Situation Awareness Technology Of Power System Operations Based On Multi-source Heterogeneous Data Mining

The power industry is an important basic energy industry of the country’s development.Ensuring the safe and stable operation of power systems is of great significance to the safety of the country.The power system is a complex multi-dimensional nonlinear system,and its safe and reliable operation depends on the “brain” of the power system–the grid dispatch and control system.The core of this system lies in the perception of the operation status and trend of the power grid through data mining and making corresponding control decisions based on situation awareness.However,with the continuous expansion of power grid structure,electric power big data presents the characteristics of massive multi-source heterogeneous.Traditional data mining methods have been difficult to deal with changeable data sources and multiple structural forms.With the advent of data-driven artificial intelligence technology represented by deep learning,great progress has been made in the feature mining of high-dimensional complex data.Thus,how to use artificial intelligence technology to combine multi-source heterogeneous data mining with the demand for situation awareness and decision support in the dispatch and control system has become the key to this article.Therefore,in this article,short-term wind power forecasting,power equipment infrared image recognition and fault detection,and user power consumption behavior prediction and identification are selected as typical situation awareness scenarios on the power generation side,transmission side,and user side,respectively,and develop the corresponding deep learning technologies based on heterogeneous data mining.Besides,how to use the knowledge graph to effectively combine the system operation situation awareness with the decision-making requirements in the power grid dispatch and control system is also explored in this article.The main work and contributions of this thesis are as follows:(1)In view of the volatility and randomness of wind power generation on the power generation side,a multi-step wind power forecasting method using the attention-based gated recurrent unit network is proposed.On the one hand,a sequence-to-sequence-based multi-step wind power forecasting structure is proposed,which can take advantage of the inherent correlations between multi-step prediction tasks and consider a variety of factors that affect wind power forecasting to improve the accuracy and stability of multi-step wind power forecasting;on the other hand,a feature selection method based on attention mechanism is designed to identify the most important factors that affect the accuracy of wind power forecasting under different situations,which could increase the forecasting performance of the proposed method.(2)From the perspective of power equipment infrared image data mining,a power equipment identification and fault detection technology based on Edge-Oriented Generative Adversarial Network is proposed.First,in view of the lack of available data in practical industrial applications and the uneven distribution of data sets,a method for generating infrared images of electrical equipment based on weakly supervised learning using the edge features of electrical equipment objects is proposed.The edge features extracted from infrared image equipment are used as prior knowledge to guide the generation of infrared images,thereby providing effective data augmentation for data-driven artificial intelligence methods.Then,a framework for infrared image recognition and fault detection of power equipment based on deep learning is designed.Combined with the proposed data augmentation technology,the proposed method can intelligently identify power equipment and accurately detect faults.(3)For the grid user side,a feature mining model of electricity consumption data based on convolutional gated recurrent network is proposed.First,for the feature mining of electricity consumption data,the global perception advantages of convolutional neural networks are used to extract the spatial characteristics between different users,while the internal time characteristics of the power consumption time series of users are extracted through the gated recurrent unit network.At the same time,in order to make the training process of the model more specialized and improve the learning ability of the model,the K-means clustering method is introduced to first roughly divide the electricity consumption data into several clusters according to its behavior tendency,and then train the corresponding feature mining model using the divided data.Finally,real electricity consumption data are used to evaluate the effectiveness and superiority of the proposed method in load forecasting and abnormal electricity consumption behavior detection.(4)On the basis of situation awareness,a power knowledge graph framework and its construction method is proposed for power systems fault handling.Based on the proposed knowledge graph,a quick query of fault events,the timely push of disposal suggestions and decision support for fault handling could be achieved for typical failure scenarios of power grid lines and equipment.Finally,the proposed method can combine the advantages of real-time situation awareness of power systems operation to dynamically update the fault handling knowledge graph,which guarantees the inheritance and accumulation of operating knowledge in power systems.