Aiming at the issues of poor real-time performance in phase picking under high background noise, low accuracy in low signal-to-noise ratio (SNR) data , and limited deployment on edge devices, the paper proposes an edge computing (EC)-based seismic phase picking (SPP) method. Firstly, to achieve efficient data denoising, a dual-domain seismic signal denoising algorithm based on deep neural networks (DNNs) is designed. The time-domain network extracts waveform features for denoising, while the frequency-domain network corrects spectral features. The results from both domains are dynamically and adaptively fused based on real-time SNR estimation. Additionally, to address the resource constraints of edge devices, a lightweight SPP algorithm based on time-domain convolutional networks (TCNs) is designed. It combines the innovative inverted residual blocks (IRBs) with TCN to model temporal dependencies, while also introducing a channel attention mechanism to dynamically enhance key features. Particularly, through structured pruning and mixed-precision quantization, the model is compressed, enabling significantly reduced computational complexity while providing high-precision P-wave and S-wave picking in real-time at the edge. The results demonstrated that the F1 scores of P-wave and S-wave phase picking reached 92.48% and 90.05% before noise reduction. Especially under low SNR data, the F1 scores were significantly improved by 3.42% and 9.34%. The model was deployed to the edge device RK3566, verifying the real-time performance of local data processing and the accuracy of phase picking method.

The microtremor survey method (MSM) holds great potential in the exploration geophysics for characterizing by its lack of dependence on artificial seismic sources, safety and environmental protection. However, the traditional MSM has the defect of serious lag in imaging process, which hinders its development. In the paper, we present an edge microtremor survey instant imaging method (MSIIM), which sets seismic edge servers (SES) to manage seismic sensor nodes (SSN) and perform instant imaging in SESs. Particularly, considering the heterogeneity of edge devices and the limited computing resources, we propose a task allocation method to minimize the total time required for instant imaging and ensure stability. We design a multi-nodes collaborative computing (MNCC) framework to model task allocation as a constraint satisfaction problem. Moreover, the heuristic enhanced genetic algorithm (HEGA) is proposed to solve the task allocation problem under the MNCC framework. Simulations in the EdgeCloudSim show that the HEGA reduces instant imaging total time by 23.36% compared with the traditional Genetic Algorithm (GA). Furthermore, the HEGA maintains 100% task coverage, which underscores the superiority of it in terms of delay optimization effectiveness and stability. The real seismic data test results prove that the MSIIM has excellent imaging ability, and its RMSE can reach 0.085. In addition, the memory footprint of the MSIIM running in resource-constrained edge devices is only 53%, which proves that it has the ability to run well in resource-constrained scenarios.

The large-scale microtremor profiling method (LSMPM) operates under a “distributed data-collection and centralized data-recovery” framework, which has led to significant delays in retrieving shear wave velocity structural information, diminished detection efficacy, and the absence of prompt local computation and processing mechanisms for real-time imaging. Effectively harnessing the computational capabilities of acquisition nodes to facilitate wireless, multinode, low-latency collaborative computing remains a significant challenge. We present a task delay optimization and scheduling algorithm based on modified particle swarm optimization (MPSO) to achieve real-time imaging of shear wave velocity structures at the edge of the sensor network. It enhances the mutation process of the standard particle swarm optimization (PSO), utilizing the delay improvement coefficient as a metric to optimize the delay within computing task queues, thereby avoiding local convergence, augmenting global search capabilities, and consequently reducing the maximum completion time$(\textrm {Makespan})$of tasks. The MPSO algorithm has undergone extensive simulation testing within the CloudSim environment, and the results indicate that, relative to the PSO task scheduling algorithm, the proposed MPSO task scheduling algorithm reduces the standard deviation of the Makespan by approximately 90%, and when the number of edge servers ranges from 10 to 30, the delay improvement coefficient of the proposed MPSO algorithm can surpass 60%, which underscores the superiority of the MPSO in terms of delay optimization efficacy and algorithmic stability.

Surge Protective Devices (SPDs) play a crucial role in Industrial Internet of Things (IIoT) by protecting equipment from lightning-induced overvoltage. However, lightning surge can cause SPD damage, thus interfering with the operation of equipments in IIoT, causing huge losses. Therefore, accurate prediction of SPD RUL is a key technology to ensure the safety of the equipments in IIoT. We conduct degradation testing on SPD to extract degradation parameters and model the degradation curve. Subsequently, we introduce a SPD RUL prediction model utilizing SVM-BaLSTM. Our approach integrates SPD degradation curve with RUL prediction models to enhance predictive accuracy. Furthermore, we establish a real-time monitoring system for SPD to gather degradation parameters, facilitate data input for RUL prediction models. Our experimental results demonstrate a high degree of accuracy, with a Root Mean Square Error (RMSE) of approximately 0.00098, indicating a close alignment between predicted and actual RUL values, and in contrast experiments with other prediction models, our prediction model also shows the best performance.