LOAD FORECASTING. Amanpreet Kaur, CSE 291 Smart Grid Seminar

Size: px

Start display at page:

Download "LOAD FORECASTING. Amanpreet Kaur, CSE 291 Smart Grid Seminar"

Molly Stewart
5 years ago
Views:

1 LOAD FORECASTING Amanpreet Kaur, CSE 29 Smart Grid Seminar

2 Outline Introduction Motivation Types Factors Affecting Load Inputs Methods Forecast Algorithm Example

3 Load forecasting is way of estimating what future electric load will be for a given forecast horizon based on the available information about the state of the system.

4 CAISO Outlook

5 Motivation Operation and planning by ISOs and utility companies Load following Real time dispatch Operating Reserves Trading in electricity market Smart Grid- Automation and Control

Forecast (month - years ) System Independent System Operator e.

6 Types Forecast Horizon Short Term Load Forecast ( one hour - week) Mid Term Load Forecast ( a week - month) Long Term Load Forecast (month - years ) System Independent System Operator e.g. ERCOT Utility e.g. SDG&E Organization e.g. UCSD Building e.g. SDSC

Power Load and Generation Forecas For High Solar Penetration Commun Amanpreet Kaur and Carlos F. M.

7 varying between 2 GW to 7 GW. As compared to CAISO, there is high demand even in the months of January and February. ERCOT demand are at their peaks. In this research we present integrated load/generation forecasting methodologies and the associated economic gains for such communities. UCUCSD San Diego! Power Load and Generation Forecas For High Solar Penetration Commun Amanpreet Kaur and Carlos F. M. Coimbra Jacobs School of Engineering, University of California, San D HAM load forecast using Hourly Ensemble Forecast. The above SDSC Introduction! forecast is for week starting from Thursday to Wednesday UCaSan Diego and UC Merced meet substantial parts of of annual energy demand from highly variable solar April823. their Minute Data power plants. High solar penetration communities emulate 7 Load profile for UCSD campus for the year 2 with approximately -2% solar penetration. UCM UC Merced! 5ïMinute Average Data the future grid scenario for CA, which aims at 33% 2 HAM renewable MAEenergy MBE RMSE MAPE penetration by 22.RThe major challenge with this level of penetration of renewable is that the variability in solar power production ERCOT resources has direct impact on the load demand from the grid, especially during diurnal periods when energy prices.27 and LS-Hourly demand are at their peaks. In this research we present 6 Load [kw] integrated load/generation forecasting methodologies and the associated economic gains for such communities. Conclusion! 5-minute ahead load forecast and power load for the UCSD campus for every 5-minute time interval. This type of load forecast is critical for deploying Automated and Fast Demand Response (ADR/FDR) strategies. UCtoSan Diego An improved methodology compute hour!ahead forecasts for 9 two major Independent (ISO) and 3/9 3/2 3/2System 3/22Operators 3/23 3/24 i.e. CAISO 3/25 ERCOT has been shown. Since 6 March 23 till now, there has been 4% improvement in the forecast. The HAM forecasts Load profile for UCM campus for the year 2 with 5-2% solar penetration. The UCM campus has a unique load shape because it

8 Factors affecting Load Weather Time Economic Random Time series of CAISO hourly (29 22) and daily peak load (22). Heinemann, G. T.; Nordmian, D. A.; Plant, E. C., "The Relationship Between Summer Weather and Summer Loads - A Regression Analysis," Power Apparatus and Systems, IEEE Transactions on, vol.pas-85, no., pp. 44,54, Nov. 966 Gross, G.; Galiana, F.D., "Short-term load forecasting," Proceedings of the IEEE, vol.75, no.2, pp.558,573, Dec. 987

9 Inputs Meteorological forecast e.g. Temperature, Relative Humidity, Wind Speed, Dew Point, etc. Type of day e.g. Weekday, Holiday, Festival, etc. Time of the day Hippert, H.S.; Pedreira, C.E.; Souza, R.C., "Neural networks for short-term load forecasting: a review and evaluation," Power Systems, IEEE Transactions on, vol. 6, no., pp.44,55, Feb 2

10 Methods Regression Stochastic Time Series Fuzzy Logic Artificial Intelligence/ Machine Learning Hybrid H. K. Alfares and M. Nazeeruddin, Electric load forecasting: literature survey and classification of methods, International Journal of System Science, vol. 33, no., pp , 22. L. Suganthi and A. A. Samuel, Energy models for demand forecasting review, Renewable and Sustainable Energy Reviews, vol. 6, no. 2, pp , 22.

