判别类条件贝叶斯网络分类器的量子粒子群优化参数学习

Quantum-behaved particle swarm optimizing parameter learning in class-condition Bayesian network

针对贝叶斯网络分类器在处理多属性分类问题时,存在分类精度下降、算法运行时间过长等问题,提出一种判别类条件贝叶斯网络模型。该模型在类条件贝叶斯模型的基础上,将条件对数似然函数以对数形式重新参数化,并使用量子粒子群优化算法最大化目标函数。新模型采用判别参数学习方法,直接计算条件概率,对于分类问题更加高效。本研究将判别类条件贝叶斯网络模型与TAN分类器相结合,使用量子粒子群算法进行优化,用于对液体火箭发动机的故障诊断与分类中。针对某型号火箭的仿真数据进行故障诊断与分类,与传统的贝叶斯分类器相比,改进的分类器在处理分类问题时,准确率和学习效率更高。

When dealing with multi-attribute classification problem,the classification accuracy of Bayesian network classifier is degraded and running time of the algorithm is very long.A discriminative conditional Bayesian network model is proposed.Based on the class-condition Bayesian,the model re-parameterizes the conditional log-likelihood function in logarithmic form and uses the quantum-behaved particle swarm optimization algorithm to maximize the objective function.The new model uses the discriminative parameter learning approaches to calculate the conditional probability directly,which is more efficient for the classification problem.In this paper,the discriminative parameter learning method of Bayesian network classifier is combined with the TAN classifier for fault diagnosis and classification of liquid rocket engine by optimizing with quantum-behaved particle swarm is presented.The fault diagnosis and classifications on the simulation data of a certain type of rocket are carried out.Compared with traditional classifiers,the improved classifier has higher accuracy and learning efficiency.