Energy Time Series Forecasting (eBook)

Lars Dannecker holds a diploma in media computer science from the Technische Universität Dresden and is pursuing a doctorate as a member of the Database Technology Group led by Prof. Dr.-Ing. Wolfgang Lehner. 

Preface 6
Acknowledgements 8
Contents 10
List of Figures 13
List of Tables 16
Chapter 1 Introduction 17
Chapter 2 The European Electricity Market: A Market Study 26
2.1 Current Developments in the European Electricity Market 27
2.1.1 Structure of the European Electricity Market 27
2.1.2 Development of Renewable Energy Sources in Europe and Germany 28
2.1.3 Impact of Volatile Renewable Energy Sources 32
2.1.4 How to Keep the Electricity Grid in Balance 35
2.1.5 Extending the Transmission Grid and Energy Storage 40
2.1.6 Demand-Side Management and Demand-Response 45
2.1.7 Changes on the European Electricity Market 47
2.1.8 Improvements in Forecasting Energy Demand and Renewable Supply 52
2.2 The MIRABEL Project: Exploiting Demand and Supply Side Flexibility 56
2.2.1 Flex-Offers 56
2.2.2 Architecture of MIRABEL’s EDMS 58
2.2.3 Basic and Advanced Use-Case 60
2.3 Conclusion 61
Chapter 3 The Current State of Energy Data Management and Forecasting 63
3.1 Data Characteristics in the Energy Domain 64
3.1.1 Seasonal Patterns 65
3.1.2 Aggregation-Level-Dependent Predictability 67
3.1.3 Time Series Context and Context Drifts 70
3.1.4 Typical Data Characteristics of Energy Time Series 72
3.2 Forecasting in the Energy Domain 73
3.2.1 Forecast Models with Autoregressive Structures 73
3.2.2 Exponential Smoothing 77
3.2.3 Machine Learning Techniques 80
3.3 Forecast Models Tailor-Made for the Energy Domain 82
3.3.1 Exponential Smoothing for the Energy Domain 83
3.3.2 A multi-equation forecast model using autoregression 84
3.4 Estimation of Forecast Models 86
3.4.1 Optimization of Derivable Functions 87
3.4.2 Optimization of Arbitrary Functions 88
3.4.3 Incremental Maintenance 90
3.4.4 Local and Global Forecasting Algorithms Used in this book 91
3.5 Challenges for Forecasting in the Energy Domain 96
3.5.1 Exponentially Increasing Search Space 96
3.5.2 Multi-Optima Search Space 97
3.5.3 Continuous Evaluation and Estimation 98
3.5.4 Further Challenges 99
Chapter 4 The Online Forecasting Process: Efficiently Providing Accurate Predictions 100
4.1 Requirements for Designing a Novel Forecasting Process 100
4.2 The Current Forecasting Calculation Process 102
4.3 The Online Forecasting Process 107
4.3.1 The Forecast Model Repository 109
4.3.2 A Flexible and Iterative Optimization for Forecast Models 112
4.3.3 Evaluation 121
4.4 Designing a Forecasting System for the New Electricity Market 126
4.4.1 Integrating Forecasting into Data Management Systems 127
4.4.2 Creating a Common Architecture for EDMSs 128
4.4.3 Architecture of an Integrated Forecasting Component 130
Chapter 5 Optimizations on the Logical Layer: Context-Aware Forecasting 133
5.1 Context-Aware Forecast Model Materialization 134
5.1.1 Case-based Reasoning and Context-Awareness in General 134
5.1.2 The Context-Aware Forecast Model Repository 136
5.1.3 Decision Criteria 137
5.1.4 Preserving Forecast Models Using Time Series Context 139
5.1.5 Forecast Model Retrieval and Assessment 144
5.1.6 Evaluation 149
5.2 A Framework for Efficiently Integrating External Information 153
5.2.1 Separating the Forecast Model 154
5.2.2 Reducing the Dimensionality of the External Information Model 155
5.2.3 Determining the Final External Model 158
5.2.4 Creating a Combined Forecast Model 160
5.2.5 Integration with the Online Forecasting Process 161
5.2.6 Experimental Evaluation 163
5.3 Exploiting Hierarchical Time Series Structures 168
5.3.1 Forecasting in Hierarchies 169
5.3.2 Approach Outline 170
5.3.3 Classification of Forecast Model Coefficients and Parameters 171
5.3.4 Aggregation in Detail 173
5.3.5 Applying the System to Real-World Forecast Models 176
5.3.6 Hierarchical Communication 178
5.3.7 Experimental Evaluation 179
5.4 Conclusion 184
Chapter 6 Optimizations on the Physical Layer: A Forecast-Model-Aware Time Series Storage 186
6.1 Related Work 187
6.1.1 Optimizing Time Series Management 187
6.1.2 Special Purpose DMS 188
6.1.3 Summarizing comparison 190
6.2 Creating an Access-Pattern-Aware Time Series Storage 191
6.2.1 Model Access Patterns 192
6.2.2 Access-Pattern-Aware Storage 195
6.3 Applying the Access-Pattern-Aware Storage to Real-World Forecast Models 200
6.3.1 Optimized Storage for Single-Equation Models 200
6.3.2 Optimized Storage for Multi-Equation Models 203
6.4 Evaluation 206
6.4.1 Single-Equation Models 207
6.4.2 Multi-Equation Models 209
6.5 Conclusion 214
Chapter 7 Conclusion and Future Work 216
References 221