关键词: 区间二型模糊神经网络, 差分进化算法, 梯度下降算法, 软测量, 淀粉利用率

Abstract: Aiming at the problem that some key parameters of complex industrial processes unable to be detected effectively and online in real-time,a soft sensor modeling method based on Interval Type-2 Fuzzy Neural Network(IT2FNN)optimized by Multi-Strategy and Adaptive Differential Evolution algorithm(MSADE)was proposed.To solve the problem that the Differential Evolution(DE)algorithm used a single strategy,fixed scaling factor and crossover probability that leaded to the lack of late search ability,an MSADE algorithm was proposed.In this algorithm,the Root Mean Square Error(RMSE)of the IT2FNN model was used as fitness function to determine the structure(number of rules)and initial parameters of the IT2FNN by searching for the RMSE values under different numbers of rules.Then,the parameters of the IT2FNN model were learned by the gradient descent method.The proposed model was applied to the prediction of Mackey-Glass chaotic time series and the soft sensor of starch utilization rate in wine making process.The simulation results verified the effectiveness and superiority of the proposed method.

Key words: interval type-2 fuzzy neural network, multi-strategy and adaptive differential evolution algorithm, gradient descent algorithm, soft sensor, starch utilization rate