
once and remain static?
4 At what temporal frequency are forecasts required?
5 Are we forecasting single or multi step predictions?
Kostas Hatalis Probabilistic Forecasting 2019 7 / 20
Point Forecasting
Traditional form of statistical numeric prediction.
The goal is to forecast a single expected value in the future ˆy = E[y|x].
Many methods exist:
Regression: ordinary least squares, ridge, lasso, polynomial, etc.
Time Series: autoregression, ARMA, exponential smoothing, etc.
Note: Time series analysis and regression analysis share many models,
and both are used for forecasting, but they are theoretically different!
Time series methods account for autocorrelation, regression methods
assume iid in features and no serial correlation.
Kostas Hatalis Probabilistic Forecasting 2019 8 / 20
Types of Uncertainty Forecasting
1 Probabilistic Forecasting: predicts future densities. Most popular in
wind power.
2 Risk Indexes: finds the expected level of forecasting error.
3 Scenario Forecasting: creates several point forecasts (scenarios) for
a future period to find spatial-temporal correlations.
Kostas Hatalis Probabilistic Forecasting 2019 9 / 20
Parametric vs Nonparametric
From ”Stochastic predictive control of battery energy storage for wind farm dispatching: Using probabilistic wind power
forecasts.” Renewable Energy (2015).
Parametric - assume a predefined shape of the density.
Nonparametric - we don’t assume any shape.
Kostas Hatalis Probabilistic Forecasting 2019 10 / 20
Types of Forecasts
Types of probabilistic forecasts:
Quantiles: dividing the range of a probability distribution into
intervals.
Prediction Intervals: a range of specified coverage probability under
that distribution.
Predictive Densities: the full distribution.
Kostas Hatalis Probabilistic Forecasting 2019 11 / 20
Quantiles
If we define Ft as the cumulative distribution function (CDF) of the
random variable xt and Ft is a strictly increasing function, the quantile qτ
t
with proportion α ∈ [0, 1] of xt is uniquely defined as the value x such that:
P(Pt < x) = τ
or equivalently as:
qτ
t = F−1
t (τ)
Kostas Hatalis Probabilistic Forecasting 2019 12 / 20
Prediction Intervals
A prediction interval ˆIβ
t+k produced at time t for future time t + k is
defined by its lower and upper bounds, which are the quantile forecasts:
ˆIβ
t+k = ˆqαl
t+k, ˆqαu
t+k
Kostas Hatalis Probabilistic Forecasting 2019 13 / 20
Prediction Density
If predicting quantiles we smooth between them to get the CDF. Else we
can use density estimation techniques.
Kostas Hatalis Probabilistic Forecasting 2019 14 / 20
Forecasting Paradigms
How are the intervals centered?
Centered on the median
Centered on the mean
What type of data do we use for prediction?
Renewable Power
Numerical Weather Forecasts (NWP)
Past Time Series Data
Forecasting done using either current NWP or a combination of
past power and NWP data.
Kostas Hatalis Probabilistic Forecasting 2019 15 / 20
Probabilistic Forecast Evaluation
Reliability Score - measures if PIs cover the observations.
Prediction Interval Coverage Probability (PICP):
PICP =
1
Ntest
Ntest
i=1
ci where ci =
1, ti ∈ Iτ
t (xi )
0, ti /∈ Iτ
t (xi )
Average coverage Error (ACE):
ACE = |PICP − 100 × (1 − τ)|
Kostas Hatalis Probabilistic Forecasting 2019 16 / 20
Probabilistic Forecast Evaluation
Sharpness Score - measures how wide PIs are.
Interval Score (IS):
IS =
1
Ntest
Ntest
i=1
(ˆq1−τ
t − ˆqτ
t )
Skill Score - a rule based score of the density.
Quantile Score (QS):
Q(qτ , y) =
τ
100 − 1 (y − qτ ) if y < qτ
τ
100(y − qτ ) if y ≥ qτ
where y is observed power and qτ is the quantile forecast.
Kostas Hatalis Probabilistic Forecasting 2019 17 / 20
Approaches to Probabilistic Forecasting
Quantile Regression (QR): predict quantiles based on exogenous
inputs.
Kernel Density Estimation (KDE): smooths a histogram of predicted
Gaussian densities.
Machine Learning: use of supervised and unsurprised learning
methods.
Kostas Hatalis Probabilistic Forecasting 2019 18 / 20
Quantile Regression
Uses the pin-ball loss function to apply asymmetric weights to prediction
errors to compute conditional quantiles.
ρτ (u) =
τu if u ≥ 0
(τ − 1)u if u < 0
, 0 < τ < 1 (1)
The pinball function is always positive. The lower the loss, the better the
quantile forecast.
Kostas Hatalis Probabilistic Forecasting 2019 19 / 20
Quantile Regression
Given predictors xi , slope mi and intercept b coefficients in a linear
regression equation we get the conditional τ quantile ˆqτ :
ˆqτ (t) =
k
i=1
mi xi + b (2)
which is be estimated by minimizing the following error cost where y(t) is
the observed value:
Eτ =
1
N
N
t=1
ρτ (y(t) − ˆqτ (t)) (3)
Kostas Hatalis Probabilistic Forecasting 2019 20 / 20

