針對將圓錐體壓入薄壁筒的控制技術(shù)中， 對位移傳感器和壓力傳感器在線實時檢測及執(zhí)行機(jī)構(gòu)氣液增力缸的控制問題， 在分析受外界干擾的情況下， 為保證系統(tǒng)的穩(wěn)定性和收斂性， 采用基于遞推最小二乘法的多層前向神經(jīng)網(wǎng)絡(luò)算法， 并對它的條件參數(shù)進(jìn)行了一般性的推導(dǎo)。該神經(jīng)網(wǎng)絡(luò)為3 層前向網(wǎng)絡(luò)， 其隱含層神經(jīng)元數(shù)為10 個， 輸出層神經(jīng)元數(shù)1 個。包括輸出層與隱含層、輸入層與隱含層間權(quán)的調(diào)整算法。用典型PID 算法和神經(jīng)網(wǎng)絡(luò)的控制算法分別仿真結(jié)果可見， 神經(jīng)網(wǎng)絡(luò)控制解決了壓合過程純滯后的非線性問題， 達(dá)到了預(yù)期的技術(shù)要求。
關(guān) 鍵 詞 ： 最小二乘法； 神經(jīng)網(wǎng)絡(luò)；PID 算法
Abstract: Aiming at a series of control problems about putting a cone into thin canister, online and real time test of displacement sensor and pressure sensor, and the control of increasing pressure jar as executive machine, the multi-layer front neural network algorithm based on least square method (LSM) was applied to ensure stability and astringency of the system under the conditions of he system disturbed by the environment, and its some of condition parameters were deduced. There are 3 layers in multi-layer front neural network, there are 10 nodes in hide layer, and is 1 node in the output layer. The network contains adjusted algorithm of the weight between output and hide layer, and between input and hide layer. The emulation results of typical PID algorithm and neural network control algorithm show that the nonlinear problem for pressed process lag was resolved with the neural network control, the expected technical requirement was achieved.
Key Words: LSM; Neural network; PID algorithm