Online demand peak shaving with machine-learned advice in digital twins

As the use of physical instruments grows,control algorithms are being increasingly deployed to enhance efficiency and reliability through digital twin technology.Demand load management is central to energy systems within digital twins,which significantly impacts operational costs.Peak demand loads can lead to substantial monthly utility expenses without proper management.AMPAMOD,a randomised online algorithm incorporating machine-learned insights is introduced to optimise battery operations and mitigate peak demand loads.AMPAMOD leverages limited-bit information from ma-chine learning models to inform its online decision-making process for cost-effective load management.We provide theoretical evidence demonstrating that AMPAMOD maintains minimal advice complexity,has a linear computational cost,and achieves a bounded competitive ratio.Extensive trace-driven experiments with real-world household data reveal that AMPAMOD successfully reduces peak loads by over 90%,outperforming other benchmarks by at least 50%.These experimental findings align with our theoretical assertions,showcasing the effectiveness of AMPAMOD.

A Collaborative Medical Diagnosis System Without Sharing Patient Data

As more medical data become digitalized,machine learning is regarded as a promising tool for constructing medical decision support systems.Even with vast medical data volumes,machine learning is still not fully exploiting its potential because the data usually sits in data silos,and privacy and security regulations restrict their access and use.To address these issues,we built a secured and explainable machine learning framework,called explainable federated XGBoost(EXPERTS),which can share valuable information among different medical institutions to improve the learning results without sharing the patients’ data.It also reveals how the machine makes a decision through eigenvalues to offer a more insightful answer to medical professionals.To study the performance,we evaluate our approach by real-world datasets,and our approach outperforms the benchmark algorithms under both federated learning and non-federated learning frameworks.

Game Theory based Optimal Defensive Resources Allocation with Incomplete Information in Cyber-physical Power Systems against False Data Injection Attacks

Modern power grid is fast emerging as a complex cyber-physical power system(CPPS)integrating physical current-carrying components and processes with cyber-embedded computing,which faces increasing cy-berspace security threats and risks.In this paper,the state(i.e.,voltage)offsets resulting from false data injection(FDI)attacks and the bus safety characterization are applied to quantify the attack consequences.The state offsets are obtained by the state estimation method,and the bus safety characterization considers the power net-work topology as well as the vulnerability and connection relationship of buses.Considering the indeterminacy of attacker’s resource consumption and reward,a zero-sum game-theoretical model from the defender’s perspective with incomplete information is explored for the optimal allocation of limited defensive resources.The attacker aims to falsify measurements without triggering threshold alarms to break through the protection,leading to load shedding,over-voltage or under-voltage.The defender attempts to ensure the estimation results to be as close to the actual states as possible,and guarantee the system’s safety and efficient defensive resource utilization.The proposed solution is extensively evaluated through simu-lations using the IEEE 33-bus test network and real-time digital simulator(RTDS)based testbed experiments of the IEEE 14-bus network.The results demonstrate the effec-tiveness of the proposed game-theoretical approach for optimal defensive resource allocation in CPPS when lim-ited resources are available when under FDI attacks.Index Terms—Optimal strategy,game theory,Nash equilibrium,CPPS,FDI attack.

A rank-based multiple-choice secretary algorithm for minimising microgrid operating cost under uncertainties

The increasing use of distributed energy resources changes the way to manage the electricity system.Unlike the traditional centralized powered utility,many homes and businesses with local electricity generators have established their own microgrids,which increases the use of renewable energy while introducing a new challenge to the management of the microgrid system from the mismatch and unknown of renewable energy generations,load demands,and dynamic electricity prices.To address this challenge,a rank-based multiple-choice secretary algorithm(RMSA)was proposed for microgrid management,to reduce the microgrid operating cost.Rather than relying on the complete information of future dynamic variables or accurate predictive approaches,a lightweight solution was used to make real-time decisions under uncertainties.The RMSA enables a microgrid to reduce the operating cost by determining the best electricity purchase timing for each task under dynamic pricing.Extensive experiments were conducted on real-world data sets to prove the efficacy of our solution in complex and divergent real-world scenarios.

Quick-MIMIC:A Multimodal Data Extraction Pipeline for MIMIC with Parallelization

Medical big data with artificial intelligence are vital in advancing digital medicine.However,the opaque and non-standardised nature embedded in most medical data extraction is prone to batch effects and has become a significant obstacle to reproducing previous works.This paper aims to develop an easy-to-use time-series multimodal data extraction pipeline,Quick-MIMIC,for standardised data extraction from MIMIC datasets.Our method can fully integrate different data structures into a time-series table,including structured,semi-structured,and unstructured data.We also introduce two additional modules to Quick-MIMIC,a pipeline parallelization method and data analysis methods,for reducing the data extraction time and presenting the characteristics of the extracted data intuitively.The extensive experimental results show that our pipeline can efficiently extract the needed data from the MIMIC dataset and convert it into the correct format for further analytic tasks.