More Related Content

What's hot

Missing data and non response pdf
Missing data and non response pdfMissing data and non response pdf
Missing data and non response pdfAnuj Bhatia
 
Statistics lecture 8 (chapter 7)
Statistics lecture 8 (chapter 7)Statistics lecture 8 (chapter 7)
Statistics lecture 8 (chapter 7)jillmitchell8778
 
Point Estimate, Confidence Interval, Hypotesis tests
Point Estimate, Confidence Interval, Hypotesis testsPoint Estimate, Confidence Interval, Hypotesis tests
Point Estimate, Confidence Interval, Hypotesis testsUniversity of Salerno
 
Statistical inference: Estimation
Statistical inference: EstimationStatistical inference: Estimation
Statistical inference: EstimationParag Shah
 
Lecture: Monte Carlo Methods
Lecture: Monte Carlo MethodsLecture: Monte Carlo Methods
Lecture: Monte Carlo MethodsFrank Kienle
 
Exploratory data analysis
Exploratory data analysisExploratory data analysis
Exploratory data analysisGramener
 
Lesson 5 arima
Lesson 5 arimaLesson 5 arima
Lesson 5 arimaankit_ppt
 
McKinsey Survey: Australian consumer sentiment during the coronavirus crisis
McKinsey Survey: Australian consumer sentiment during the coronavirus crisisMcKinsey Survey: Australian consumer sentiment during the coronavirus crisis
McKinsey Survey: Australian consumer sentiment during the coronavirus crisisMcKinsey on Marketing & Sales
 
Fundamentals Probability 08072009
Fundamentals Probability 08072009Fundamentals Probability 08072009
Fundamentals Probability 08072009Sri Harsha gadiraju
 
Estimation and hypothesis testing 1 (graduate statistics2)
Estimation and hypothesis testing 1 (graduate statistics2)Estimation and hypothesis testing 1 (graduate statistics2)
Estimation and hypothesis testing 1 (graduate statistics2)Harve Abella
 
Lesson 2 stationary_time_series
Lesson 2 stationary_time_seriesLesson 2 stationary_time_series
Lesson 2 stationary_time_seriesankit_ppt
 
Time Series Analysis.pptx
Time Series Analysis.pptxTime Series Analysis.pptx
Time Series Analysis.pptxSunny429247
 
regression assumption by Ammara Aftab
regression assumption by Ammara Aftabregression assumption by Ammara Aftab
regression assumption by Ammara AftabUniversity of Karachi
 

What's hot (20)

Missing data and non response pdf
Missing data and non response pdfMissing data and non response pdf
Missing data and non response pdf
 
Timeseries forecasting
Timeseries forecastingTimeseries forecasting
Timeseries forecasting
 
Time series analysis
Time series analysisTime series analysis
Time series analysis
 
Hypothesis testing
Hypothesis testingHypothesis testing
Hypothesis testing
 
Statistics lecture 8 (chapter 7)
Statistics lecture 8 (chapter 7)Statistics lecture 8 (chapter 7)
Statistics lecture 8 (chapter 7)
 
Point Estimate, Confidence Interval, Hypotesis tests
Point Estimate, Confidence Interval, Hypotesis testsPoint Estimate, Confidence Interval, Hypotesis tests
Point Estimate, Confidence Interval, Hypotesis tests
 
PCA Final.pptx
PCA Final.pptxPCA Final.pptx
PCA Final.pptx
 
Panel data
Panel dataPanel data
Panel data
 
Statistical inference: Estimation
Statistical inference: EstimationStatistical inference: Estimation
Statistical inference: Estimation
 
