Research on SOM-DBN Based Fault Early Warning System For Dispatching Automation

With the development of computer and network communication technology, substation comprehensive automation system has been widely applied in power network, thus power dispatching center can collect and monitor a lot of power network operation information. How to draw useful diagnosis rules from history and real time operational state data of power network, how to use these rules to find important information and exceptional situation from power network real time state data and give early warning of potential fault in time and correctly, become a significant and urgent research problem of dispatching automation system. This paper presents a framework of SOM-DBN based fault early warning system for dispatching automation, including data monitoring, diagnosis and warning, rule drawing, management control center, memory unit. Data monitoring module is used to monitor real time operational data of power network; diagnosis and warning module is designed to build diagnosis model, identify important information and exceptional situation, and send out warning information according to diagnosis rules; rule drawing module uses data mining methods find potential knowledge and rules from history and real time data; management control center as the system backbone aims at the whole management and cooperation; storage unit, composed of database, knowledge-base and rule-base, is used to store operational state data, expert domain knowledge and diagnosis rules. The proposed system can monitor power network state and realize early fault warning through two stages. In the first stage, it constructs fault diagnosis model for power network. Firstly, it clusters history data by SOM and finds clustered feature variables; secondly, through rule drawing module, it can identify fault zone and work out corresponding measures according to clustered feature variables combined with expert domain knowledge; then it uses LVQ algorithm to train SOM network model; next it uses the fault zone sample data to train diagnosis Bayesian network model and learns model parameters. In the second stage, the system can monitor power network state and realize fault diagnosis and warning. Through data monitoring module, it monitors the environment state variables and determines the need for situation assessment; through diagnosis and warning module, it propagates event cues, projects events, diagnoses situation, warns fault and prorides decision-marking. The SOM-DBN based fault early warning system has been applied into dispatching automation system ON2000, and effectively improved the diagnosis and warning capability of dispatching automation system.