An Adaptive Machine Learning Framework for Behind-the-Meter Load PV Disaggregation
Abstract : A significant amount of distributed photovoltaic (PV) generation is invisible to distribution system operators since it is behind the meter on customer premises and not directly monitored by the utility. The generation essentially adds an unknown varying negative demand to the system, which causes additional uncertainty in determining the total load. This uncertainty directly impacts system reliability, cold load pickup, load behavior modeling, and hence cost of operation. This paper proposes an adaptive machine learning framework to disaggregate PV generation and load from the net measurement. The framework estimates PV generation and load based on measurements/data collected from smart transformers, smart meters, other sensors, and weather stations. The proposed framework core idea is to transform the data such that: a) the machine learning model can effectively utilize the time dependency of measurements, and b) the measurements are transformed into a lower-dimensional space to reduce complexity while maintaining accuracy. The transformed measurements are then used to train the machine learning models for load/PV dis-aggregation. Machine learning models investigated include linear regression (LR), decision tree (DT), random forest (RF), and Multilayer Perceptron (MLP). Several test/ training split scenarios, including 90%-10% Split, One-Month-Out, One-Season-Out, and Panel-Independent Split, are presented to provide a thorough evaluation of the proposed framework. Results show that the proposed framework can estimate PV generation with high accuracy using low-complexity methods, and random forest is found to provide superior performance compared to the other ML models investigated.
? In this paper, net demand of individual customers is constructed by subtracting real BTM PV generation from real native demand.
? The hourly native demand and PV generation data are from Midwest U.S. utilities. This data is available online
? The graph partitioning can be conducted in different ways according to different objective functions.
? In this paper, the objective function is to roughly maximize the dissimilarity between the different graph clusters while minimizing the similarity within each cluster
? This invisibility, along with the stochastic nature of solar power, can cause new problems for utilities, such as inaccurate load forecasting and estimation, inefficient service restoration, and sub-optimal network expansion decisions.
? Thus, it is of significance to disaggregate PV generation from net demand to enhance grid-edge observability.
? One solution to this problem is to monitor each single rooftop PV generation by installing extra metering devices.
? However, due to the large number of distributed PVs, this option comes at a significant cost for utilities.
? In a data-driven approach is proposed based on dimension reduction and mapping functions using PV generation measurement data from temporarily-installed sensors.
? In a linear proxybased estimator is developed to disaggregate a solar farm generation from feeder-level measurement using µPMU data, along with the measured power profile of nearby PV plants, and global horizontal irradiance (GHI) proxy data.
? In a PV generation disaggregation approach is presented for groups of residential customers, under the assumption that their aggregate active power is measured at the point of common coupling (PCC) to the grid
? The solar-demand disaggregation accuracy of the DD-based SSS and the proposed RGVP-based SSS .
? It can be seen that the proposed RGVP based SSS outperforms DD-based SSS in terms of MAP E.
? The reason for this better performance is that the RGVP-based method can identify the candidate exemplars that are highly correlated with the BTM real load/solar powers.
? The proposed BTM PV generation and native demand disaggregation approach is also applied to secondary transformers.
|