Lecture: Monte Carlo Methods
Lecture: Monte Carlo MethodsLecture: Monte Carlo Methods
Lecture: Monte Carlo Methods
 
Exploratory data analysis
Exploratory data analysisExploratory data analysis
Exploratory data analysis
 
Lesson 5 arima
Lesson 5 arimaLesson 5 arima
Lesson 5 arima
 
McKinsey Survey: Australian consumer sentiment during the coronavirus crisis
McKinsey Survey: Australian consumer sentiment during the coronavirus crisisMcKinsey Survey: Australian consumer sentiment during the coronavirus crisis
McKinsey Survey: Australian consumer sentiment during the coronavirus crisis
 
Fundamentals Probability 08072009
Fundamentals Probability 08072009Fundamentals Probability 08072009
Fundamentals Probability 08072009
 
Estimation and hypothesis testing 1 (graduate statistics2)
Estimation and hypothesis testing 1 (graduate statistics2)Estimation and hypothesis testing 1 (graduate statistics2)
Estimation and hypothesis testing 1 (graduate statistics2)
 
Stat 3203 -pps sampling
Stat 3203 -pps samplingStat 3203 -pps sampling
Stat 3203 -pps sampling
 
Lesson 2 stationary_time_series
Lesson 2 stationary_time_seriesLesson 2 stationary_time_series
Lesson 2 stationary_time_series
 
Time series
Time seriesTime series
Time series
 
Time Series Analysis.pptx
Time Series Analysis.pptxTime Series Analysis.pptx
Time Series Analysis.pptx
 
regression assumption by Ammara Aftab
regression assumption by Ammara Aftabregression assumption by Ammara Aftab
regression assumption by Ammara Aftab
 

Similar to Probabilistic Forecasting: How and Why?

Developing and validating statistical models for clinical prediction and prog...
Developing and validating statistical models for clinical prediction and prog...Developing and validating statistical models for clinical prediction and prog...
Developing and validating statistical models for clinical prediction and prog...Evangelos Kritsotakis
 
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016Quantopian
 
Estimating Financial Frictions under Learning
Estimating Financial Frictions under LearningEstimating Financial Frictions under Learning
Estimating Financial Frictions under LearningGRAPE
 
Probabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering NetworksProbabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering NetworksTomaso Aste
 
Towards the Validation of National Risk Assessments against Historical Observ...
Towards the Validation of National Risk Assessments against Historical Observ...Towards the Validation of National Risk Assessments against Historical Observ...
Towards the Validation of National Risk Assessments against Historical Observ...Global Risk Forum GRFDavos
 
V. pacáková, d. brebera
V. pacáková, d. breberaV. pacáková, d. brebera
V. pacáková, d. breberalogyalaa
 
Forecasting stock price movement direction by machine learning algorithm
Forecasting stock price movement direction by machine  learning algorithmForecasting stock price movement direction by machine  learning algorithm
Forecasting stock price movement direction by machine learning algorithmIJECEIAES
 
Understanding the effect of Financing Variability Using Chance- Constrained A...
Understanding the effect of Financing Variability Using Chance- Constrained A...Understanding the effect of Financing Variability Using Chance- Constrained A...
Understanding the effect of Financing Variability Using Chance- Constrained A...IRJET Journal
 
Predictive Modeling in Insurance in the context of (possibly) big data
Predictive Modeling in Insurance in the context of (possibly) big dataPredictive Modeling in Insurance in the context of (possibly) big data
Predictive Modeling in Insurance in the context of (possibly) big dataArthur Charpentier
 
Smooth Pinball based Quantile Neural Network
Smooth Pinball based Quantile Neural NetworkSmooth Pinball based Quantile Neural Network
Smooth Pinball based Quantile Neural NetworkKostas Hatalis, PhD
 
Calibration of risk prediction models: decision making with the lights on or ...
Calibration of risk prediction models: decision making with the lights on or ...Calibration of risk prediction models: decision making with the lights on or ...
Calibration of risk prediction models: decision making with the lights on or ...BenVanCalster
 
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...cscpconf
 
FPP 1. Getting started
FPP 1. Getting startedFPP 1. Getting started
FPP 1. Getting startedRob Hyndman
 
Maxentropic and quantitative methods in operational risk modeling
Maxentropic and quantitative methods in operational risk modelingMaxentropic and quantitative methods in operational risk modeling
Maxentropic and quantitative methods in operational risk modelingErika G. G.
 
