Research and Improvement of K2 Algorithm Based on Topological Sorting

In the complex chemical process, some key process features need to be monitored, such as safety which is the key to a stable development in modern process industries. Due to the increasing difficulties of modem processes, it is becoming more and more difficult in fault tracing and diagnosing. BN (Bayesian Network) is a promising model for tracing and inferring the fault. The traditional method to build a BN in structure learning is the K2 Algorithm. However, this algorithm depends on the given input awfully. To solve this problem, this paper put forward an improved algorithm to develop the performance in building a BN structure. The main idea of this proposed method is topological sorting. The proposed algorithm is called a topological-sorting based K2 Algorithm (TS-K2) where the algorithm inputs are improved. One of the inputs is the maximal size of parent node which is the algorithm termination condition. The other input is the node order in the searching space, which will be the key feature of the K2 Algorithm. It determines the quality of the algorithm output accuracy. The ASIA Network is used to verify the performance of the proposed algorithm. The simulation result shows that TS-K2 Algorithm has better performance than the traditional K2 Algorithm.