In order to analyze the influence of uncertain factors on power system, PCA (polynomial chaos approximation) method, which is both fast and accurate, is widely used in probabilistic power flow calculation. The polynomial chaotic approximation method requires that the probability density function of the random input variable is known, and the random input variable must satisfy the independent condition. A probabilistic power flow method based on DDPCA (data driven polynomial chaos approximation) was proposed for the known random input variables which are historical data. First, DDPCA selects the optimal orthogonal polynomial according to the historical data, and then determines the Gaussian sample considering the nonlinear correlation of random input variables, and then computes the weights with Monte Carlo integral. Then, a small amount of power flow was calculated based on Gaussian samples, and the approximation coefficient was solved according to the power flow results and weights, and then the statistical characteristics of the random output variables were obtained. The proposed method was compared with the point estimation method, and the effectiveness of the proposed method was verified by the results of three examples.

To solve the problem of large steady-state error and poor parameter robustness of MPFC system predicted by traditional model of PMSM (permanent magnet synchronous motor), a multi-voltage vector selection method based on stator flux prediction error vector analysis was proposed. Firstly, the multi-voltage vector selection criteria for determining the region where the flux error vector is located were established by dividing the sectors according to the axis in the two-phase stationary coordinate system. Then, the predicted value of stator flux and the value function of stator flux in two-phase stationary coordinate system were used to calculate the action time of each voltage vector. In addition, a discrete sliding mode stator flux observer considering the mismatch of resistance and inductance parameters was designed, which further improves the parameter robustness of the system. Finally, the effectiveness and feasibility of the proposed predictive stator flux control method are verified by simulation and experiments. The proposed method still has good steady-state performance under the condition of system parameter mismatch, and significantly reduces the stator flux and electromagnetic torque ripple.

With the increasing penetration of distributed power sources in the distribution network, active distribution has become the mainstream direction of the power grid in the future. Microgrids and active stations that are slightly smaller than their scale will gradually increase. Whether these subsystems can be used as energy to schedule becomes the key to improving the economy and stability of power grid operation. Therefore, aiming at the overall optimal scheduling of microgrid and active station area with distributed power supply, a scheduling model solution method based on improved target cascade method was proposed, which mainly takes the optimal benefits generated by different stakeholders as the final scheduling target, and adopts the opportunity constraint description for the processing of uncertain factors of wind and solar. The upper layer is the distribution network and the optimal target of the distribution network, and the lower layer is the microgrid and active station area with the ability to participate in the scheduling. Based on the modeling of distribution network and microgrid, the improved target cascade method was introduced, and the interactive power was used as a shared variable to equivalent the generator and virtual load, so as to realize the decoupling and independent optimization of the upper and lower layers. The comparison of the experimental results shows that the target cascade method with the balance coefficient can obtain better results in the number of iterations, convergence performance, anti-interference performance and overall economic evaluation.

To address the voltage fluctuations caused by the integration of renewable energy into the distribution network, a voltage coordinated control method based on MPC (model predictive control) was proposed to ensure safe operation. Aiming at the uncertainty of wind and photovoltaic output, the AP-K-Medoids clustering algorithm was proposed to generate and reduce output scenarios, and a model was established with the optimization objective of minimizing system network loss. Adopting multi time scale voltage control through the integration of on load voltage regulating transformers, capacitor banks, static reactive power compensators, wind solar reactive power output, and energy storage charging and discharging coordination. Long term scale optimization control solves the output of each device in the system through multi-step rolling optimization, with wind solar output and load demand prediction as the premise. On the basis of short-term and long-term rolling optimization, the increment of output is solved. The optimization control model is a non convex and nonlinear model, which utilizes a second-order cone programming model to solve nonlinear problems. Using an improved IEEE33 node distribution network system for case analysis, the research results demonstrate the feasibility of the proposed voltage optimization control strategy.

Aiming at the technical problems of "untimely perception, poor transmission and difficult equipment deployment" in the monitoring and early warning of mountain disasters in the complex environment of the Qinghai-Tibetan Plateau, a UAV-throwing monitoring device, LoRa networking and edge computing gateway, as well as other embedded hardware and software equipment applicable to deformation and micro-motion monitoring of high-level and high-risk mountain disasters were developed, and focused on the research of the system low-power adaptive data acquisition algorithm and RF frequency adaptive technology, were developed the self-organised network routing algorithm based on LoRa and Beidou RDSS, as well as the multimodal communication intelligent switching technology, so as to solve the problems of timeliness of data perception in complex environments and the problems of low-power consumption and environmental adaptability. The results show that the developed system had a good on-site pilot application effect, which meeting the requirements for long-term monitoring of mountain disasters in alpine mountainous areas, and the average packet loss rate of data transmission in extreme environments is 2.328 8 percent, providing new technologies and methods for disaster prevention and mitigation in the construction and operation of major projects in alpine and complex mountainous areas.

