Sky Image GHI Estimation for Behind-the-Meter PV Forecasting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
As the prevalence of distributed photovoltaic (PV) energy resources increases, power grids face challenges in accurately managing energy from PV sources, leading to issues like voltage deviation, frequency oscillation, and unplanned islanding, due to stochastic deviations in solar radiation, especially on cloudy days, and the lack of direct access to behind-the-meter PV systems.
Innovation Solution
A system utilizing a convolutional neural network (CNN)-based image regression model to estimate global horizontal irradiance (GHI) from sky images and a Bayesian Structural Time Series (BSTS) model to predict PV power output, enabling real-time or near-real-time adjustments in power distribution network operations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If distributed PV systems are widely deployed to reduce grid dependence, then energy independence and cost savings are improved, but grid management accuracy and stability deteriorate due to stochastic solar radiation variations
Solution Approach 1:
The patent implements a feedback mechanism by using sky imaging devices to continuously monitor cloud cover and solar radiation conditions, then feeding this information back to the grid management system. This allows real-time adjustment of grid operations based on actual PV generation conditions, resolving the stability issue while maintaining distributed PV deployment.
Solution Approach 2:
The patent introduces an intermediary estimation system that acts as a mediator between distributed PV systems and grid management. By using sky images and machine learning models to estimate PV power output, the system provides accurate forecasting information that enables better grid scheduling and stability management without requiring direct access to behind-the-meter PV systems.
2Measurement precision
If sky imaging and machine learning models are used to estimate PV power output, then measurement precision and grid management accuracy are improved, but device complexity and computational requirements increase
Solution Approach 1:
The patent employs self-service principles by using readily available sky images from existing infrastructure and open-source machine learning models. The system serves itself by automatically processing sky images through pre-trained models to generate PV power estimates, reducing the need for complex custom-built systems while maintaining high measurement precision.
Solution Approach 2:
The patent applies universal principles by using multi-functional sky imaging devices that can serve both as weather monitoring tools and as inputs for PV power estimation. The machine learning models are trained to perform multiple functions including cloud detection, irradiance estimation, and power forecasting from the same sky image data, reducing overall system complexity.
3Productivity
If real-time PV power estimation is implemented to improve grid management, then productivity and response time are improved, but data processing requirements and computational energy consumption increase
Solution Approach 1:
The patent applies preliminary action by pre-training machine learning models offline using historical sky image and PV generation data. Once trained, the models can rapidly process new sky images in real-time with minimal computational energy requirements. This preliminary preparation enables fast, energy-efficient real-time PV power estimation for improved grid management productivity.
Solution Approach 2:
The patent uses computationally efficient models that can be rapidly retrained or replaced if needed, rather than relying on complex, energy-intensive systems. The approach uses lightweight machine learning models that consume minimal computational energy during operation, allowing for easy updates and adaptations without significant energy investment.
Data Source
AI summary
An example device is configured to determine, based on a sky image of a portion of sky over a power distribution network and using a convolutional neural network (CNN)-based image regression model, an estimated global horizontal irradiance (GHI) value and manage or control the power distribution network using the estimated GHI value. The device may also be configured to determine, based on GHI values and aggregate load values for at least a portion of the power distribution network, using a Bayesian Structural Time Series model, an estimated photovoltaic power output value for the at least a portion of the power distribution network. The device may manage or control the power distribution network using the estimated photovoltaic power output value.