Energy produc-on forecas-ng based on renewable sources of energy

Similar documents
Importance of Numerical Weather Prediction in Variable Renewable Energy Forecast

Analysis and validation of 24 hours ahead neural network forecasting of photovoltaic output power

CARLOS F. M. COIMBRA (PI) HUGO T. C. PEDRO (CO-PI)

Systems Operations. PRAMOD JAIN, Ph.D. Consultant, USAID Power the Future. Astana, September, /6/2018

Current best practice of uncertainty forecast for wind energy

ANN and Statistical Theory Based Forecasting and Analysis of Power System Variables

Multi-Model Ensemble for day ahead PV power forecasting improvement

Short- term solar forecas/ng with sta/s/cal models and combina/on of models

AN INTERNATIONAL SOLAR IRRADIANCE DATA INGEST SYSTEM FOR FORECASTING SOLAR POWER AND AGRICULTURAL CROP YIELDS

Alberto Troccoli, Head of Weather and Energy Research Unit, CSIRO, Australia ICCS 2013 Jamaica, 5 December 2013 (remotely, unfortunately)

CHAPTER 6 CONCLUSION AND FUTURE SCOPE

PowerPredict Wind Power Forecasting September 2011

The Center for Renewable Resource Integration at UC San Diego

COMPARISON OF CLEAR-SKY MODELS FOR EVALUATING SOLAR FORECASTING SKILL

Short term wind forecasting using artificial neural networks

Bringing Renewables to the Grid. John Dumas Director Wholesale Market Operations ERCOT

Short Term Load Forecasting Based Artificial Neural Network

Short-term wind forecasting using artificial neural networks (ANNs)

Predic've Analy'cs for Energy Systems State Es'ma'on

Forecasting of Renewable Power Generations

SOLAR POWER FORECASTING BASED ON NUMERICAL WEATHER PREDICTION, SATELLITE DATA, AND POWER MEASUREMENTS

ANN based techniques for prediction of wind speed of 67 sites of India

Prashant Pant 1, Achal Garg 2 1,2 Engineer, Keppel Offshore and Marine Engineering India Pvt. Ltd, Mumbai. IJRASET 2013: All Rights are Reserved 356

An Operational Solar Forecast Model For PV Fleet Simulation. Richard Perez & Skip Dise Jim Schlemmer Sergey Kivalov Karl Hemker, Jr.

About Nnergix +2, More than 2,5 GW forecasted. Forecasting in 5 countries. 4 predictive technologies. More than power facilities

CS 229: Final Paper Wind Prediction: Physical model improvement through support vector regression Daniel Bejarano

Energy Forecasting Customers: Analysing end users requirements Dec 3rd, 2013 Carlos Alberto Castaño, PhD Head of R&D

SHORT TERM LOAD FORECASTING

Heat Load Forecasting of District Heating System Based on Numerical Weather Prediction Model

Recent US Wind Integration Experience

The document was not produced by the CAISO and therefore does not necessarily reflect its views or opinion.

Short and medium term solar irradiance and power forecasting given high penetration and a tropical environment

Explanatory Information Analysis for Day-Ahead Price Forecasting in the Iberian Electricity Market

Speedwell High Resolution WRF Forecasts. Application

FORECASTING: A REVIEW OF STATUS AND CHALLENGES. Eric Grimit and Kristin Larson 3TIER, Inc. Pacific Northwest Weather Workshop March 5-6, 2010

Solar Irradiance Prediction using Neural Model

David John Gagne II, NCAR

Wind power and management of the electric system. EWEA Wind Power Forecasting 2015 Leuven, BELGIUM - 02/10/2015

Improving the accuracy of solar irradiance forecasts based on Numerical Weather Prediction

Integrated Electricity Demand and Price Forecasting

Big Data Analysis in Wind Power Forecasting

Modelling Wind Farm Data and the Short Term Prediction of Wind Speeds

Short Term Load Forecasting Of Chhattisgarh Grid Using Artificial Neural Network

UniResearch Ltd, University of Bergen, Bergen, Norway WinSim Ltd., Tonsberg, Norway {catherine,

Temporal Wind Variability and Uncertainty

Forecast solutions for the energy sector

Fine-grained Photovoltaic Output Prediction using a Bayesian Ensemble

Short-Term Irradiance Forecasting Using an Irradiance Sensor Network, Satellite Imagery, and Data Assimilation

WIND POWER FORECASTING: A SURVEY

Country scale solar irradiance forecasting for PV power trading

Solar Nowcasting with Cluster-based Detrending

Power Engineering II. Fundamental terms and definitions

Irradiance Forecasts for Electricity Production. Satellite-based Nowcasting for Solar Power Plants and Distribution Networks

A Novel Method for Predicting the Power Output of Distributed Renewable Energy Resources

Influence of knn-based Load Forecasting Errors on Optimal Energy Production

SYSTEM OPERATIONS. Dr. Frank A. Monforte

California Independent System Operator (CAISO) Challenges and Solutions

2014 HIGHLIGHTS. SHC Task 46 is a five-year collaborative project with the IEA SolarPACES Programme and the IEA Photovoltaic Power Systems Programme.

