This study presents a microbial activation strategy to repurpose coal gangue (CG) for enhancing soil quality in aeolian sandy soil. Phosphate-solubilizing bacteria (20-3) and nitrogen-fixing bacteria (SGD06) were employed to produce two microbially activated CGs (CG20-3 and CGSGD06). Adsorption experiments elucidated their colonization mechanisms on CG, while seed germination and pot trials with pakchoi validated remediation efficacy. Even after pakchoi harvest, both CG20-3 and CGSGD06 significantly increased soil bioavailable nutrients by 1.13-6.86-fold. CGSGD06 exhibited superior nitrogen delivery through biological fixation, providing 82.39% higher available nitrogen than CG20-3. Conversely, CG20-3 showed faster adsorption kinetics and stronger binding capacity. Both strains mediated nutrient liberation via acid-driven dissolution of CG clay minerals, including kaolinite, muscovite, and phosphates. Treated soils enhanced pakchoi germination (3.08%) and biomass (1.14-1.89-fold) through improved plant resistance and biochemical indicators, namely ascorbic acid, chlorophyll, and soluble sugar. The activated CGs reshaped the rhizosphere bacterial community and stimulated the secretion of organic acids, phytohormones, and key enzymes, collectively establishing a synergistic growthpromoting system. Additionally, rhizospheric micronutrient bioavailability increased 4.52-43.15-fold, and key soil enzyme activities were upregulated 1.42-267-fold. This work provides a preliminary basis for using microbially activated CG as a potential amendment for aeolian sandy soil, highlighting opportunities for solid waste valorization and soil restoration under controlled conditions.
The high cost of the power collection system (PCS) remains a notable challenge for constructing floating offshore wind farms (FOWFs). This letter introduces a new type of device called the current aggregator in the topology design of the PCS for FOWF, aiming to reduce the use of expensive dynamic cables and their additional components. Firstly, considering the cost components of the current aggregator from multiple perspectives, its life-cycle cost model is established. On this basis, the life-cycle cost model for the PCS system integrating the current aggregator is built. Then, a general solving scheme is presented. Finally, the proposed PCS design is implemented on an irregular 21 FOWF and a regular 6 & times; 6 FOWF, and its economic feasibility is preliminarily verified, achieving cost reduction for irregular and regular FOWFs. The letter may provide a practical new way of designing the PCS of FOWFs, which could reduce the high cost.
Graph Neural Networks (GNNs) often struggle with node classification under class imbalance because standard methods fail to adequately leverage graph topology. Specifically, majority nodes dominate message passing, causing the aggregated information from minority nodes to become statistically diluted, which masks relational signals associated with under-represented classes. Traditional techniques such as oversampling and cost-sensitive learning mitigate data skew but typically neglect structural dependencies, leading to under-representation of minority classes or overfitting to local graph substructures. We propose CL3E-GNN, a novel curriculum learning framework that addresses class imbalance through three synergistic phases: Engage, Enact, and Embed. The Engage phase employs multi-head attention to identify critical node and edge connections, generating enhanced feature representations via dual graph convolution network (GCN) and graph attention network (GAT) aggregation. The Enact phase refines minority class representations through three complementary mechanisms, each targeting a distinct source of imbalance: importance sampling for active feature learning, metapath propagation for knowledge transfer across heterogeneous relations, and anomaly detection for identifying structural irregularities. The Embed phase introduces a memory-augmented reinforcement learning mechanism that progressively refines node embeddings via reward-modulated updates. A dual-loss optimisation strategy combining cross-entropy with adaptive curriculum loss ensures gradual difficulty progression, offering a principled alternative to simple loss functions or data augmentation alone. Extensive evaluations on eight benchmark datasets, with results averaged over five independent runs, demonstrate consistent improvements over state-of-the-art methods. These results establish CL3E-GNN as a new benchmark for imbalanced node classification, effectively preserving topological integrity while mitigating class distribution skew. The implementation code is available at https://github.com/afofanah/CL3E-GNN.
Capturing the rich, hierarchical information embedded in complex multivariate time series (MTS) necessitates modeling at multiple temporal scales. However, prevalent multi-scale forecasting methods suffer from critical limitations: a uniform scaling approach that overlooks individual variable characteristics, inconsistent modeling of temporal patterns and inter-variable dependency leading to modeling spurious correlations, and a neglect of global context. To overcome these challenges, we introduce LMSNet, a Lightweight Multi-Scale graph learning framework. Specifically, LMSNet introduces a novel variable-specific scaling mechanism, which leverages frequency-domain analysis to assign a tailored analytical window to each variable based on its intrinsic dynamics, establishing a robust foundation for accurately modeling multi-scale inter-variable dependencies. Furthermore, LMSNet synergizes global and local information and performs concurrent temporal and inter-variable dependency modeling at every network layer, aided by a momentum update mechanism that refines deep-layer graph node representations to mitigate the risk of modeling spurious correlations. LMSNet establishes a new state-of-the-art on eight benchmark datasets. Crucially, it achieves a 2.94% reduction in MAE with reducing GPU memory usage by 83% and accelerating training by 60% compared to strong baseline-TimeMixer. Our source code is provided at https://github.com/05Pikachu24/TSF-LMSNet.
