Energy ICT/V2G/Smart Grid

                           (에너지 정보통신기술(ICT)/차량과 전력망 간의 에너지 교환(V2G)/스마트 그리드)

Energy ICT/Vehicle-to-Grid(G)(V2G)/Smart Grid

              (에너지  정보통신기술/차량과  전력망 간의 에너지 교환(V2G)/ 스마트 그리드)


Main Content of Smart Grid/V2G/G2V/Energy Grid:



Smart Grid and Smart Grid Communication Networks 


The U.S. DOE defines smart grid as “an automated, widely distributed energy delivery network incorporating the benefits of distributed computing and communications to deliver real-time information and enable the near-instantaneous balance of supply and demand.”

The Energy Independence and Security Act of 2007 (EISA-2007) signed  to law by President George W. Bush in 2007 characterized a smart grid: (1) Increased use of digital information and controls technology to improve reliability, security, and efficiency of the electric grid. (2) Dynamic optimization of grid operations and resources, with full cyber-security. (3) Deployment and integration of distributed resources and generation, including renewable resources. (4) Development and incorporation of demand response, demand-side resources, and energy-efficiency resources. (5) Deployment of 'smart' technologies (real-time, automated, interactive technologies that optimize the physical operation of appliances and consumer devices) for metering, communications concerning grid operations and status, and distribution automation. (6) Integration of 'smart' appliances and consumer devices. (7) Deployment and integration of advanced electricity storage and peak-shaving technologies, including plug-in electric and hybrid electric vehicles, and thermal storage air conditioning. (8) Provision to consumers of timely information and control options. (9) Development of standards for communication and interoperability of appliances and equipment connected to the electric grid, including the infrastructure serving the grid. (10) Identification and lowering of unreasonable or unnecessary barriers to adoption of smart grid technologies, practices, and services. [EISA-2007] 

Fig. 1 shows the smart grid communication physical architecture. The conventional power grid  consists of bulk power generators, power transmission system, power distribution system, and customer premise, while in the smart grid  various renewable energy sources, micro grids, and energy storage systems are connected to the power grid, enabling the both-way power transmission. In order to support  the both-way power exchange, we need  a rather complex smart grid communication layer consisting of wide area netorks, neighborhood area networks, field area networks, and customer premise networks including home area networks, building area networks, and industrial area networks. 

Figure 1   Smart Grid Communication Physical Architecture (Source: IEEE P2030)

A smart utility network  is required to support home-to-home, home-to-grid, and home area communications. We proposed a multichannel CSMA/CA scheme for IEEE 802.15.4g and evaluate the performance in the smart utility network environment. We develop link level simulation in terms of  packet error rate and system level simulation in terms of throughput, collision probability, mean tramsfer delay, and packet loss probability.

In the early development stage of Korean smart grid, we started to study various aspects of smart grid and carried out research projects  and also taught  a new graduate-level course: smart grid networks with the following content:


Prediction of Solar Irradiance [K.Y. Bae]

We collected meteorological data for past 26 months (2012.01 - 2014.04) from the weather station of the Korea Meteorological Administration, which is located in Yuseong-gu, Daejeon (36.37°N, 127,37°E). The data sampling interval of weather report is 1 hour.  Meteorological parameters and value ranges in Daejeon, Korea  and distributed PV solar power generators in the KAIST campus are shown in Table 1  and  Figure 2.

We investigated the correlation coefficient between PV output power and various weather factors. We can find how much weather factors affect PV output power by comparing the correlation coefficient between the PV output power and each meteorological parameter. Fig. 3 shows the correlation coefficient for PV output power  with  respect to various weather parameters.  The result shows that Cloud factor and humidity are highly correlated with the PV output power throughout the year. On the other hand, temperature has a low correlation value with the exception of July and wind speed has a nearly zero correlation value.