Short-Term Load Forecasting Using ARIMA Model For Karnataka State Electrical Load

Integration of Behind-the-Meter PV Fleet Forecasts into Utility Grid System Operations

Short Term Load Forecasting Using Multi Layer Perceptron

P. M. FONTE GONÇALO XUFRE SILVA J. C. QUADRADO DEEA Centro de Matemática DEEA ISEL Rua Conselheiro Emídio Navarro, LISBOA PORTUGAL

Sun to Market Solutions

Predicting the Electricity Demand Response via Data-driven Inverse Optimization

FORECAST OF ENSEMBLE POWER PRODUCTION BY GRID-CONNECTED PV SYSTEMS

Wind Forecasts in Complex Terrain Experiences with SODAR and LIDAR

The Use of Analog Ensembles to Improve Short-Term Solar Irradiance Forecasting

Multi-Plant Photovoltaic Energy Forecasting Challenge with Regression Tree Ensembles and Hourly Average Forecasts

Wind Power Production Estimation through Short-Term Forecasting

Improving Efficiency of PV Systems Using Statistical Performance Monitoring

Prediction of Hourly Solar Radiation in Amman-Jordan by Using Artificial Neural Networks

WIRE: Weather Intelligence for Renewable Energies

SOLAR RADIATION FORECAST USING NEURAL NETWORKS FOR THE PREDICTION OF GRID CONNECTED PV PLANTS ENERGY PRODUCTION (DSP PROJECT)

Bayesian Based Neural Network Model for Solar Photovoltaic Power Forecasting

A new method for short-term load forecasting based on chaotic time series and neural network

1.3 STATISTICAL WIND POWER FORECASTING FOR U.S. WIND FARMS

A SOLAR AND WIND INTEGRATED FORECAST TOOL (SWIFT) DESIGNED FOR THE MANAGEMENT OF RENEWABLE ENERGY VARIABILITY ON HAWAIIAN GRID SYSTEMS

Study on Impact of Solar Photovoltaic Generation by Atmospheric Variables

Probabilistic Energy Forecasting

CustomWeather Statistical Forecasting (MOS)

Creation of a 30 years-long high resolution homogenized solar radiation data set over the

Modelling residual wind farm variability using HMMs

Benchmark of forecasting models

MODELLING ENERGY DEMAND FORECASTING USING NEURAL NETWORKS WITH UNIVARIATE TIME SERIES

Wind Power Forecasting using Artificial Neural Networks

SDG&E Meteorology. EDO Major Projects. Electric Distribution Operations

COMPARISON OF CLEAR-SKY MODELS FOR EVALUATING SOLAR FORECASTING SKILL

Solar Generation Prediction using the ARMA Model in a Laboratory-level Micro-grid

Short-Term Demand Forecasting Methodology for Scheduling and Dispatch

Wind Assessment & Forecasting

Integration of WindSim s Forecasting Module into an Existing Multi-Asset Forecasting Framework

Dr SN Singh, Professor Department of Electrical Engineering. Indian Institute of Technology Kanpur

Climate services in support of the energy transformation

Application of Artificial Neural Network for Short Term Load Forecasting

Research Article Hybrid Power Forecasting Model for Photovoltaic Plants Based on Neural Network with Air Quality Index

Power Forecasting and Dynamic Line Rating

WMO Aeronautical Meteorology Scientific Conference 2017

Value of Forecasts in Unit Commitment Problems

Research and application of locational wind forecasting in the UK

Wind resource assessment and wind power forecasting

Transcription:

Energy produc-on forecas-ng based on renewable sources of energy S. Leva Politecnico di Milano, Dipar1mento di Energia Via La Masa 34, 20156 Milano, Italy sonia.leva@polimi.it, www.solartech.polimi.it

