Short-Term Forecasting for Small-Scale Distributed Energy Resources (DERs) in Power System = 소규모 분산자원의 단기예측기법에 관한 연구

다국어 초록 (Multilingual Abstract)

Short-term forecasting (STF) is a critical process for power system operation and planning. Due to importance of forecasting in power system, various forecasting methods have been developed at the transmission level, and its performance has been enhanced gradually. On the other hand, the forecasting for small-scale resources is still a challenging issue, because a relatively small amount of power makes the higher uncertainty and volatility. Considering the increasing number of the distributed energy resources (DERs) and usefulness in power system operation, it is required to enhance the forecasting performance for the small-scale electricity load and small-scale photo-voltaic (PV) generation.
In case of small-scale electricity load, due to changes in household demographics, lifestyles, and the number of electric appliances, and varying weather conditions, the power consumption is becoming diverse on a daily basis. In order to overcome these uncertainty, sufficient historical data is required. On the other hand, most of small-scale PV plants are installed at remote locations from the weather data centers (WDCs), and they do not usually have their own measuring apparatus to record for weather conditions. In fact, small-scale load and PV forecasting are suffering from adequate data collection issues to enhance its forecasting performance. Thus, a novel STF method is required to be developed for reducing higher forecasting errors that should assist several areas of the power system.
As deep learning enables higher forecasting accuracy either with immense data or with homogeneous data for time series analysis, new hybrid STFs for both single household load series and small-scale PV generations have been proposed in this dissertation. For single households, a convolution neural network (CNN) based hybrid forecasting model is proposed using a data augmentation strategy. The proposed data augmentation technique can artificially enlarge the training data by incorporating homogeneous residual load series for the CNN-based model and work out the historical data requirement problem effectively. For small-scale PV plants, a long-short term memory (LSTM) based hybrid forecasting model is proposed that combines both a weather data mixing model and a similar days detection (SDD) method. The presented weather data mixing models compute adequate weather data from the relatively nearby weather data centers (WDCs) using proposed inverse distance and inverse correlation techniques. The newly developed SDD method detects similar days considering the different impact from the weather variables on PV power generation. Both weather data mixing models and SDD are combined for small-scale PV forecasting to lower the impact from long-distance and weather uncertainty problems. The simulation results of both presenting hybrid algorithms are compared with corresponding conventional methods that indicate the effectiveness and application of the proposed algorithms in the power system.

목차 (Table of Contents)

• ABSTRACT
• 1. INTRODUCTION
• A. Distributed Energy Resources 1
• B. Objective and Contributions 6
• C. Dissertation Outline 7
• 2. SHORT-TERM FORECASTING PROBLEMS ON DISTRIBUTED ENERGY RESOURCES
• A. Forecasting in Power System 8
• B. Load Forecasting Problems for Single Household 14
• C. Forecasting Problems for Small-Scale PV Plants 20
• 3. SHORT-TERM FORECASTING METHOD FOR SINGLE HOUSEHOLD ELECTRICITY LOAD
• A. Concept of Data Augmentation 26
• B. Data Augmentation Method for Single Household Electricity Load 27
• C. Proposed Single Household Load Forecasting Method 34
• 4. SHORT-TERM FORECASTING METHOD FOR SMALL-SCALE PV GENERATION
• A. Weather Data Collection 43
• B. Proposed Weather Data Mixing Models 45
• C. Proposed Similar Day Detection (SDD) Method 48
• D. Proposed Small-Scale PV Forecasting Method 57
• 5. SIMULATION RESULTS AND DISCUSSION
• A. Forecasting Results of Single Household Electric Load 59
• B. Forecasting Results of Small-Scale PV Generation 72
• 6. CONCLUSION AND FUTURE WORK 86
• REFERENCES 88
• 국문초록 96
• ACKNOWLEDGEMENT

참고문헌 (Reference)

1. T. Essentials of the self-organizing map, Kohonen, 37 , 52 ? 65, , 2013

2. Wong D.W. An adaptive inverse-distance weighting spatial interpolation technique, Lu , G.Y, 34 ( 9 ) , 1044 ? 1055 ., , 2008

3. Time Series Forecasting by Evolving Deep Belief Network with Negative Correlation Search, Lin , Y, , 2018

4. Pro Deep Learning with TensorFlow : A Mathematical Approach to Advanced Artificial Intelligence in Python, Pattanayek , S., 1st edpp . 153 ? 222, , 2017