Core Concepts and Methods in Load Forecasting: With Applications in Distribution Networks

Foreword
Preface
Acknowledgments
Contents
1 Introduction
1.1 Motivation
1.2 Demand Forecasting for LV Systems
1.3 Why Do We Need This Book
1.4 Aims and Overview of This Book
1.5 How to Read This Book
1.6 Note for a Semester Delivery Course
Reference
2 Primer on Distribution Electricity Networks
2.1 The Electricity Distribution Network and Core Concepts
2.2 Low Voltage Networks
2.3 Some Features of Distribution Networks
2.4 Managing the Distribution Network
2.5 Questions
References
3 Primer on Statistics and Probability
3.1 Univariate Distributions
3.2 Quantiles and Percentiles
3.3 Multivariate Distributions
3.4 Nonparametric Distribution Estimates
3.5 Sample Statistics and Correlation
3.6 Questions
References
4 Primer on Machine Learning
4.1 Definitions and Related Concepts
4.2 Machine Learning Taxonomy and Terms
4.2.1 Supervised Learning
4.2.2 Unsupervised Learning
4.2.3 Reinforcement Learning
4.3 Introduction to Optimisation with Gradient Descent
4.4 Questions
References
5 Time Series Forecasting: Core Concepts and Definitions
5.1 Time Series: Basic Definitions and Properties
5.2 Time Series Forecasting: Definitions
5.3 Types of Forecasts
5.4 Notation
5.5 Questions
6 Load Data: Preparation, Analysis and Feature Generation
6.1 Preparation and Pre-processing
6.1.1 Outlier Identification
6.1.2 Imputation
6.1.3 Normalisation and Transformations
6.1.4 Other Pre-processing
6.2 Feature Selection and Engineering
6.2.1 Domain Knowledge
6.2.2 Visual Analysis
6.2.3 Assessing Linear Relationships
6.2.4 Temporal Correlation Analysis
6.2.5 Basic Functions as Features
6.2.6 Common Features in Load Forecasting
6.3 Questions
Reference
7 Verification and Evaluation of Load Forecast Models
7.1 Point Forecast Error Measures
7.2 Probabilistic Forecast Error Measures
7.3 Causes of Forecast Error
7.4 Skill Scores
7.5 Residual Checks and Forecast Corrections
7.6 Questions
8 Load Forecasting Model Training and Selection
8.1 General Principles for Forecasts Trials
8.1.1 Benchmarking
8.1.2 Bias-Variance Tradeoff
8.1.3 Cross-Validation Methods
8.2 Training and Selecting Models
8.2.1 Least-Squares and Maximum Likelihood Model Fitting
8.2.2 Information Criterion
8.2.3 Hyper-Parameter Tuning
8.2.4 Weight Regularisation
8.2.5 Other Regularisation Methods
8.3 Questions
Reference
9 Benchmark and Statistical Point Forecast Methods
9.1 Benchmarks Methods
9.2 Exponential Smoothing
9.3 Multiple Linear Regression
9.4 ARIMA and ARIMAX Methods
9.5 SARIMA and SARIMAX Models
9.6 Generalised Additive Models
9.7 Questions
References
10 Machine Learning Point Forecasts Methods
10.1 k-Nearest Neighbour Regression
10.2 Support Vector Regression
10.3 Tree-Based Regression Methods
10.3.1 Decision Tree Regression
10.3.2 Random Forest Regression
10.3.3 Gradient-Boosted Regression Trees
10.4 Artificial Neural Networks
10.4.1 Feed-Forward Neural Networks
10.4.2 Recurrent Neural Networks
10.5 Deep Learning
10.5.1 Modern Recurrent Neural Networks
10.5.2 Convolutional Neural Networks
10.5.3 Temporal Convolutional Networks
10.5.4 Outlook
10.6 Feature Importance and Explainable Machine Learning
10.7 Questions
References
11 Probabilistic Forecast Methods
11.1 The Different Forms of Probabilistic Forecasts
11.2 Estimating Future Distributions
11.2.1 Notation
11.3 Parametric Models
11.3.1 Simple Univariate Distributions
11.3.2 Mixture Models
11.4 Quantile Regression and Estimation
11.5 Kernel Density Estimation Methods
11.6 Ensemble Methods
11.6.1 Residual Bootstrap Ensembles (Homoscedasticity)
11.6.2 Residual Bootstrap Ensembles (Heteroskedasticity)
11.7 Copula Models for Multivariate Forecasts
11.8 Questions
References
12 Load Forecast Process
12.1 Core-Steps for Forecast Development
12.2 Which Forecast Model to Choose?
13 Advanced and Additional Topics
13.1 Combining Forecasts
13.2 Hierarchical Forecasting
13.3 Household Level Forecasts
13.4 Global Verses Local Modeling
13.5 Forecast Evaluation: Statistical Significance
13.6 Other Pitfalls and Challenges
13.6.1 Collinearity and Confounding Variables
13.6.2 Special Days and Events
13.6.3 Concept Drift
13.6.4 Unrealistic Modelling and Data Leakage
13.6.5 Forecast Feedback
13.7 Questions
References
14 Case Study: Low Voltage Demand Forecasts
14.1 Designing Forecast Trials
14.2 Residential Low Voltage Networks
14.2.1 Initial Experimental Design
14.2.2 Data Analysis
14.2.3 Model Selection
14.2.4 Testing and Evaluation
14.3 Example Code
14.4 Summary
14.5 Questions
References
15 Selected Applications and Examples
15.1 Battery Storage Control
15.1.1 Data
15.1.2 Forecast Methods
15.1.3 Analysis of Forecasts
15.1.4 Application of Forecasts in Energy Storage Control
15.1.5 Results
15.2 Estimating Effects of Interventions and Demand Side Response
15.3 Anomaly Detection
15.4 Other Applications
15.5 How to Use Forecasts in Applications
References
Appendix A Stationary Tests for Time Series
Appendix B Weather Data for Energy Forecasting
B.1 Weather Variables and Types
B.2 Numerical Weather Forecasts
B.2.1 How Weather Forecasts Are Produced
B.2.2 Preprocessing
Appendix C Load Forecasting: Guided Walk-Through
Appendix D Further Reading
D.1 Time Series Analysis and Tools
D.2 Methods: Time Series and Load Forecasting
D.3 Low Voltage Forecasting Examples
D.4 Data and Competitions
References
Index