Goal and outline 2 The goal of this speech is to analyze how, starting from weather forecast, we can predict in term of hourly-curve the energy production by RES for one day two days, a week ahead. 1. Introduc-on: the energy produc-on forecas-ng and the role of RES set up by the interna-onal energy agency 2. The energy forecas-ng from RES 3. Weather forecas-ng 4. The PV forecas-ng and error defini-ons, some examples 5. The wind forecas-ng, some examples 6. Conclusions

3 1. Introduc-on: the energy produc-on forecas-ng and the role of RES in the world and in Italy 2. The energy forecas1ng from RES 3. Weather forecas1ng 4. The PV forecas1ng and error defini1on, some examples 5. The wind forecas1ng, some examples 6. Conclusions

Introduc/on: the energy produc/on forecas/ng and the role of RES The IEA forecasts confirm that the demand for energy (not just electricity) will grow especially in non- OECD Share of global energy demand 4 Global energy demand rises by over one- third in the period to 2035, underpinned by rising living standards in China, India & the Middle East

Introduc/on: the energy produc/on forecas/ng and the role of RES IEA predic1ons for the future (scenario "reference"): oil, gas, coal con1nue to dominate the energy (not just electricity) produc1on 5

Introduc/on: the energy produc/on forecas/ng and the role of RES IEA predic1ons of how will be sa1sfied the demand of electricity in the world. 6 «KING» COAL!

Introduc/on: the role of RES in Italy 7 In five years the electricity genera1on by RES in Italy has doubled. Hydro Geothermal Bioenergy Wind Solar The data are really up to date: august 2013!

Introduc/on: the role of RES in Italy 8 Electricity generation in Italy in the first seven monthes of 2013 Bioenergy Hydro wind PV geothermal Thermoelectric fossil Number of plants producing electricity passes in a decade from 1 thousand to 550,000 Centralized system tends towards a mixed system of genera1on (distributed genera1on) A growing number of households and factories now are involved in electricity genera1on

9 1. Introduc1on: the energy produc1on forecas1ng and the role of RES in the world and in Italy 2. The energy forecas-ng from RES 3. Weather forecas1ng 4. The PV forecas1ng and error defini1on, some examples 5. The wind forecas1ng, some examples 6. Conclusion

The energy forecas/ng from RES 10 Distributed system: grid- connected RES installa1ons are decentralized RESs energy produc1on has a stochas1c behavior. RESs are much smaller than tradi1onal u1lity generators Today's available transforma1on and storage capabili1es for electric energy are limited and cost- intensive. Challenges of controlling and maintaining energy from inherently intermittet sources involves many aspects: efficicency, reliability, safety, stability of the grid and ability to forecast energy production.

The energy forecas/ng from RES 11 Forecasting of PV/wind is an estimation from expected power production of the plant in the future. For monitoring and maintenance purposes To help the grid operators to be\er manage the electric balance between power demand and supply and to improve embedding of distributed renewable energy sources. In stand alone hybrid systems energy forecas1ng can help to size all the components and to improve the reliability of the isolated systems.

Time scale classifica/on for RES Forecas/ng 12 Time horizon Very short- term Short- term Medium- term Range Few seconds to 30minutes ahead 30 minutes to 6 hours ahead 6 hours to 1 day ahead Applica1ons - Control and adjustment ac1ons - Economic Dispatch Planning - Load Increment/Decrement Decisions - Generator Online/Offline Decisions - Opera1onal Security in Day- Ahead - Electricity Market Long- term 1 day to 1 week or more ahead - Unit Commitment Decisions - Reserve Requirement Decisions - Maintenance Scheduling to Obtain Op1mal Opera1ng Cost

13 1. Introduc1on: the energy produc1on forecas1ng and the role of RES in the world and in Italy 2. The energy forecas1ng from RES 3. Weather forecas-ng 4. The PV forecas1ng and error defini1on, some examples 5. The wind forecas1ng, some examples 6. Conclusion

Weather forecast 14 Forecasts of RES produc/on is based on weather forecasts. This is an orthogonal step to a grid operator: weather data is usually obtained from meteorological services. The most influencing factor for output determina1on are: solar energy produc1on: global irradia1on forecast. wind energy produc1on: wind speed amplitude and direc1on forecast, pressure forecast The use of precise weather forecast models is essen1al before reliable energy output models can be generated.

