Unified SCADA Load Forecasting via Kalman and Savitzky-Golay Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current load forecasting techniques fail to provide precise near-term (five-minute to one hour ahead) real-time load forecasts, leading to inaccurate day-ahead generation scheduling and inefficient use of power generation resources, as they are plagued by erroneous and noisy SCADA measurements, and lack a unified framework to integrate near-term and short-term forecasts effectively.
Innovation Solution
The implementation of a unified framework that combines a two-stage modified Kalman filter and an augmented Savitzky-Golay filter to smooth and filter SCADA data, allowing for the use of observed loads when valid and model-predicted loads when outliers are detected, and blending autoregressive near-term and non-autoregressive short-term models to bridge the gap between different forecast horizons.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If statistical modeling is used for short-term demand forecasting, then forecast coverage is improved, but near-term forecast precision deteriorates
Solution Approach 1:
The patent divides the forecasting system into two distinct segments: a statistical modeling component for short-term demand forecasting and a manual crafting component for near-term load forecasting. This segmentation allows each component to optimize for its specific time horizon, with the statistical model handling longer-term trends and the manual process capturing real-time operational nuances that statistical models miss.
Solution Approach 2:
The patent introduces an intermediary blending mechanism that combines outputs from both statistical models and manual forecasting processes. This intermediary layer reconciles the strengths of both approaches, allowing the system to leverage statistical forecast coverage while incorporating manual adjustments that improve near-term precision without completely discarding either methodology.
2Adaptability or versatility
If different forecasting techniques are used for next-day and near-term forecasts, then forecast versatility is improved, but system complexity deteriorates
Solution Approach 1:
The patent merges previously separate forecasting processes into a unified blended forecast system. By combining next-day statistical forecasts with near-term manually crafted forecasts into a single integrated system, the patent reduces operational complexity while maintaining the versatility of using different techniques for different time horizons. The blending mechanism provides a single coherent forecast output that incorporates both approaches.
Solution Approach 2:
The patent creates a universal forecasting framework that can handle multiple time horizons (next-day and near-term) within a single system. This multi-functional system allows operators to use the same blended forecast output for both day-ahead generation scheduling and real-time operational decisions, eliminating the need for completely separate forecasting systems while maintaining adaptability to different time scales.
3Measurement precision
If real-time adjustments are made to generation schedule, then operational accuracy is improved, but forecast integration deteriorates
Solution Approach 1:
The patent implements a feedback mechanism where real-time adjustments made by operators are captured and fed back into the forecasting system. This feedback loop ensures that valuable operational information from manual adjustments is not lost but instead incorporated into future forecast iterations, improving both operational accuracy and forecast integration over time.
Solution Approach 2:
The patent performs preliminary blending of statistical and manual forecasts before final operational use. By pre-integrating the components and creating a blended forecast output in advance, the system prepares a unified forecast that reduces the need for last-minute adjustments while preserving the ability to make real-time modifications when necessary.
Data Source
AI summary
Techniques for near-term data filtering, smoothing and forecasting are described herein. In one example, data is received from supervisory control and data acquisition (SCADA) measurements available in an electrical grid. The data may be filtered according to a two-stage Kalman filter, which may include a ramp rate filter test and a load level filter test. The filtered data may then be smoothed according to an augmented Savitzky-Golay filter. Within the filter, a lift multiplier may correct for bias, which may have been introduced by load changes (e.g., an early morning increase in load). In one example, the lift multiplier may be calculated as a ratio between a smoothed load from a centered Savitzky-Golay moving average and a right hand side constrained Savitzky-Golay moving average. The filtered and smoothed data may be used in forming near-term forecast(s), which may be performed by autoregressive model(s).