Abstract: The accurate forecasting of daily natural gas load is pivotal to the reasonable supply and dispatch of energy in the city. This study proposes a multistage hybrid model based on Fuzzy Coding of Genetic Algorithms (FCGA) and the improved LSTM-BPNN residual correction model since gas load data is periodic but random and the single-stage and single-forecasting model has a limited role. In the first stage, the gas load is forecasted by the LSTM model to calculate its residual value. In the second stage, the residual value is predicted by the BPNN model, and then the learning rate of the LSTM-BPNN residual model is automatically adjusted with the Adam algorithm regarding the adaptive learning rate to accelerate fitting. Afterward, the initial weights and thresholds of the BPNN are optimized by the fuzzy coding of genetic algorithms to find the global optimal solution. Finally, the sum of the forecasting values in the two stages is taken as the final gas load. Comparative experiments prove that the model in this study ensures higher prediction accuracy than the single model and the original two-stage forecasting model.
Abstract: During the operation of the power grid, key devices such as converter valves continue to generate heat, affecting the stability and safety of the system. Then it is crucial to ensure the stable operation of those devices. As a major component in the cooling system, the valve cooling system releases the heat energy from the equipment with water of high thermal conductivity as the medium. The stable operation of the converter valve can be ensured by monitoring the temperature and pressure of the cooling water. Also, with inlet valve temperature in the valve cooling system as the main predictive index, the historical data of the system is fully mined for predicting the operating state of the power grid. An ARIMA-SVM hybrid model integrating the traditional time series model and machine learning is compared with the traditional ARIMA model, the SVM model and the GRU neural network model with regard to the time series analysis of the real valve cooling data from China Southern Power Grid. The comparative experimental results demonstrate that the above four models can all clearly indicate the trend of the inlet valve temperature. However, the ARIMA-SVM hybrid model behaves better in the evaluation indicators including the root mean square error, the mean square error and the mean absolute error than the other three, with a more accurate prediction of inlet valve temperature.
Abstract: In recent years, driven by the progress in artificial intelligence, deep learning models have been widely applied to ECG data analysis (especially the detection of atrial fibrillation). This study proposes an algorithm based on the multi-head attention mechanism to classify atrial fibrillation, which is trained and validated through the public data set of the PhysioNet 2017 Challenge. Firstly, the local features of the ECG signal are extracted through the deep residual network. Then, the bidirectional long short-term memory network is built to extract the global features on this basis. Finally, the multi-head attention mechanism layer is used to extract the key features, and cascade modules greatly improve the performance of the overall model. The experimental results show that the proposed heads-8 model can achieve precision of 0.861, recall of 0.862, F1 score of 0.861, and accuracy of 0.860, which is better than the latest methods based on ECG signals for classifying atrial fibrillation.
Abstract: The diverse and complex trend of public opinion has long made it difficult to manage. Negative public opinion will intensify contradictions, bringing adverse effects to social stability. Then a method of public opinion deduction based on the event knowledge graph is proposed. The causal logic of the event is mined through the neural network, and the event knowledge graph is drawn after causal events are connected. Vectorized event nodes can merge into similar nodes to reduce map redundancy while enhancing map generalization. Besides, the evolution of target public opinion events can be predicted based on the deductive logic indicated in the event knowledge graph. With a public opinion event related to a natural disaster as an example, the experimental results prove that the proposed method can reliably predict the trend of the event, supporting public opinion supervision.
Abstract: The abuse of Unmanned Aerial Vehicles (UAVs) brings great security risks to the low altitude area. Then the research on detection of UAVs’ illegal intrusion has become important for a low-altitude defense system. In this study, a multi-sensor information fusion technique based on radar and a RGB camera is designed to detect small objects in the low altitude range. After that, the Single Shot multibox Detector (SSD) for deep learning is introduced to train the UAV detection model and predict the category and location of objects captured by the RGB camera. An experimental platform is built to verify that the information fusion method can collect the location, speed, appearance of targets, and the deep learning model can determine the categories of suspicious targets.
Abstract: The original U-Net integrates a jumping structure with high-level and low-level image information, which makes the U-Net model perform well in segmentation, but the results still present poor segmentation, over-segmentation, and under-segmentation at the edges of cervical nucleus. Then an improved U-Net network for image segmentation is proposed. First, the densely connected DenseNet is introduced into the encoder of U-Net to solve the problem that the encoder is too simple to extract abstract high-level semantic features. Then different weights are given to the cervical nucleus nuclei and background in the binary cross-entropy loss function, so that the network pays more attention to the learning of nuclear characteristics. Finally, during the pooling operation, reasonable weights are assigned to the pixel values in the pooling domain to avoid losing information in the pooling layer. Experimental results reveal that the improved U-Net network can behave better in cervical cell segmentation with a more robust model, and the proportions of over-segmentation and under-segmentation are also smaller.
You are the firstVisitors
Copyright: Institute of Software, Chinese Academy of Sciences Beijing ICP No. 05046678-3
Address：4# South Fourth Street, Zhongguancun,Haidian, Beijing,Postal Code：100190
Phone：010-62661041 Fax： Email：csa (a) iscas.ac.cn
Technical Support：Beijing Qinyun Technology Development Co., Ltd.