Table 1  Meteorological Parameters and Value Ranges in Daejeon, Korea

 Figure 2  Distributive PV Solar Power Generators at KAIST Campus

Figure 3   Correlation Coefficient for PV Output Power  with  respect to Various Weather Parameters


We proposed to classify the input and output data into three clusters (Cluster 1, Cluster 2, and Cluster3) according to sunny days, partially sunny/cloudy days, and cloudy/rainy days, respectively. Input and output data of the same cluster are used to support supervised learning based on the SVM regression for making each training model of solar irradiance prediction. One of SVM regression models, which are built in the training step, is selected according to the weather type for solar irradiance prediction.  We evaluate the prediction accuracy of the support vector machine(SVM) scheme for Clusters 1, 2, and 3 and obtain the predicted solar irradiance with the measured result and the corrsponding scatter plot of the measured and predicted solar irradiance. Fig. 4  shows the predicted solar irradiance with the measured result for sunny days (Cluster 1). Fig. 5  shows the scatter plot of the measured predicted solar irradiance. We can observe from Figures  that small errors occur between the measured and predicted solar irradiance. The R^2 metric is equal to 0.9749 in sunny days.  Figures 6 and 7 for partially sunny/cloudy days (Cluster 2) show that the prediction result of the proposed SVM regression is still accurate with an R^2 metric of 0.9318. Figures 8 and 9  for cloudy/rainy days (Cluster 3) show that some prediction errors occur due to a sudden change of solar irradiance in cloudy or rainy days.

We compare the numerical results for various prediction schemes: non-linear auto-regressive(NAR) scheme, artificial neural network(ANN) scheme, and the proposed SVM in terms of the average MRE, average RMSE, and R2. Table 2 shows the  performance evaluation results of various solar irradiance prediction schemes.  The proposed SVM scheme with weather type based data clustering yields best accuracy, compared with that of  the conventional prediction schemes. In addition, the numerical results show that the RMSE and MRE for the proposed SVM regression are much better than those of the NAR and the ANN schemes, respectively. The R^2 metric, which is equal to 0.8685, is smaller than that of the ANN scheme because the proposed scheme also can not fully avoid quite large errors in cloudy or rainy days due to a sudden change of irradiance.


Figure 4  The predicted and measured data for  Cluster 1                           Figure 5   The scatter plot of the predicted and                                                                                                                                    measured solar  irradiance for                                                                                                                                               Cluster 1 (Sunny days)

Figure 6  The predicted and measured data for  Cluster 2                                 Figure 7  The scatter plot of the predicted and                                                                                                                                     measured solar  irradiance for                                                                                                                                           Cluster 2 (Partially sunny/cloudy days)

Figure 8  The predicted and measured data for  Cluster 3                                 Figure 9   The scatter plot of the predicted and                                                                                                                                       measured solar  irradiance for                                                                                                                                   Cluster 3 (Cloudy/rainy days)

Table 2    Performance Evaluation  Results for Various Solar Irradiance Prediction Schemes

A Solar Power Prediction Scheme Based on Satellite Images [H.S. Jang]

We proposed  a solar power prediction scheme based on satellite images and support vector machine.  Atmosheric motion vectors(AMVs) provide the overall information of atmospheric motion and wind direction and speed of lower, middle, and upper wind fields. The red, green, and blue colors on the weather satellite image shown in Fig. 10  indicate the lower, middle, and upper wind fields, respectively.

The second figure shows the amount of clouds in the range of  1~100 on each pixel. The cloud information is a very important factor for the prediction of future condition of clouds on the target area, and is used for input variables of machine learning (ML). The irradiance image in the third figure is based on the amount of light intensity reflected  by the ground  and the irradiance is measured in Watts per square meter(W/m^2) in the range of 0~1000.

Since the irradiance value almost linearly affects photovoltaic (PV) power generation, it can be converted into the power generation from PV solar panels.