Weather forecast models 15 Numerical Weather Prediction (NWP) Complex global NWP models are used to predict a number of variables describing the dynamic of the atmosphere and then to derive the weather at a specific point of interest. Post processing techniques are applied to obtain down scaled models (1.5 km). European Center for Medium-Range Weather-Forecasts Model (ECMWF) Global Forecast System (GFS), North American Mesoscale Model (NAM) 3-6 hors Cloud Imagery Influence of local cloudiness is considered to be the most critical factor for estimation of solar irradiation. The use of satellite provide high-quality medium term forecast. Satellite-based (METEOSAT), Total Sky Imager, 24h-48h Statistical Methods based on historical observation data using time series regression models ARIMA Articial Neural Networks (ANN), Fuzzy Logic (FL), ARMA/TDNN ANFIS RBFNN MLP Long term

Weather forecast 16 Meteorology remains a field of uncertainty. Time horizon is a crucial aspect. Sunshine and wind speed can only be predicted with accuracy a few days in advance. The number and type of variables describing the physics and dynamic of the atmosphere are fundamental topics. Cloudy index or irradia;on are two indexes that can impact on the forecast in a different way.

17 1. Introduc1on: the energy produc1on forecas1ng and the role of RES in the world and in Italy 2. The energy forecas1ng from RES 3. Weather forecas1ng 4. The PV forecas-ng and error defini-on, some examples 5. The wind forecas1ng, some examples 6. Conclusion

The PV forecast: different Models. 18 Physical Models to describe the rela1on between environmental data and power - highly sensitive to the weather prediction - have to be designed specifically for a particular energy system and location Sta-s-cal Models are based on persistent predic1on or on the 1me series' history Persistent predic1on, Similar- days Model Stochas1c Time Series Machine Learning Ar1ficial neural network (ANN) learn to recognize pa\erns in data using training data sets. They need historical weather forecas1ng data and PV- plant measured data for their training Hybrid Models are any combina;on of two or more of the previously described methods. They could be two different stochas1c models or a stochas1c model and a physical model.

The PV forecast: Physical Models. 19 Weather forecast Physical Algorithm PV energy forecast PV energy forecast Global Irradiation, Cloud cover, Temperature, ecc Plant Description; Monitoring System Measured data

The PV forecast: Sta/s/cal Models 20 TRAINED NEURAL NETWORK Environmental temperature

Error Defini/ons 21 In order to correctly define the accuracy of the predic1on and the rela1ve error it is necessary to analyze different defini1ons of error. The star1ng point reference is the hourly error e h : P m,h is the average power produced in the hour (or energy kwh) P p,h is the predic1on provided by the forecas1ng model From this basic defini1on, other error defini1ons have been inquired: Absolute error based on the hourly output expected power (p=predicted) [AEEG]: e pu, p, h e = P P mh, ph, h absolute error based on the hourly output produced power (m=measured) [AEEG]: e h m, h p, h = P pu, m, h P = P = e P ph, ph, e P h mh, AEEG=Italian Authority for Electricity and Gas

Error Defini/ons 22 Mean absolute error [AEEG et al]: 1 MAE = P P N h = 1 mh, ph, Normalized mean absolute error NMAE, based on net capacity of the plant C [AEEG et al] NMAE % N N 1 Pmh, Pph, = 100 N C h= 1 C could be the rated power, the maximum observed or expected power!!!!

Error defini/ons 23 Weighted mean absolute error WMAE % based on total energy produc1on [AEEG et al.]: WMAE % N P 1 mh, Pph, h= = 100 N P h= 1 mh, Normalized root mean square nrmse, based on the maximum observed power [Urlicht et al]: nrmse = 1 N N h= 1 P P mh, ph, m ax( P ) mh, 2 23

Some examples: Hybrid Models (ANN+Physical) 24 ü Plant data validation: Theoretical Solar Irradiance (clear sky) ü Physical data: Theoretical Solar Irradiance (clear sky), Sunrise-, Sunset-hour ü weather forecasts ü weather forecasts data analysis: evaluation of their reliability. ü Comparison between ANN forecasts and other methods ü Ensembled methods ü Error definitions ü Accuracy assessment of the obtained results

A. Hybrid Models (ANN+Physical) at SolarTech Lab 25 TRAINED NEURAL NETWORK Environmental temperature Clear Sky Physical Model 4.4kW, Milano, Italy

A. Some Results: Solar Tech Lab 26 NMAEp%= 3.08% NMAEp% = 30.1% pink line: there was an error in the weather forecast.

B. Hybrid Models (ANN+Physical) PV Plant in Cuneo 27 285kW PV plant, Cuneo (Italy) Meteo dataset: Day, hour, Environmental temperature, wind direc1on, wind speed, global solar irradia1on Goals: Analysis of the error due to the weather forecas1ng Ensembles method: use more than one trials of stochas1c methods to make the forecast Absolute hourly error based on predicted power vs measured power

B. Hybrid Models (ANN+Physical) PV Plant in Cuneo 28 Error due to the weather forecas1ng: difference between the irradia1on given by weather service and the irradia1on measured

B. Hybrid Models (ANN+Physical) PV Plant in Cuneo 29 Error due to the weather forecas1ng: Absolute hourly errors of GI are sorted from largest to smallest. Absolute hourly error based on expected global irradiation (predicted) and on the measured global irradiation. Solar Radia;on forecas;ngs are affected by a great error!