Application of Exponential Gamma Distribution in Modeling Queuing Data
Application of Exponential Gamma Distribution in Modeling Queuing DataApplication of Exponential Gamma Distribution in Modeling Queuing Data
Application of Exponential Gamma Distribution in Modeling Queuing Dataijtsrd
 
A Maximum Entropy Approach to the Loss Data Aggregation Problem
A Maximum Entropy Approach to the Loss Data Aggregation ProblemA Maximum Entropy Approach to the Loss Data Aggregation Problem
A Maximum Entropy Approach to the Loss Data Aggregation ProblemErika G. G.
 
Naïve Bayes Machine Learning Classification with R Programming: A case study ...
Naïve Bayes Machine Learning Classification with R Programming: A case study ...Naïve Bayes Machine Learning Classification with R Programming: A case study ...
Naïve Bayes Machine Learning Classification with R Programming: A case study ...SubmissionResearchpa
 

Similar to Probabilistic Forecasting: How and Why? (20)

XGBoostLSS - An extension of XGBoost to probabilistic forecasting, Alexander ...
XGBoostLSS - An extension of XGBoost to probabilistic forecasting, Alexander ...XGBoostLSS - An extension of XGBoost to probabilistic forecasting, Alexander ...
XGBoostLSS - An extension of XGBoost to probabilistic forecasting, Alexander ...
 
Developing and validating statistical models for clinical prediction and prog...
Developing and validating statistical models for clinical prediction and prog...Developing and validating statistical models for clinical prediction and prog...
Developing and validating statistical models for clinical prediction and prog...
 
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
Market Timing, Big Data, and Machine Learning by Xiao Qiao at QuantCon 2016
 
Estimating Financial Frictions under Learning
Estimating Financial Frictions under LearningEstimating Financial Frictions under Learning
Estimating Financial Frictions under Learning
 
GDRR Opening Workshop - Gradient Boosting Trees for Spatial Data Prediction ...
GDRR Opening Workshop -  Gradient Boosting Trees for Spatial Data Prediction ...GDRR Opening Workshop -  Gradient Boosting Trees for Spatial Data Prediction ...
GDRR Opening Workshop - Gradient Boosting Trees for Spatial Data Prediction ...
 
Probabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering NetworksProbabilistic Modelling with Information Filtering Networks
Probabilistic Modelling with Information Filtering Networks
 
Towards the Validation of National Risk Assessments against Historical Observ...
Towards the Validation of National Risk Assessments against Historical Observ...Towards the Validation of National Risk Assessments against Historical Observ...
Towards the Validation of National Risk Assessments against Historical Observ...
 
V. pacáková, d. brebera
V. pacáková, d. breberaV. pacáková, d. brebera
V. pacáková, d. brebera
 
Forecasting stock price movement direction by machine learning algorithm
Forecasting stock price movement direction by machine  learning algorithmForecasting stock price movement direction by machine  learning algorithm
Forecasting stock price movement direction by machine learning algorithm
 
Understanding the effect of Financing Variability Using Chance- Constrained A...
Understanding the effect of Financing Variability Using Chance- Constrained A...Understanding the effect of Financing Variability Using Chance- Constrained A...
Understanding the effect of Financing Variability Using Chance- Constrained A...
 
Predictive Modeling in Insurance in the context of (possibly) big data
Predictive Modeling in Insurance in the context of (possibly) big dataPredictive Modeling in Insurance in the context of (possibly) big data
Predictive Modeling in Insurance in the context of (possibly) big data
 
Smooth Pinball based Quantile Neural Network
Smooth Pinball based Quantile Neural NetworkSmooth Pinball based Quantile Neural Network
Smooth Pinball based Quantile Neural Network
 
Data science
Data scienceData science
Data science
 
Calibration of risk prediction models: decision making with the lights on or ...
Calibration of risk prediction models: decision making with the lights on or ...Calibration of risk prediction models: decision making with the lights on or ...
Calibration of risk prediction models: decision making with the lights on or ...
 
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...
ENTROPY-COST RATIO MAXIMIZATION MODEL FOR EFFICIENT STOCK PORTFOLIO SELECTION...
 