The overview of weather satellite images based irradiance forecast process is shown in Fig. 11. The first step is to extract atmospheric motion vectors according to three different color vectors and to estimate the wind direction as well as wind speed on each location on the satellite image. The second step is to calculate the impact factor on a target area. In order to extract cloud factors in search areas, we calculate a set of AMV parameters W(i,k): the average wind direction, speed, and angle of the i-th search area for color k; calculate the average amount of clouds Ci of the i-th search area; calculate the shaded area after time t; calculate the cloud impact factor of the i-th search area on the target area after time t; and then calculate the total cloud impact factor on the target area after t minutes. The third step is to perform training with SVM regression. SVM is one of high performance ML schemes, which was originally proposed by Vapnik; SVM is designed to not only minimize the error, but also maximize the separation margin among different classes; SVM can be applied to regression methods in supervised learning. In the training, the first object is to predict the future amount of clouds at the target area; the other objective is to predict the future irradiance at the target area.  Depending on the prediction objectives, we need to configure the input and output data sets in a different way for ML trainings. In the prediction of future amount of clouds, we need the input data vector for the amount of clouds with a set of the current amount of clouds at the search areas, and the total cloud impact factor for the prediction horizon of t minutes.; the input column vector consists of the current amount of clouds at the l search areas; the total cloud impact factors comprises a l+1 column vector. A single output value corresponds to the amount of clouds at the target area after time t minutes. N input and output data sets are used  for SVM regression trainer for predicting  the amount of clouds at the target area after t min. In the prediction of future irradiance, we additionally insert the solar altitude angle at the specific date and time and the current irradiance of the target area into the input data vector; Finally, the input column vector is configured as {cloud(search areas), impact factor, solar altitude angle, current irradiance}, with i+3 dimensions; and the irradiance of the target area after t minutes is set to a single output value. 

Figure 10  Weather  Satellite Images: (1) Atmospheric motion vector (AMR) (2) Amount of clouds (3) Irradiance image

Figure 11  Satellite Images Based Solar Irradiance Forecast Process

We evaluate the performance of the proposed SVM-based prediction model, a persistence model, a nonlinear autoregressive recurrent model, and an artificial neural network model for the amount of clouds and irradiance in terms of prediction accuracy metrics such as root mean square error (RMSE), mean relative error (MRE), and the coefficient of determination, R^2. 

Table 3 compares the prediction results for the amount of clouds with a prediction horizon of 60 minutes. The proposed SVM model outperforms the other three models in terms of RMSE, MRE< and R^2. 

Fig. 12 compares the predicted amount of clouds of the proposed SVM-based model with the measured results and also shows the scatter plot of the measured and predicted amount of clouds for the proposed SVM-based prediction model. Even though some prediction errors occur, the predicted results generally agree well with the measured values. The scatter plot results show  that a larger amount of clouds (near 100) and a smaller amount of clouds (near 0) are well predicted, while it is rather hard to predict the medium amount of clouds. 

Table 4 summarizes the prediction results for the amount of clouds based on the proposed SVM-based model with varying prediction time horizons. The very short term (15 and 30 minutes horizon) and short term (60 minutes horizon) prediction forthe amount of clouds  show high prediction accuracy, while the prediction accuracy of 90~300 minutes horizon degrades gradually because new formation and disappearance of clouds  amy occur and they are not considered in the prediction model.

Table 3 Prediction results for the amount of clouds after 60 minutes                               Table 4  Prediction results for                                                                                                                                                        the amount of clouds                                                                                                                                                         with varying horizons

 Figure 12 Satellite images based prediction results for the amount of clouds: (1)Comparison of the measured and predicted amount of clouds; (2) Scatter plot of the measured and predicted amount of clouds based on the proposed SVM-based model with t = 60 minutes

Table 5 compares the prediction results for the irradiance with a prediction horizon of 60 minutes. The proposed SVM model outperforms the other three models in terms of RMSE, MRE< and R^2. 

Fig. 13 compares the predicted irradiance of the proposed SVM-based model with the measured result and also shows the scatter plot of the measured and predicted irradiance for the proposed SVM-based prediction model. Even though some prediction errors occur, the predicted results generally agree well with the measured values. The scatter plot converges to the red line, which represents the high accuarcy for the future irradiance.

