创造负荷预测的全新前沿.pdf

1、Load Forecasting:Application-Driven Evaluation of Machine Learning Models across Spatial and Temporal ScalesTianzhenHong,HanLiBuilding Technology and Urban SystemsDivision Lawrence BerkeleyNational Laboratory2021 Pacific Northwest Heat DomeThe actual peak demand for the Northwest Power Pool(NWPP)reg

2、ion on June 28,2021,was around 24,700 MW,which was about 12%higher than the predicted value.The actual peak demand for the Bonneville Power Administration(BPA)service area on June 28,2021,was around 22,400 MW,which was about 12%higher than the predicted value.2OutlineBERKELEY LAB Intro to load forec

3、asting What are challenges of conventional methods?Why AI/ML-based methods?A case study Summary and future researchIntroductionWhat?Predicting future electricity demand based on historical data and influencing factors such as weather,economic activity,and consumer behavior.BERKELEY LAB4Electric Load

4、 ForecastingPrediction HorizonVery Short-TermForm minutes to a few hoursFor minute-to-minute load balancingShort-TermFrom hours to afew daysFor daily operationMedium-TermFrom a week to a few monthsFor strategicdecisions andinfrastructuredevelopmentLong-TermFrom a few months to several yearsIntroduct

5、ionWhy?Grid Reliability:Ensures sufficient supply to meet demand,preventing blackouts or overloads.Cost Efficiency:Helps optimize generation,transmission,and distribution,reducing operational costs.Integration of Renewables:Facilitates the transition to clean energy by managing variabilityand uncert

6、ainty.Policy and Planning:Supports strategic investments in energy infrastructure and decarbonization goals.Consumer Benefits:Enables demand-side management,leading to energy savings andbetter service.BERKELEY LAB5IntroductionBERKELEY LAB6Technical ApproachesPhysics-basedData-drivenConventional Meth