The blade mass imbalance fault is one of the common faults of the DFIG (Doubly-Fed Induction Generator) wind turbines (WTs). In this paper, considering the spatiotemporal distribution of natural wind speed and the influence of wind shear and tower shadow effect, the influence of blade mass imbalance faults on the electrical characteristics of DFIG WTs is analyzed. Firstly, the analytical expressions and variation characteristics of electromagnetic torque and electromagnetic power under blade mass imbalance are derived before and after consideration of the spatiotemporal distribution of wind speed. Then simulations on the MATLAB/Simulink platform were done to verify the theoretical analysis results. The theoretical analysis and simulation results show that, considering the spatiotemporal distribution of wind speed and the influence of wind shear and tower shadow effect, the blade mass imbalance fault will cause fluctuation at the frequency of 1P (P = the frequency of rotor rotation), 3P, and 6P on electromagnetic power. Fluctuation at 1P is caused by mass imbalance while fluctuation at 3P and 6P are caused by wind speed spatiotemporal distribution; the amplitude of fluctuation at 1P is proportional to the degree of the imbalance fault. Since the equivalent wind speed has been used in this paper instead of the average wind speed, the data is more suitable for the actual operation of the WT in the natural world and can be applied for fault diagnosis in field WT operation.

State estimation is integral to the continuous monitoring and management of the energy consumption in gas turbines. Recent research about multistate prediction adopts separate predictions and uses few connections between signals. The complex physical mechanism of the gas turbine makes it difficult to extract and obtain the spatial information between states as prior knowledge, and the spatial information has irregularities as the operating conditions change. To overcome these limitations, a graph structure with a fusion strategy was adopted to solve the irregularity of multisource sensor data, and then the adaptive adjacency-matrix diffusion graph network (AdapDGN) was proposed. First, an adaptive adjacency matrix module learned through a query and key and based only on the input data can capture the temporal trends of spatial information. Second, the diffusion graph recurrent module can remember the history sequences. The two aforementioned modules are combined in a unified framework, and the entire framework is learned end to end. The performance of AdapDGN for multistate prediction is demonstrated on a real-world sensor network of a gas turbine control system under variable working conditions. The results show a 2% improvement in prediction accuracy of the fusion strategy to the separate prediction, and the learned connections between signals under diverse working conditions are also visualized.

Accurate prediction of multisource sensor states in engine control systems is critical for operational stability and performance optimization. While Transformer architectures have demonstrated exceptional capabilities in long-sequence prediction, conventional temporal attention-based frameworks face dual challenges of excessive parameterization and inadequate spatial relationship modeling across multivariate control signals. This paper introduces a Spatial-Compression Attention Transformer (SCA-Transformer) featuring three architectural innovations to address these limitations. Firstly, a spatiotemporal coupling mechanism that dynamically captures cross-variable dependencies through compressed attention matrices, achieving 90% parameter reduction compared to standard Transformers. Secondly, a channel-adaptive denormalization module enables seamless prediction of heterogeneous sensor signals. Thirdly, a landscape-aware regularization technique enhances model generalization by monitoring loss surface geometry during gradient updates. Extensive experiments on two engine datasets demonstrate the framework’s capability of delivering 460-step predictions with less than 0.05 mean absolute percentage error, outperforming state-of-the-art benchmarks while maintaining 1.95ms/step inference speed. Besides, the effects of different input and output lengths on the prediction results are compared. Finally, the prediction trends of various types of sensors such as temperature, pressure, and speed at different prediction lengths and the spatial attention matrix under diverse working conditions are visualized. The proposed paradigm establishes a new architectural foundation for reliable long-sequence prediction in safety-critical engine systems.