【摘要】 针对传统的铝电解异常状态诊断算法结构复杂、CPU占用率高、准确率低等问题,设计了一种布谷鸟支持向量机和深度学习算法级联的异常状态诊断算法。首先运用支持向量机算法处理二分类问题快速准确的特点,将生产状态诊断为正常或异常,并运用新一代布谷鸟算法对其进行优化,加快收敛速度,若诊断结果为正常则直接输出,异常则将数据送入下一级。下一级是由分类效果良好的改进的深度置信神经网络充当,该级接受上一级的信息,对铝电解异常状态类型做全面的分析,诊断出具体的异常状态类型。最后运用Matlab进行实验仿真,结果表明,该算法对计算机CPU的平均使用率远远低于传统算法,对于训练速度,准确率,预报的提前量都有明显的提升。更多还原

【Abstract】 Aiming at problems of traditional aluminum electrolysis abnormal state diagnosis algorithm, such as complex structure, high CPU usage and low accuracy, a cuckoo support vector machine and deep learning cascade abnormal state diagnosis algorithm are designed. Firstly, based on the fast and accurate characteristics of the support vector machine algorithm to deal with the two-category problem, the production status is diagnosed as normal or abnormal, and it is optimized by the new generation cuckoo algorithm to accelerate the convergence speed. If the diagnosis result is normal, the diagnosis result is directly output. If abnormal, the data shall be sent to the next level. The next level is played by an improved deep confidence neural network with good classification effect. This level accepts the information of the upper level, and comprehensively analyzes the type of aluminum electrolysis abnormal state, and diagnoses the specific abnormal state type. Finally, Matlab is used to carry out experimental simulation, the results show that the average usage rate of the algorithm for computer CPU is much lower than the traditional algorithm, while the training speed, accuracy rate and forecast advance value are all obviously improved.更多还原