Some Results: Power Plant 30 ANN are stochastic methods: Different trials give different forecas1ng curves. Ensemble: power/energy forecast is calculated considering the hourly average value of different (here 10) trials. Absolute hourly error based on expected output power (predicted) and on the measured output power. Ensemble methods reduce the error! The error based on the measured power is bigger than the one based on the predicted! Hourly sample (from sunrise to sunset)

Some Results: Power Plant 31 NMAEp% = 10 NMAEr% = 5.86 WMAEp% = 16.58 NMAEp% = 29.14 NMAEr% = 15 WMAEp% = 50 NMAEp% = 16 NMAEr% = 7.33 WMAEp% = 28.7

Some Results: Power Plant 32 expected output power (predicted) and versus measured output power. 1 year: NMAEp = 12.15%, NMAEr%=7,34%

33 1. Introduc1on: the energy produc1on forecas1ng and the role of RES in the world and in Italy 2. The energy forecas1ng from RES 3. Weather forecas1ng 4. The PV forecas1ng and error defini1on, some examples 5. The wind forecas-ng, some examples 6. Conclusion

Wind Forecas/ng 34 Forecas1ng of wind is an es1ma1on from expected power produc1on of the wind turbines in the future. This power produc1on is expressed in kw or MW depending on the nominal capacity of the wind farm. Forecas1ng methods described for PV can be applied Error defini1ons described for PV are used Kalman or Kolmogorov- Zurbenko are usually adopted to be\er ex1mate the wind speed elimina1ng the effects of noise and systema1c errors Hybrid approaches (ANN + CFD computa1onal fluid dynamics sonware) can improve the accuracy of the forecas1ng 34

Example: Cas/glione Messer Marino Wind Farm 35 Input parameters: Inviromental temperature [ C] Atmospheric pressure [hpa] Wind speed intensity [m/s] Humidity [%] Cloud coverage [%] Performance parameters WMAE NMAE Implemented feed-forward ANN with details on input, output, and hidden layers.

Some Results: Cas/glione Messer Marino Wind Farm 36 16 14 Wind plant forecast P m,h P p,h 12 Power (MW) 10 8 6 v 1000 iterations: NMAEp = 40.2 % NMAEr= 14% 4 2 0 84 86 88 90 92 94 96 Day

Hybrid methods: computa/onal fluid dynamics so[ware 37 The use of tools of CFD (computational fluid dynamics software) can improve the predictive capability of forecasting systems. The computational cost greatly limits its practical applicability for wind farms with a large number of wind turbines. Expensive measurement systems (see anemometer towers) to model the field.

The most promising method: Hybrid methods 38 Plant Description ANN Historical Wind data Historical Power data Ground description Physical algorithm CFD Analysis by GSE, ANEMOS.plus

39 1. Introduc1on: the energy produc1on forecas1ng and the role of RES in the world and in Italy 2. The energy forecas1ng from RES 3. Weather forecas1ng 4. The PV forecas1ng and error defini1on, some examples 5. The wind forecas1ng, some examples 6. Conclusions

Conclusions 40 The meteorological services have an important influence on the power forecas1ng system for PV and wind energy. The input data analysis is very important and cost- intensive Hybrid forecas1ng method are the most promising methods both for PV and Wind energy forecas1ng PV. Clear sky data are very useful to reduce error. Wind. The use of special filters (eg Kalman or KZ) may be useful for the removal of systema1c errors of the forecasts of wind speed provided by the NWP and used as input to sta1s1cal methods. The performance of the forecas1ng models are strongly related to the 1me horizon of the forecast and to the characteris1cs of the land on which the plant/farm is placed. The need for energy forecas1ng from RES is a recent topic!!!

41 THANK YOU! www.solartech.polimi.it Diapartimento di Energia Via Lambruschini, 4 20133 Milano e-mail: sonia.leva@polimi.it e-mail: giampaolo.manzolini@polimi.it Tel. +39 02 2399 3800 (Centralino) 3709 (Leva) 3810 (Manzolini)

Some Results: Power Plant 42 Absolute hourly error based on expected output power (predicted) and on the measured output power.