Residential-level short-term load forecasting (STLF) is significant for power system operation. Data-driven forecasting models, especially machine-learning-based models, are sensitive to the amount of data. However, privacy and security concerns raised by supervision departments and users limit the data for sharing. Meanwhile, the limited data from the newly built houses are not sufficient to support building a powerful model. Another problem is that the data from different houses are in a non-identical and independent distribution (non-IID), which makes the general model fail in predicting accurate load for the specific house. Even though we can build a model corresponding to each house, it costs a large computation time. We first propose a federated transfer learning approach applied in STLF, deep federated adaptation (DFA), to deal with the aforementioned problems. This approach adopts the federated learning architecture to train a global model without undermining privacy, and then the model leverage multiple kernel variant of maximum mean discrepancies (MK-MMD) to fine-tune the global model, which makes the model adapted to the specific house’s prediction task. Experimental results on the real residential datasets show that DFA has the best forecasting performance compared with other baseline models and the federated architecture of DFA has a remarkable superiority in computation time. The framework of DFA is extended with alternative transfer learning methods and all of them achieve good performances on STLF.
The Aquila Optimizer (AO) algorithm is a meta-heuristic algorithm with excellent performance, although it may be insufficient or tend to fall into local optima as as the complexity of real-world optimization problems increases. To overcome the shortcomings of AO, we propose an improved Aquila Optimizer algorithm (IAO) which improves the original AO algorithm via three strategies. First, in order to improve the optimization process, we introduce a search control factor (SCF) in which the absolute value decreasing as the iteration progresses, improving the hunting strategies of AO. Second, the random opposition-based learning (ROBL) strategy is added to enhance the algorithm’s exploitation ability. Finally, the Gaussian mutation (GM) strategy is applied to improve the exploration phase. To evaluate the optimization performance, the IAO was estimated on 23 benchmark and CEC2019 test functions. Finally, four real-world engineering problems were used. From the experimental results in comparison with AO and well-known algorithms, the superiority of our proposed IAO is validated.
This paper introduces the election optimization algorithm (EOA), a meta-heuristic approach for engineering optimization problems. Inspired by the democratic electoral system, focusing on the presidential election, EOA emulates the complete election process to optimize solutions. By simulating the presidential election, EOA introduces a novel position-tracking strategy that expands the scope of effectively solvable problems, surpassing conventional human-based algorithms, specifically, the political optimizer. EOA incorporates explicit behaviors observed during elections, including the party nomination and presidential election. During the party nomination, the search space is broadened to avoid local optima by integrating diverse strategies and suggestions from within the party. In the presidential election, adequate population diversity is maintained in later stages through further campaigning between elite candidates elected within the party. To establish a benchmark for comparison, EOA is rigorously assessed against several renowned and widely recognized algorithms in the field of optimization. EOA demonstrates superior performance in terms of average values and standard deviations across the twenty-three standard test functions and CEC2019. Through rigorous statistical analysis using the Wilcoxon rank-sum test at a significance level of 0.05, experimental results indicate that EOA consistently delivers high-quality solutions compared to the other benchmark algorithms. Moreover, the practical applicability of EOA is assessed by solving six complex engineering design problems, demonstrating its effectiveness in real-world scenarios.
9 November 2023 in IEEE Internet of Things Journal
Anomaly detection in energy consumption for buildings plays a critical role in improving energy efficiency. In distributed systems, federated learning (FL) has gained widespread adoption due to its ability to train robust models while preserving privacy. However, most FL algorithms face challenges in handling heterogeneous data across different buildings, where distribution gaps exist. Moreover, addressing the heterogeneity in data necessitates additional efforts in model personalization, resulting in significant computational overhead, especially in large-scale FL settings. To address these challenges, we propose a sparse attentive aggregation-based federated anomaly detection approach (SAA-FAD). SAA-FAD comprises an extraction network and a lightweight recognition network. The extraction network employs a variational Bayes scheme to extract statistical characteristics from the raw data for computing similarities between clients while preserving privacy. Then the recognition network is trained with the extracted statistical features. The model is trained in a sparse attentive aggregation manner, which leads to superior anomaly detection performance on heterogeneous data. Additionally, SAA-FAD achieves a time and memory complexity of O(N∙logN) for computing similarity. Simulation results demonstrate that SAA-FAD outperforms other baseline models, particularly in scenarios with heterogeneous data, while offering a significant advantage in terms of computation time.
21 April 2025 in IEEE Transactions on Instrumentation and Measurement
The precision positioning equipment is essential in many manufacture areas. This paper develops a novel noncontact six-degree-of-freedom maglev planar actuator (MLPA) for applications in precision manufacturing area. To achieve the tracking and positioning of MLPA, a model predictive control (MPC) algorithm is introduced. However, the high frequency feedback control of maglev system up to kilohertz level poses the challenges in the online optimization of MPC. We argue that the field-programmable gate array (FPGA), a high throughput and low latency processor, accelerates the computation of MPC, and the High-Level synthesis (HLS) tool simplifies the implementation of the optimization algorithm, which is the most complex module in MPC. In this work, the dual-gradient projection method is introduced to realize the online optimization with its characteristics of parallelism. Benefiting from the parallel architecture of FPGA, the dual-gradient projection method successfully runs in the millisecond level with the acceleration performance highlighted by comparing with active set and interior point methods. Furthermore, the FPGA-based MPC is deployed for MLPA to inhibit the trajectory tracking error and improve the step response speed. The synthesis of the proposed FPGA-based MPC scheme indicates the prediction horizon of MPC is able to theoretically reach about 20, and the experiments undertaken on the real-time system shows that tracking performance of MLPA is improved with the prediction horizon expanding.
18 February 2026 in Expert Systems with Applications