FPP 1. Getting started
FPP 1. Getting startedFPP 1. Getting started
FPP 1. Getting started
 
Maxentropic and quantitative methods in operational risk modeling
Maxentropic and quantitative methods in operational risk modelingMaxentropic and quantitative methods in operational risk modeling
Maxentropic and quantitative methods in operational risk modeling
 
Application of Exponential Gamma Distribution in Modeling Queuing Data
Application of Exponential Gamma Distribution in Modeling Queuing DataApplication of Exponential Gamma Distribution in Modeling Queuing Data
Application of Exponential Gamma Distribution in Modeling Queuing Data
 
A Maximum Entropy Approach to the Loss Data Aggregation Problem
A Maximum Entropy Approach to the Loss Data Aggregation ProblemA Maximum Entropy Approach to the Loss Data Aggregation Problem
A Maximum Entropy Approach to the Loss Data Aggregation Problem
 
Naïve Bayes Machine Learning Classification with R Programming: A case study ...
Naïve Bayes Machine Learning Classification with R Programming: A case study ...Naïve Bayes Machine Learning Classification with R Programming: A case study ...
Naïve Bayes Machine Learning Classification with R Programming: A case study ...
 

Recently uploaded

Determinants of health, dimensions of health, positive health and spectrum of...
Determinants of health, dimensions of health, positive health and spectrum of...Determinants of health, dimensions of health, positive health and spectrum of...
Determinants of health, dimensions of health, positive health and spectrum of...shambhavirathore45
 
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779Best VIP Call Girls Noida Sector 39 Call Me: 8448380779
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779Delhi Call girls
 
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779Best VIP Call Girls Noida Sector 22 Call Me: 8448380779
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779Delhi Call girls
 
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptx
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptxBPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptx
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptxMohammedJunaid861692
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130Suhani Kapoor
 
VidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxVidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxolyaivanovalion
 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz1
 
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Online
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service OnlineCALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Online
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Onlineanilsa9823
 
April 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's AnalysisApril 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's Analysismanisha194592
 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxolyaivanovalion
 
BigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptxBigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptxolyaivanovalion
 
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...Delhi Call girls
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfLars Albertsson
 
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Callshivangimorya083
 
Carero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptxCarero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptxolyaivanovalion
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Callshivangimorya083
 
Generative AI on Enterprise Cloud with NiFi and Milvus
Generative AI on Enterprise Cloud with NiFi and MilvusGenerative AI on Enterprise Cloud with NiFi and Milvus
Generative AI on Enterprise Cloud with NiFi and MilvusTimothy Spann
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...amitlee9823
 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptxAnupama Kate
 

Recently uploaded (20)

Determinants of health, dimensions of health, positive health and spectrum of...
Determinants of health, dimensions of health, positive health and spectrum of...Determinants of health, dimensions of health, positive health and spectrum of...
Determinants of health, dimensions of health, positive health and spectrum of...
 
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779Best VIP Call Girls Noida Sector 39 Call Me: 8448380779
Best VIP Call Girls Noida Sector 39 Call Me: 8448380779
 
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779Best VIP Call Girls Noida Sector 22 Call Me: 8448380779
Best VIP Call Girls Noida Sector 22 Call Me: 8448380779
 
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptx
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptxBPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptx
BPAC WITH UFSBI GENERAL PRESENTATION 18_05_2017-1.pptx
 
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
VIP Call Girls Service Miyapur Hyderabad Call +91-8250192130
 
VidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptxVidaXL dropshipping via API with DroFx.pptx
VidaXL dropshipping via API with DroFx.pptx
 
Abortion pills in Doha Qatar (+966572737505 ! Get Cytotec
Abortion pills in Doha Qatar (+966572737505 ! Get CytotecAbortion pills in Doha Qatar (+966572737505 ! Get Cytotec
Abortion pills in Doha Qatar (+966572737505 ! Get Cytotec
 
Invezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signalsInvezz.com - Grow your wealth with trading signals
Invezz.com - Grow your wealth with trading signals
 
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Online
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service OnlineCALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Online
CALL ON ➥8923113531 🔝Call Girls Chinhat Lucknow best sexual service Online
 
April 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's AnalysisApril 2024 - Crypto Market Report's Analysis
April 2024 - Crypto Market Report's Analysis
 
CebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptxCebaBaby dropshipping via API with DroFX.pptx
CebaBaby dropshipping via API with DroFX.pptx
 
BigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptxBigBuy dropshipping via API with DroFx.pptx
BigBuy dropshipping via API with DroFx.pptx
 
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
Call Girls in Sarai Kale Khan Delhi 💯 Call Us 🔝9205541914 🔝( Delhi) Escorts S...
 