Table 6 summarizes the prediction results for the irradiance based on the proposed SVM-based model with varying prediction time horizons. The very short term (15 and 30 minutes horizon) and short term (60 minutes horizon) prediction forthe amount of clouds  show high prediction accuracy, while the prediction accuracy of 90~300 minutes horizon degrades gradually because new formation and disappearance of clouds  amy occur and they are not considered in the prediction model.  The irradiance of 15 ∼ 150 minutes horizons can be well predicted with R^2 = 0.9770 ∼ 0.8641, while the prediction accuracy degrades from the 180 minutes horizon to R^2 =0.8338. 


Table 5 Prediction results for the irradiance after 60 minutes                                 Table 6  Prediction results for irradiance with                                                                                                                                  varying prediction time horizon                                                                                                                                                                                                                                                                    

Figure 13  Satellite Images Based Prediction of Future Irradiance: (1)Comparison of the measured and predicted irradiance; (2) Scatter plot of the measured and predicted irradiance based on the proposed SVM-based model with t = 60 minutes

Application of Electric Vehicles (EVs) to Frequency Regulation Service in Electricity Market [K.S. Ko]

Vehicle-to-grid (V2G)  technology  is to push back battery energy in electric vehicles (EVs) to grid in order to provide various applications such as load leveling, smoothing renewable energy, and frequency regulation service  in power grid.  In future, EV aggregators manage  and control a group of EV batteries and may play an important role in frequency regulation markets as  V2G operators.

Fig. 14 shows a research domain representing the application of EVs to frequency regulation service.  Each EV aggregator manage and control a group of EVs in terms of  charging (G2V) and discharging (V2G). It can participate in frequency regulation market. Power grid consists of power generators, transmission systems, and distribution systems. It can provide consumers with electricity. System operator manages electricity market including frequency regulation service and overall energy management system (EMS).

Ther are some research issues related to the provision of frequency regulation service with a large number of EVs.


Figure 14    A Research Domain Representing  the Application of EVs to Frequency Regulation Service

In this research area, we investigate the following research issues in detail: [K.S. Ko's dissertation]


KOSEP Microgrid Testbed Project

Overview of KOSEP-KAIST Microgrid Field Demonstration Project

Figure 15   Munji Campus Microgrid

Figure 16 The Second Stage Munji Campus Microgrid(2016~2017) (Source: Seondo Electric Co., Ltd.)

Planning of KAIST Campus Energy Grid Project

KAIST was  selected as a preliminary business operator for the ‘Smart Grid Expansion Project’ promoted by the Ministry of Trade, Industry and Energy in 2013. This project, in which 19 institutions participated as a consortium, was one of the largest projects promoted on university campuses. It would be promoted from 2015 to 2017 if the detailed project panning was  accepted by the government after a preliminary feasibility study.

We proposed to build a KAIST Campus Energy Grid with self-sufficient energy supply. In the proposal, we planned to invest a matching fund of  19.2 billion won (in a total budget of 45  billion won) over three years from 2015 to build fundamental energy grid systems such as ▲ a renewable power generation platform ▲ a campus energy management system ▲ a smart grid data operation center ▲ an electric vehicle operation system. (Figs. 17, 18, 19, and 20)

We proposed to install a 3 MW fuel cell power plant for power production and cooling/heating supply, a 2 MW solar power generation facility using rooftop and outdoor parking lot, a Smart Grid Total Operation Center,LED bulb replacements, a Zero Energy Building, a Fuel Cell Research Center,a  KAIST Energy Mix Promotion Center, and a solar power infrastructure for electric vehicles.

We submitted a final proposal to our government after a preliminart study in 2013. However, we failed to get approval from our government.

Figure 17   Overview of KAIST Campus Energy Grid Project

Figure 18   KAIST Zero Energy Building  and KAIST Campus Energy Management System (CEMS)

Figure  19  Future KAIST Campus Energy Grid System  and  Research Issues

Figure 20   Reference Model: University of California, Irvine Campus Grid

(1) Journal Papers

(2) International Conference Papers (국제 학회 논문)

 

(3) Patents (특허기술)