Aiming at the problems that the large-scale pre-training language model faces when dealing with news headlines, such as huge parameters, inefficient use of contextual semantic features and circular convolution neural network’s neglect of the importance of initial input elements, a news headline classification method that combines ERNIE(enhanced representation through knowledge integration) of mixture-of-expert model and recurrent convolution neural network with attention mechanism were proposed. Firstly, the text was encoded with the help of MoE’s improved ERNIE technology, and then the text was classified with attention RCNN (recurrent convolutional neural networks)on the basis of preserving the word order and characteristics of the text. In order to improve the classification ability, RCNN was improved by calculating the input fusion context weight. In the process of calculating the weights of experts in MoE, Gumbel-Softmax was selected as a new gating function to improve the traditional Softmax function, so as to better control the smoothness. According to the experimental results, it is found that compared with the traditional classification methods, the classification method proposed in this study shows significant advantages and greatly reduces the number of parameters. On this basis, the F₁ value is increased by 0.51% compared with the traditional model. After the ablation experiment, the feasibility of this classification method in the classification task has been confirmed.

Ship targets in remote sensing images have multi-scale characteristics, changeable backgrounds, and complex meteorological characteristics, which lead to low accuracy, false detection, and missed detection of small target ships. In response to the above situation, an improved small-target ship detection model based on YOLOv5s was proposed. First, in order to solve the problems of scale changes and background variability in ship detection, the ASFF(adaptive spatial feature fusion) module was introduced. Secondly, in order to reduce the calculation amount and parameter amount of the detection network, the BoTNet attention mechanism was introduced, and then in order to improve the overall network to improve the detection accuracy, the EIoU border loss function was used, and finally the Slim-neck network was introduced to ensure the overall lightweight of the network. Experiments show that on the main data set LEVIR-Ship, compared with the benchmark YOLOv5s, mAP@0.5 increased by 7.1% to 81.3%, the number of parameters is reduced by 0.44 M, the calculation amount is reduced by 0.6GFLOPs, and the weight was reduced by 0.9 M. The proposed method performs better in various key indicators and achieves high-precision small target ship detection in complex environments. Comparative experiments are conducted on the verification data set McShips. The experiments show that the proposed method still performs better, verifying the universal applicability of the proposed method.

BCI (brain computer interface) is one of the important research methods in the fields of brain cognition, brain medicine and brain-like research,where the precise implantation of microelectrodes is a necessary prerequisite and important guarantee. With the rapid development of robotics, machine vision and artificial intelligence, surgical robots are gradually used in brain-computer interface implantation surgery. To meet the demand for precise implantation of microelectrodes in the somatosensory and cerebral motor cortex of SD(Sprague-Dawley) rats, a vision-guided precision implantation system for brain-computer interface microelectrodes was presented. Based on the method of machine vision to identify the key points of the rat skull, the coordinate system was established to obtain the point cloud information of the rat skull, to realize the high-precision identification and localization of the target points, and to guide the actuator to complete the electrode implantation operation. Through model analysis and animal experiments, it has been demonstrated that the implantation system can accurately identify surgical targets on the subjects, guide the actuator to swiftly penetrate the skull, and accurately and stably implant the electrodes into the target area, which effectively improves the accuracy of microelectrode implantation.

In order to improve the accuracy and efficiency of inventory counting in the process of monitoring and auditing biological assets, a biological asset detection model YOLOSC incorporating the attention mechanism and loss function optimization was proposed. Firstly, the SENet attention mechanism was introduced into the backbone network of the YOLOv5s model to enhance the ability of extracting the key features in the pictures of the biological assets. Secondly, the CIoU was adopted as the regression of the detection frames with the loss function to enhance the regression speed and localization accuracy of the detection frame during the training process. Finally, a biological asset datasets was constructed for targeted training of the proposed model to enhance the model detection effect. The experimental results show that compared with the YOLOv5model, the precision, recall, F₁ value and AP of YOLOSC are improved by 2.3%, 2.1%, 2.7% and 1.6%, respectively, which proves the effectiveness of the proposed biological asset detection model YOLOSC.

In response to the low detection accuracy and high model complexity of existing road damage detection algorithms in complex environments, a lightweight road damage detection algorithm named LDC-YOLOv5 (lightweight deformable convolution YOLOv5) was proposed based on YOLOv5.To address the complexity of real road surface damages, a lightweight feature extraction module was designed using Deformable Conv (deformable convolution) and Depthwise Conv (depthwise convolution) to replace the C3 module in the original network backbone, enabling convolutional kernels to focus on irregular crack damages and enhancing feature extraction for damage detection. To reduce algorithm complexity in the feature fusion stage, a lightweight feature fusion module was constructed using GhostConv to replace the C3 module in the original network neck, lowering network parameters and complexity. Additionally, to prevent missed detections caused by uneven lighting and shadow obstruction, a lightweight attention mechanism, TripletAttention, was introduced in the backbone network to improve the algorithm's understanding of damage information and context. Experiments conducted on the IEEE open dataset RDD2022 and the Kaggle open dataset Road Damage demonstrate that, compared to YOLOv5s, the proposed LDC-YOLOv5 achieves a 1.4% and 4.2% improvement in mAP50 on the two datasets, respectively, with only 67.6% of the model parameters of YOLOv5s.