Schema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdfSchema on read is obsolete. Welcome metaprogramming..pdf
Schema on read is obsolete. Welcome metaprogramming..pdf
 
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
꧁❤ Greater Noida Call Girls Delhi ❤꧂ 9711199171 ☎️ Hard And Sexy Vip Call
 
Carero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptxCarero dropshipping via API with DroFx.pptx
Carero dropshipping via API with DroFx.pptx
 
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip CallDelhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
Delhi Call Girls Punjabi Bagh 9711199171 ☎✔👌✔ Whatsapp Hard And Sexy Vip Call
 
Generative AI on Enterprise Cloud with NiFi and Milvus
Generative AI on Enterprise Cloud with NiFi and MilvusGenerative AI on Enterprise Cloud with NiFi and Milvus
Generative AI on Enterprise Cloud with NiFi and Milvus
 
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
Chintamani Call Girls: 🍓 7737669865 🍓 High Profile Model Escorts | Bangalore ...
 
100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx100-Concepts-of-AI by Anupama Kate .pptx
100-Concepts-of-AI by Anupama Kate .pptx
 

Probabilistic Forecasting: How and Why?

  • 1. Probabilistic Forecasting: How and Why? By Kostas Hatalis hatalis@gmail.com Dept. of Electrical & Computer Engineering Lehigh University, Bethlehem, PA 2019 Kostas Hatalis Probabilistic Forecasting 2019 1 / 20
  • 2. Motivation - Application Standpoint Demand for power, and usage of renewables is both increasing rapidly. Source: EIA Annual Energy Outlook 2018 Kostas Hatalis Probabilistic Forecasting 2019 2 / 20
  • 3. Motivation - Application Standpoint Several challenges can be identified with a higher penetration of renewables energy into the smart grid: 1) managing variability and uncertainty 2) balancing supply and demand 3) security All will require the use of forecasting. Applied motivation of my research. Kostas Hatalis Probabilistic Forecasting 2019 3 / 20
  • 4. Motivation - Methodology Standpoint Goal:develop novel forecasting frameworks that can answer most of these questions. 1 How do we make robust nonparametric probabilistic forecasts? 2 Can a forecasting model be domain independent? 3 How can nonlinearity and nonstationary be captured in data? 4 Can forecasts be made multi-step? 5 Can the model scale to big data sets? 6 Can we autonomously engineer features? 7 Can we conduct easy model selection (have few parameters)? Kostas Hatalis Probabilistic Forecasting 2019 4 / 20
  • 5. Motivation - Methodology Standpoint Kostas Hatalis Probabilistic Forecasting 2019 5 / 20
  • 6. Forecasting Setup Forecasting involves taking models fit on historical data and using them to predict future observations. ˆyi output prediction = forecast model f (xi ) where xi = xi,1, xi,2, ..., xi,k input features A forecast model is fit on a set of training samples in the form of (xi , yi ) where xi is the predictor variable, yi is the target variable, and i = 1...N. Kostas Hatalis Probabilistic Forecasting 2019 6 / 20
  • 7. Forecasting Questions When forecasting, it is important to understand your goal: 1 How much data do we have available and are we able to gather it all together? 2 What is the time horizon of predictions that is required? Short, medium, or long term? 3 Can forecasts be updated frequently over time or must they be made once and remain static? 4 At what temporal frequency are forecasts required? 5 Are we forecasting single or multi step predictions? Kostas Hatalis Probabilistic Forecasting 2019 7 / 20
  • 8. Point Forecasting Traditional form of statistical numeric prediction. The goal is to forecast a single expected value in the future ˆy = E[y|x]. Many methods exist: Regression: ordinary least squares, ridge, lasso, polynomial, etc. Time Series: autoregression, ARMA, exponential smoothing, etc. Note: Time series analysis and regression analysis share many models, and both are used for forecasting, but they are theoretically different! Time series methods account for autocorrelation, regression methods assume iid in features and no serial correlation. Kostas Hatalis Probabilistic Forecasting 2019 8 / 20
