Power System Fault Diagnosis Based On Rough Set Theory And Bayesian Network

With the continuous development of national economy, demands of the users for power quality are increasing.However, on account of climate, human factors, and so on, the power failure is inevitable. Therefore, when the power system goes wrong, the operators should judge the reasons rapidly, insulate the elements of fault and restore the normal operation of the system, in order to reduce the damage to the electrical elements, and guarantee the secure and reliable electricity supply to users. With the expanding of the power system's scale and the increasing complication of its structure, a lot of alarm informations goes to the dispatching center,far exceeding the capabilities of operators.It is a prerequisite for rapid restoration of power supply that the power system fault diagnosis system can help operators find the faulted element.In this paper, based on studing various artificial intelligence methods in depth which are applied to power system fault diagnosis, we advance a method combining Rough Set and Bayesian Network. Firstly, the signals of protection and circuit breaker are looked as a condition attributes set of fault classification and a decision table is constructed by considering every possible fault, we reduce the condition attributes, and construct a new decision table;We establish Bayesian Network models one element by one element,according to the physical topology of power system and the operation principle of the protection device;Then we reduce the Bayesian Network models based on the new decision table;At last, we accomplish the parameter settings and calculation of the Bayesian Network models using subjective Bayesian method in order to achieve fault diagnosis.The fault information can be greatly simplified by the attribute reduction of rough set theory when the classification ability doesn't change, and the Bayesian Network models can be greatly simplified; Bayesian network and subjective Bayesian method are two uncertainty reasoning methods which are used commonly, Bayesian network is clear, easy to understand and easy to express causal relationship between data, but its parameters are often difficult to obtain. The subjective Bayesian method transforms the knowledge of importation into evaluation of two parameters-LS and LN, which avoids a lot of data statistic work. Based on the combination of the subjective Bayesian approach and Bayesian network, the problem which Bayesian network's parameters are difficult to obtain is solved, and Bayesian network can give full play to its advantages on reasoning;Furthermore, this algorithm has fault tolerance.The fault signals are often incomplete during the fault diagnosis process, as long as the key signal is not lost, we still can achieve a correct diagnosis.In actual power system, the obtained information is often incomplete,but the rational method based on subjective Bayesian approach can calculate by using the information of its former point when the information of the point is losed. It enhances the fault tolerance of the approach.The example shows that the approach has good ability to handle complete fault condition and uncomplete fault condition, simply fault and complex fault, and have good application prospects.