Modified Pattern Sequence-based Forecasting for Electric Vehicle Charging Stations
Three algorithms for the forecasting of energy consumption at individual EV charging outlets have been applied to real world data from the UCLA campus. Out of these three algorithms, namely k-Nearest Neighbor (kNN), ARIMA, and Pattern Sequence Forecasting (PSF), kNN with k=1, was the best and PSF was the worst performing algorithm with respect to the SMAPE measure. The advantage of PSF is its increased robustness to noise by substituting the real valued time series with an integer valued one, and the advantage of NN is having the least SMAPE for our data. We propose a Modified PSF algorithm (MPSF) which is a combination of PSF and NN; it could be interpreted as NN on integer valued data or as PSF with considering only the most recent neighbor to produce the output. Some other shortcomings of PSF are also addressed in the MPSF. Results show that MPSF has improved the forecast performance.
- Research Organization:
- City of Los Angeles Department
- Sponsoring Organization:
- USDOE Office of Electricity (OE)
- DOE Contract Number:
- OE0000192
- OSTI ID:
- 1332712
- Report Number(s):
- DOE-UCLA-00192-21
- Resource Relation:
- Conference: IEEE International Conference on Smart Grid Communications (SmartGridComm '14) Venice, Italy 3-6 Nov. 2014
- Country of Publication:
- United States
- Language:
- English
Similar Records
Incomplete Data in Smart Grid: Treatment of Values in Electric Vehicle Charging Data
Automated Dynamic Demand Response Implementation on a Micro-grid