A5020861976- Time Series Analysis and Forecasting
- Neural Networks and Applications
- Energy Load and Power Forecasting
- Machine Learning and ELM
- Advanced Algorithms and Applications
- Web Data Mining and Analysis
- Semantic Web and Ontologies
- Currency Recognition and Detection
- Text and Document Classification Technologies
- Topic Modeling
- Data Visualization and Analytics
- Rough Sets and Fuzzy Logic
- Advanced Clustering Algorithms Research
- Wireless Sensor Networks and IoT
- Advanced Sensor and Control Systems
- Advanced Computational Techniques and Applications
- IoT-based Smart Home Systems
- Data Quality and Management
- Neural Networks and Reservoir Computing
- Anomaly Detection Techniques and Applications
- Mobile Agent-Based Network Management
- Natural Language Processing Techniques
China Southern Power Grid (China)
2020-2024
Zhejiang University
2021
The Echo State Networks (ESNs) is an efficient recurrent neural network consisting of a randomly generated reservoir (a large number neurons with sparse random connections) and trainable linear layer. It has received widespread attention for its simplicity effectiveness, especially time series prediction tasks. However, there no explicit mechanism in ESNs to capture the inherent multi-scale characteristics series. To this end, we propose model multi-reservoir structure named long-short term...
Accurate electricity consumption forecasting can be treated as a reliable guidance for power production. However, traditional models suffer from simultaneously capturing the periodicity and volatility of sequential data, while are important forecasting. In order to effectively model this data predict accurately, we propose multi-scale prediction (Long Short Term Memory, LSTM) algorithm based on Time-Frequency Variational Autoencoder (TFVAE-LSTM). The proposed treats superposition in...
Named Entity Recognition(NER) is one key step for constructing power domain knowledge graph which increasingly urgent in building smart grid. This paper proposes a new NER model called Att-CNN-BiGRU-CRF consists of the following five layers. A joint feature embedding layer combines character and word based on BERT to obtain more semantic information. convolutional attention local mechanism CNN capture context relationship. BiGRU extracts higher-level features metering text. global multi-head...
In recent years, knowledge graphs are applied to provide support and data for power grid monitoring decision-making. To construct a metering graph, the entities should be effectively recognized extracted. However, existing machine learning models do not fully consider situation that some entities' names partially overlapping boundaries of fuzzy. this paper, we propose Bi-order-Transformer-CRF recognize entities. Specifically, alleviate problem fuzzy entity boundaries, train our word-vectors,...
Multivariate electricity consumption series clustering can reflect trends of power changes in the past time period, which provide reliable guidance for production. However, there are some abnormal multivariate data, while outliers will affect discovery different periods. To address this problem, we propose a robust graph factorization model (RGF-MEC), performs and outlier simultaneously. RGF-MEC first obtains similarity by calculating distance among data then matrix on graph. Meanwhile, is...
Multivariate electricity consumption series clustering can reflect the trend of power changes in past time period, which provide reliable guidance for production. The dimensionality reduction-based method is an effective technology to address this problem, obtains low-dimensional features each variate or all variates multivariate clustering. However, most existing methods ignore joint learning common representations and variable-based representations. In paper, we build a extreme machine...
In the process of constructing knowledge graph in domain electric power metering, often comes from several different data sources, and because differences structure content among there are also problems multiple meanings words unclear referents integration acquired knowledge. Therefore, fusion processing is needed to eliminate ambiguity entities attributes terms denotation. This paper proposes a method based on text matching technology. The includes three core steps. Firstly it use crawler...
With the explosive growth of data on Internet, mining vertical knowledge domain material has become more complex. In order to efficiently collect, manage and exploit huge corpus, this paper proposes a semantic search engine for electric power metering. On one hand overcomes shortcomings traditional general-purpose engines in terms lack targeting specialization, other it higher accuracy stable recall than general keyword-based engines, enabling understanding relational analysis. The main...