Danhong Zhang

Accurate wind power forecasting is essential for the reliable and efficient operation of renewable energy–dominated power systems, as it directly affects power system scheduling and grid stability. Nevertheless, the inherent intermittency, strong nonlinearity, and stochastic characteristics of wind power generation pose significant challenges to short-term forecasting accuracy. To tackle these issues, this paper develops a hybrid forecasting framework that integrates Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Variational Mode Decomposition (VMD), and a deep learning predictor based on Convolutional Neural Networks and Long Short-Term Memory networks (CNN–LSTM). Specifically, CEEMDAN is first employed to decompose the original wind power time series into multiple intrinsic mode functions at different temporal scales. Subsequently, VMD is applied to the high-frequency components to further suppress noise and enhance signal stationarity. Each decomposed subsequence is then modeled and predicted using a CNN–LSTM architecture, and the final wind power forecast is obtained through reconstruction of all predicted components. Extensive experiments conducted on real-world wind farm data that the proposed hybrid model consistently outperforms several benchmark methods in terms of forecasting accuracy, thereby verifying its effectiveness and practical applicability for short-term wind power forecasting.

Driven by the energy transition and the dual carbon goals, the power system is accelerating its evolution toward a structure dominated by new energy sources. The high proportion of wind and solar integration introduces significant randomness and volatility, substantially increasing the challenges of system balancing and secure operation. Establishing a multi-energy complementary system integrating wind, solar, hydrogen, and storage can effectively address the challenge of renewable energy integration. First, a multi-energy complementary system model is constructed, proposing a day-ahead optimization dispatch strategy. Building upon this, a composite cost function—incorporating grid interaction costs, fuel costs, and resource costs—is formulated as the objective function, while power balance constraints and equipment operation constraints serve as the objective conditions. Finally, the proposed day-ahead dispatch strategy undergoes systematic validation through case studies, confirming the feasibility of the methodology.

Neural radiance fields (NeRFs) leverage a neural representation to encode scenes, obtaining photorealistic rendering of novel views. However, NeRF has notable limitations. A significant drawback is that it does not capture surface geometry and only renders the object surface colors. Furthermore, the training of NeRF is exceedingly time-consuming. We propose Depth-NeRF as a solution to these issues. Specifically, our approach employs a fast depth completion algorithm to denoise and complete the depth maps generated by RGB-D cameras. These improved depth maps guide the sampling points of NeRF to be distributed closer to the scene’s surface, benefiting from dense depth information. Furthermore, we have optimized the network structure of NeRF and integrated depth information to constrain the optimization process, ensuring that the termination distribution of the ray is consistent with the scene’s geometry. Compared to NeRF, our method accelerates the training speed by 18%, and the rendered images achieve a higher PSNR than those obtained by mainstream methods. Additionally, there is a significant reduction in RMSE between the rendered scene depth and the ground truth depth, which indicates that our method can better capture the geometric information of the scene. With these improvements, we can train the NeRF model more efficiently and achieve more accurate rendering results.

Neural radiance fields (NeRF) leverage a neural representation to encode scenes, obtaining photo-realistic rendering of novel views. However, NeRF has notable limitations. A significant drawback is that it does not capture surface geometry and only renders the object surface colors. Furthermore, the training of NeRF is exceedingly time-consuming. We propose Depth-NeRF as a solution to these issues. Specifically, our approach employs a fast depth completion algorithm to denoise and complete the depth maps generated by RGB-D cameras. These improved depth maps guide the sampling points of NeRF to be distributed closer to the scene's surface, benefiting from dense depth information. Furthermore, we have optimized the network structure of NeRF and integrated depth information to constrain the optimization process, ensuring that the termination distribution of the ray is consistent with the scene’s geometry. Compared to NeRF, our method accelerates the training speed by 18% and significantly reduces the RMSE between the rendered scene depth and the ground truth depth, which indicates that our method can better capture the geometric information of the scene. With these improvements, we can train the NeRF model more efficiently and achieve more accurate rendering results.

Under factor graph-based visual-inertial odometry (VIO), is usually used to limit the continuously increasing numbers of nodes (states) and edges (observations). However, this approach will destroy the original sparse structure of graph optimization. To solve this problem, we propose a paralleled online sparsification method in the fixed-lag smoothing VIO by minimizing Kullback–Leibler divergence (KLD) to recover nonlinear and sparse factors that optimally approximate the original dense prior induced by marginalization. In addition, we propose a new graph formulation capable of simplifying partial VIO into VO (visual odometry) with a gravity-aligned pose prior. In addition, a novel method is presented in the loop-closure phase and mapping, in which the recovered factors and the observation of landmarks can be reused for more consistent back-end optimization. The proposed method is validated based on real datasets and compared with other state-of-the-art methods to verify its efficiency and accuracy. The experimental results demonstrate that the proposed sparse method effectively reduces computational complexity when compared to the traditional tightly coupled VINS system. Furthermore, the proposed method offers a novel perspective for achieving global consistent mapping, making it an innovative addition to existing advanced systems employing sparse steps.

Submarines are the main underwater combat force in our country. In order to reduce the possibility of submarines being discovered by airborne magnetic detectors and magnetic mines when operating underwater, this paper proposes a parameter adaptive particle swarm algorithm to optimize the current of the degaussing coil of the submarine degaussing system. Using the two test functions of Ackley and Rastrigin to test the improved algorithm, it can be seen that the optimization accuracy of the parameter adaptive particle swarm optimization algorithm has been improved by 2-4 orders of magnitude. Applying the improved algorithm to the single-component and multi-component submarine degaussing current optimization model, the degaussing rate of the induced magnetic field of each component of the submarine has been significantly improved, the degaussing rate of the X component has increased by 9.56%, the degaussing rate of the Y component has increased by 8.76% and Z component degaussing rate increased by 5.53%.