11 Stochastic Time Series AutoRegressive Model (AR) AutoRegressive Model with exogenous Input (ARX) Non-Linear ARX AutoRegressive Moving Average (ARMA) ARMAX ARIMA (Seasonal Modeling) State Space Models e.g. Kalman Filter Gross, G.; Galiana, F.D., "Short-term load forecasting," Proceedings of the IEEE, vol.75, no.2, pp.558,573, Dec. 987 Taylor, J.W.; McSharry, P.E., "Short-Term Load Forecasting Methods: An Evaluation Based on European Data," Power Systems, IEEE Transactions on, vol.22, no.4, pp.223,229, Nov. 27 Carter C.K. and Kohn R., On Gibbs sampling for state space models Biometrika (994) 8(3): doi:.93/ biomet/8.3.54

12 Fuzzy Logic It is many valued logic that approximates the expected values Disadvantage: Rules for fuzzy logic are determined experimentally with hit and trial Proposed Solution: Using optimization techniques like Simulated Annealing, GA and ANN model functions. Papadakis, S.E.; Theocharis, J.B.; Kiartzis, S. J.; Bakirtzis, A.G., "A novel approach to short-term load forecasting using fuzzy neural networks," Power Systems, IEEE Transactions on, vol.3, no.2, pp.48,492, May 998 Mori, H.; Kobayashi, H., "Optimal fuzzy inference for short-term load forecasting," Power Industry Computer Application Conference, 995. Conference Proceedings., 995 IEEE, vol., no., pp.32,38, 7-2 May 995

13 Artificial Intelligence Artificial Neural Network Advantage: Ability to learn Disadvantage: Over fitting and over parameterizing k Nearest Neighbor Advantage: Good for forecasting load profile for a whole day or longer period of time Disadvantage: Requires historical data to create a database. Hippert, H.S.; Pedreira, C.E.; Souza, R.C., "Neural networks for short-term load forecasting: a review and evaluation," Power Systems, IEEE Transactions on, vol.6, no., pp.44,55, Feb 2 Lora A.C., Riquelme J.C., Ramos J.L.M., Santos J.M.R., Exposito A.G., Influence of knn-based Load Forecasting Errors on Optimal Energy Production, Progress in Artificial Intelligence, Lecture Notes in Computer Science, Volume 292, 23, pp 89-23

14 Expert & Hybrid Models Support Vector Regression (SVR) Self Organizing Map (SOM) Meta Learning Hybrid Model: Combination of various forecasting models Using various optimization techniques like particle swarm, ant colony, GAs, etc. are introduced to optimize the model parameters, input selection and model selection!! Bo-Juen Chen; Ming-Wei Chang; Chih-Jen Lin, "Load forecasting using support vector Machines: a study on EUNITE competition 2," Power Systems, IEEE Transactions on, vol.9, no.4, pp.82,83, Nov. 24 Shu Fan; Luonan Chen, "Short-term load forecasting based on an adaptive hybrid method," Power Systems, IEEE Transactions on, vol.2, no., pp.392,4, Feb. 26 Marin Matijaš, Johan A.K. Suykens, Slavko Krajcar, Load forecasting using a multivariate meta-learning system, Expert Systems with Applications, Volume 4, Issue, September 23, Pages , ISSN

the model performance and updating the forecast If performance

15 Load Forecast Algorithm Data Preprocessing Model formulation or selection Identification or updating model parameters Testing the model performance and updating the forecast If performance is unsatisfactory return to Step or Step2, else return to Step 3

applied to the ERCOT control region, which shows the robustness of the methodology. Example: Forecast HAM CAISO Model! Forecast using Ensemble Method HAM Block diagram of the forecast model.

Forecast of the various models are combined using weights (wi ) for each model computed by Least Square (LS) optimization on either hourly, weekly or model basis.

16 applied to the ERCOT control region, which shows the robustness of the methodology. Example: Forecast HAM CAISO Model! Forecast using Ensemble Method HAM Block diagram of the forecast model. Day-Ahead Market forecast (DAM) is used as an input and hour ahead forecast (fi,j ) for each model is produced. Forecast of the various models are combined using weights (wi ) for each model computed by Least Square (LS) optimization on either hourly, weekly or model basis. The final ensemble forecast (F i ) is a combination of various forecast computed as follows f, f2,... f,2 f2,2... F f,n w f2,n w2 F2.... = fm,n wn F n f, f2,... F f,n w f2,n w2 F2.... = f,2 f2,2... fm, fm,2 fm,n wn F n HAM load forecast using Hourly Ensemble Forecast. T time forecast is available at coimbra.ucs forecasting_plots/caiso_ham.php CAISO Sample Autocorrelation Least Square Sample Partial Autocorrelations DAM ARX ARMAX Box-Jenkins Polynomial Model Non-linear ARX Non-linear HW LSïHourly LSïWeekly LSïModels ï.5 2 ï.5 2 ï.5 2 ï.5 2 ï ï.5 2 ï.5 2 ï.5 2 ï.5 Model ANNïModels 2 ï.5 CAISO 2 LS-Hourl LS-Mode 2 LS-Week Lag Autocorrelation and Partial Autocorrelation betw residuals of HAM from CAISO, Ensemble (LS-Ho Weekly and LS-Models) and ANN model. The blue l 95% confidence interval. Except CAISO, residuals f fm, fm,2 models are almost white which validates that the f model is capturing all the information in the provided t about the dynamics of the system load.

17 Error Analysis Sample Autocorrelation.5 CAISO LS Hourly LS Weekly LS Models ANN Models.5 2 Sample Partial Autocorrelations Lag

18 QUESTIONS? Thanks!

Energy Consumption Prediction for Optimum Storage Utilization

Energy Consumption Prediction for Optimum Storage Utilization Eric Boucher, Robin Schucker, Jose Ignacio del Villar December 12, 2015 Introduction Continuous access to energy for commercial and industrial