Qian Gao

Recently, data-driven approaches are widely used to predict and fix the values of integer variables in unit commitment (UC) problems to reduce the computational burden. However, learning the complex pattern between the UC model characteristics and integer solutions requires a huge number of samples, which is an obstacle in the practical application. Meanwhile, the prediction error is hard to control. Facing these challenges, this paper proposes a hybrid offline-online approach to predict the UC solution using few-shot samples (typically, no more than 5). To avoid the reliance on the sample scale, the prediction task is strategically decomposed into offline and online tasks. In the offline process, the internal solution information of the branch-and-bound process is collected to determine the candidate integer variables that can be predicted using online information. In the online process, an instance-specific root relaxation method is used to determine the values to which the candidate integer variables should be fixed. A parameter tuning method of the hybrid offline-online framework is presented to improve the performance. Based on the prediction result, an accompanying model is constructed and solved in parallel to provide a better estimation of upper bound and accelerate the branch-and-bound process without compromising any optimality. Test cases on public and utility test systems show that the proposed method can achieve up to 14.69 times acceleration under a variety of conditions with guaranteeing the optimality.

In a set of similar unit commitment (UC) problems, a variable is called a stable integer variable (SIV) if its optimal value is 0 or 1 on a regular basis, which reflects an inherent pattern in UC solutions. The computational complexity can be significantly reduced by fixing SIVs with a minor chance of accuracy loss. To identify SIVs, this letter proposes a method to collect internal solution information from the initial branch-and-bound process of artificial mixed-integer programming problems generated from the given problem. A distinct advantage of this method is that because no prior offline training is needed, learning is instance-specific and has no dependency on the training set. With identified SIVs, machine learning-based acceleration is achieved. Based on 50 test cases from publicly available and practical data, the proposed method improves the percentage of solved problems within the clearing time window from 68% to 96% compared with commercial solvers, counting the total time of data collection, prediction, and optimization.

In this letter, we present an internally induced acceleration method for the branch-and-cut (B&C) process of unit commitment (UC) problem on the basis of general mixed-integer programming (MIP) solvers. The B&C process is internally guided by solving a small-scale induced MIP model formulated by combining the congestion management information with global upper and lower bounds. Based on the test cases of RTS-GMLC and practical utility data, the UC calculation can be accelerated by 2.29 and 2.06 times on average, respectively, with no influence on the solution optimality.

The latest sepsis definition includes both infection and organ failure, as evidenced by the sequential organ failure assessment (SOFA) score. However, the applicability of the pediatric SOFA score (pSOFA) is not yet determined. This study evaluated the effectiveness of both pSOFA and system inflammatory reaction syndrome (SIRS) scores in predicting sepsis-related pediatric deaths. This is a retrospective multi-center cohort study including hospitalized patients <18 years old with diagnosed or not-yet-diagnosed infections. Multivariate analyses were carried out to evaluate risk factors for in-hospital mortality. According to Youden index (YI), three sub-categories of pSOFA were screened out and a new simplified pSOFA score (spSOFA) was formed. The effectiveness and accuracy of prediction of pSOFA, SIRS and spSOFA was retrieved from the area under the receiver operating characteristic curve (AUROC) and Delong’s test. A total of 1,092 participants were eligible for this study, and carried a 23.4% in-hospital mortality rate. The 24-h elevated pSOFA score (24 h-pSOFA), bloodstream infection, and mechanical ventilation (MV) requirement were major risk factors associated with sepsis-related deaths. The AUROC analysis confirmed that the spSOFA provided good predictive capability in sepsis-related pediatric deaths, relative to the 24 h-pSOFA and SIRS. The pSOFA score performed better than SIRS in diagnosing infected children with high mortality risk. However, it is both costly and cumbersome. We, therefore, proposed spSOFA to accurately predict patient outcome, without the disadvantages. Nevertheless, additional investigations, involving a large sample population, are warranted to confirm the conclusion of this study.

Probabilistic power flow (PPF) calculation is an important power system analysis tool considering the increasing uncertainties. However, existing calculation methods cannot simultaneously achieve high precision and fast calculation, which limits the practical application of the PPF. This article designs a specific architecture of the extreme learning machine (ELM) in a model-driven pattern to extract the power flow features and therefore accelerate the calculation of PPF. ELM is selected because of the unique characteristics of fast training and less intervention. The key challenge is that the learning capability of the ELM for extracting complex features is limited compared with deep neural networks. In this article, we use the physical properties of the power flow model to assist the learning process. To reduce the learning complexity of the power flow features, the feature decomposition and nonlinearity reduction method is proposed to extract the features of the power flow model. An enhanced ELM network architecture is designed. An optimization model for the hidden node parameters is established to improve the learning performance. Based on the proposed model-driven ELM architecture, a fast and accurate PPF calculation method is proposed. The simulations on the IEEE 57-bus and Polish 2383-bus systems demonstrate the effectiveness of the proposed method.

This paper proposes a data-driven approach for optimal power flow (OPF) based on the stacked extreme learning machine (SELM) framework. SELM has a fast training speed and does not require the time-consuming parameter tuning process compared with the deep learning algorithms. However, the direct application of SELM for OPF is not tractable due to the complicated relationship between the system operating status and the OPF solutions. To this end, a data-driven OPF regression framework is developed that decomposes the OPF model features into three stages. This not only reduces the learning complexity but also helps correct the learning bias. A sample pre-classification strategy based on active constraint identification is also developed to achieve enhanced feature attractions. Numerical results carried out on IEEE and Polish benchmark systems demonstrate that the proposed method outperforms other alternatives. It is also shown that the proposed method can be easily extended to address different test systems by adjusting only a few hyperparameters.

Compared with the conventional frame-based image sensor, Dynamic Visual Sensor (DVS) only sensitive to the intensity changes in the scene, and then serially send out these changes in the form of Activity Event (AE). Taking advantage of real-time, low redundancy and small data volume of DVS, we exploit a novel method for Unmanned Aerial Vehicle (UAV) vision-aid landing. In our method, discrete AEs within a varying period are accumulated to form an event map at first. Then the template matching technology is employed to locate the landing mark. Based on the recognized landing mark, local optical flow estimation and real speed projection are conducted to provide more aux information for accurate landing. In the experiments, a simulation platform is setup verify the effectiveness of our methods under various external environments. The experimental results show that the proposed algorithm can accurately measure the location and relative speed of landing marks.