  • 9. Types of Uncertainty Forecasting 1 Probabilistic Forecasting: predicts future densities. Most popular in wind power. 2 Risk Indexes: finds the expected level of forecasting error. 3 Scenario Forecasting: creates several point forecasts (scenarios) for a future period to find spatial-temporal correlations. Kostas Hatalis Probabilistic Forecasting 2019 9 / 20
  • 10. Parametric vs Nonparametric From ”Stochastic predictive control of battery energy storage for wind farm dispatching: Using probabilistic wind power forecasts.” Renewable Energy (2015). Parametric - assume a predefined shape of the density. Nonparametric - we don’t assume any shape. Kostas Hatalis Probabilistic Forecasting 2019 10 / 20
  • 11. Types of Forecasts Types of probabilistic forecasts: Quantiles: dividing the range of a probability distribution into intervals. Prediction Intervals: a range of specified coverage probability under that distribution. Predictive Densities: the full distribution. Kostas Hatalis Probabilistic Forecasting 2019 11 / 20
  • 12. Quantiles If we define Ft as the cumulative distribution function (CDF) of the random variable xt and Ft is a strictly increasing function, the quantile qτ t with proportion α ∈ [0, 1] of xt is uniquely defined as the value x such that: P(Pt < x) = τ or equivalently as: qτ t = F−1 t (τ) Kostas Hatalis Probabilistic Forecasting 2019 12 / 20
  • 13. Prediction Intervals A prediction interval ˆIβ t+k produced at time t for future time t + k is defined by its lower and upper bounds, which are the quantile forecasts: ˆIβ t+k = ˆqαl t+k, ˆqαu t+k Kostas Hatalis Probabilistic Forecasting 2019 13 / 20
  • 14. Prediction Density If predicting quantiles we smooth between them to get the CDF. Else we can use density estimation techniques. Kostas Hatalis Probabilistic Forecasting 2019 14 / 20
  • 15. Forecasting Paradigms How are the intervals centered? Centered on the median Centered on the mean What type of data do we use for prediction? Renewable Power Numerical Weather Forecasts (NWP) Past Time Series Data Forecasting done using either current NWP or a combination of past power and NWP data. Kostas Hatalis Probabilistic Forecasting 2019 15 / 20
  • 16. Probabilistic Forecast Evaluation Reliability Score - measures if PIs cover the observations. Prediction Interval Coverage Probability (PICP): PICP = 1 Ntest Ntest i=1 ci where ci = 1, ti ∈ Iτ t (xi ) 0, ti /∈ Iτ t (xi ) Average coverage Error (ACE): ACE = |PICP − 100 × (1 − τ)| Kostas Hatalis Probabilistic Forecasting 2019 16 / 20
  • 17. Probabilistic Forecast Evaluation Sharpness Score - measures how wide PIs are. Interval Score (IS): IS = 1 Ntest Ntest i=1 (ˆq1−τ t − ˆqτ t ) Skill Score - a rule based score of the density. Quantile Score (QS): Q(qτ , y) = τ 100 − 1 (y − qτ ) if y < qτ τ 100(y − qτ ) if y ≥ qτ where y is observed power and qτ is the quantile forecast. Kostas Hatalis Probabilistic Forecasting 2019 17 / 20
  • 18. Approaches to Probabilistic Forecasting Quantile Regression (QR): predict quantiles based on exogenous inputs. Kernel Density Estimation (KDE): smooths a histogram of predicted Gaussian densities. Machine Learning: use of supervised and unsurprised learning methods. Kostas Hatalis Probabilistic Forecasting 2019 18 / 20
  • 19. Quantile Regression Uses the pin-ball loss function to apply asymmetric weights to prediction errors to compute conditional quantiles. ρτ (u) = τu if u ≥ 0 (τ − 1)u if u < 0 , 0 < τ < 1 (1) The pinball function is always positive. The lower the loss, the better the quantile forecast. Kostas Hatalis Probabilistic Forecasting 2019 19 / 20
  • 20. Quantile Regression Given predictors xi , slope mi and intercept b coefficients in a linear regression equation we get the conditional τ quantile ˆqτ : ˆqτ (t) = k i=1 mi xi + b (2) which is be estimated by minimizing the following error cost where y(t) is the observed value: Eτ = 1 N N t=1 ρτ (y(t) − ˆqτ (t)) (3) Kostas Hatalis Probabilistic Forecasting 2019 20 / 20