Written by Yishen Wang, Xiao Liang, and Fei Zhou
Recently, climate changes have raised serious concerns and urged broad actions from all sectors of society: government, industry, and academia. Energy systems have then been recognized as one key part in addressing such concerns. One critical mitigation strategy identified is to reduce the carbon emissions. In China, the “Dual Carbon” target was strategically proposed in 2020 by the government, aiming to reach a carbon emission peak by 2030 and carbon neutrality by 2060.
Similarly, the European Commission announced “Fit for 55” in 2021 to reduce greenhouse gas emissions by 55% by 2030. Energy systems have thus entered an era of achieving a new equilibrium point for the three-way balance between economics, reliability, and sustainability.
To meet such evolving trends, energy systems are coordinating the resources from the generation, network, demand, and energy storage sides with various perspectives. The whole operation and planning has extended to a more geographically dispersed, more hierarchical-level, and more complicating-coupled architecture. Leveraging the capabilities of high-performance computing and data analytics, higher volume data is being stored and processed at a faster speed as well as more comprehensive analysis.
As the hub to connect the transmission networks, distribution networks, and end-users, energy systems in smart cities have widely deployed demand-side distributed energy resources (DER) and other flexible resources to enable smart and efficient city-level energy management systems. There are many trending techniques for the computing and analytics of demand-side resources in smart cities to provide situational awareness and decision-making support, including energy consumption forecasting, profiling analytics, load modeling analytics and crowd intelligent decision-making.
First, it is essential to monitor and estimate the system operation parameters and states to support system planning, operation, and control. One basic thing is to forecast the energy consumption. That means not only electricity consumption, but gas and heat consumption also need to be predicted. The multi-energy system or integrated energy system concepts require a unified gas-heat-electric energy consumption prediction as an input. In addition, prediction at various levels, including aggregated levels and individual levels, enhance the scheduling and optimization at higher granularity. Due to behind-the-meter (BTM) behaviors with deployed DER, energy prediction becomes more challenging at the system side or highly aggregated coupling points. Non-intrusive load monitoring (NILM) in general provides some useful information. Furthermore, due to random and stochastic demand behaviors, probabilistic forecasting reveals more statistical information than the deterministic forecasting to the subsequent operation decision-making tasks, which improves the operation economics, reliability and potential carbon emission. Most of this prediction work could be achieved through machine learning, specifically supervised learning. From classical linear regression, support vector machine (SVM), gradient boosting trees (GBT), to recent deep learning, these learning algorithms have shown superior performances in terms of feature extraction and prediction accuracy. With the progress made by the artificial intelligence community and more deployed measurements, the prediction performances are expected to be further improved.
In addition to energy consumption prediction, smart city situational awareness also requires extracting deeper insights to build demand behavior profiling analytics, which could be viewed as an extension of the consumption prediction. Compared with demand time series forecasting, the demand behavior profiling mainly investigates the common and different consumption patterns for end users in the smart cities. All the demand-side resources, including distributed PV, energy storage systems (ESS), electric vehicles (EV), heat pumps, electric boilers, are contributing to generate such profiles and estimate the potential for grid services. These resources serve as efficient and economic dispatchable assets to enhance system flexibility. Utilities and ISO/RTO would benefit from the demand response, emergency control, frequency regulation, and other ancillary services. Statistical analysis, clustering, and unsupervised learning have all shown good profiling performance in recent literature. At the same time, data privacy should be paid special attention these days. Personal energy consumption data involves private information that users may not want to share. Privacy computing techniques, including federated learning, become a tool to address this new issue and protect user privacy when information is shared among multiple parties, and it should be further investigated.
Furthermore, load dynamic modeling is another timely computing and analytic topic for smart cities. Different from static perspectives from the forecasting and profiling, dynamic modeling provides a dynamic analysis tool to improve the system security and stability. Electrical load components have become more and more complex with highly penetrated power electronics equipment and devices. From the city perspective, it is impossible to build exact and detailed dynamic load models. Finding simplified load models capturing load dynamic features would be a solution. Previously, commonly adopted static load models (ZIP, exponential) or dynamic load models (ZIP+IM) need to be updated with models to represent the DER, feeder, single-phase motors and various new load characteristics. Recent proposed Western Electricity Coordinating Council (WECC) composite load models partially addresses some modeling concerns, yet it is still challenging to estimate parameters. Nonlinear optimization, deep learning, deep reinforcement learning, and meta-heuristics have shown progress to provide parameter estimation strategies. However, the modeling and corresponding online estimation methods still need further development to represent various static and dynamic load features.
In addition to these situational awareness applications, operation decision-making is also important to smart city management. One important topic is to coordinate and optimize resources from various system participants. As the smart cities develop, there are EV aggregators, virtual power plants (VPP), integrated energy service providers, retailers, microgrids, and other new participant types. These entities build aggregate demand-side resources flexibility to support system operation and improve renewable integration. Due to the large number of end users, traditional centralized operation is inadequate, and more operation schemes are developed, including local operation, distributed operation, hierarchical operation, and so on. Crowd intelligence, as an analytic tool, could model the gaming decisions among these energy participants and has shown some promising results. Furthermore, a combination of data-driven and knowledge-based methods is an alternative to handle the data and model availability issues with improved performances.
To summarize, computing and analytics for demand-side resources is becoming increasingly important for smart city energy systems. This newsletter discussed a few applications developed to improve situational awareness and operation decision-making. New research topics and techniques are expected to further improve smart city energy systems.