摘 要

在現(xiàn)代機械加工生產(chǎn)過程中，參數(shù)的選擇問題成為困擾設(shè)計者的一個難題，它關(guān)系到如何提高加工精度并直接影響產(chǎn)品性能及成本，這就使得這一問題也成為設(shè)計者積極解決的問題。通過改進人工神經(jīng)網(wǎng)絡(luò)的任意非線性映射能力，逼近誤差復(fù)映系數(shù)與工件材料、進給量等因素之間的非線性關(guān)系，對訓(xùn)練成熟的網(wǎng)絡(luò)輸入加工前毛坯誤差、工件材料硬度等，可以輸出滿足加工要求的加工次數(shù)和各次加工量。
實驗表明：基于人工神經(jīng)網(wǎng)絡(luò)的參數(shù)優(yōu)化選擇模型考慮了工件材料、進給量等因素，能夠作出比較準確的預(yù)測，它的學(xué)習(xí)特性能夠以較好的穩(wěn)定性和精度模擬輸入輸出間的非線性關(guān)系，有很廣泛的發(fā)展前景。




關(guān)鍵詞：參數(shù)優(yōu)化選擇，人工神經(jīng)網(wǎng)絡(luò)，BP網(wǎng)絡(luò)，MATLAB7.0。



ABSTRACT

In modern production machining process, the parameter optimization is problem puzzling the designer, which is related to how to improve the accuracy of processing and has a direct impact on product performance and cost. This issue attracts designers to find a positive solution.

By improving the arbitrary nonlinear mapping capability of artificial neural networks, the nonlinear connections among the error-reflection, workpiece materials, feed rate and other factors are approximated. The input of the trained network includes rough processing error, the workpiece material hardness and so on; the output includes the processing time and the processing volume, which can meet the processing requirements.
Experiments show that: the optimization model based on artificial neural network, which considers the workpiece material, the feed rate and other factors, can make comparatively accurate forecast. Its characteristics ensure that it can simulate the input-output nonlinear relationship with stability and accuracy, which has a very broad prospect for development.
Key words: parameter optimization, artificial neural network, BP Network，MATLAB7.0.



目 錄
摘要………………………………………………………………………………… Ⅰ
ABSTRACT……………………………………………………………………………Ⅱ
1.緒論……………………………………………………………………………… 1
1.1本文的目的及意義……………………………………………………………………… .1
1.2人工神經(jīng)網(wǎng)絡(luò)概率及提出……………………………………………………………… 2
1.3MATLAB的神經(jīng)網(wǎng)絡(luò)工具箱…………………………………………………………………3
1.4本文的主要研究內(nèi)容…………………………………………………………… 4
2.人工神經(jīng)網(wǎng)絡(luò)簡介…………………………………………………………… 5
2.1神經(jīng)細胞及人工神經(jīng)元的組成………………………………………………………… 5
2.2人工神經(jīng)網(wǎng)絡(luò)的功能…………………………………………………………………… 6
2.3人工神經(jīng)網(wǎng)絡(luò)的基本結(jié)構(gòu)……………………………………………………… 7
2.3.1神經(jīng)元網(wǎng)絡(luò)的簡化模型…………………………………………………………… 7
2.3.2單層神經(jīng)元網(wǎng)絡(luò)和多層神經(jīng)網(wǎng)絡(luò)…………………………………………………… 8
2.3.3激活轉(zhuǎn)移函數(shù)……………………………………………………………… 9
2.4神經(jīng)網(wǎng)絡(luò)的發(fā)展方向…………………………………………………………… 11
3. 解決誤差復(fù)映的神經(jīng)網(wǎng)絡(luò)………………………………………………… 14
3.1誤差復(fù)映問題的特點分析……………………………………………………………… 14
3.2神經(jīng)網(wǎng)絡(luò)模型的確定…………………………………………………………………… 16
3.3輸入輸出參數(shù)的確定…………………………………………………………… 16
4.BP網(wǎng)絡(luò)及算法改進………………………………………………………… 18
4.1BP網(wǎng)絡(luò)的結(jié)構(gòu)…………………………………………………………………………… 18
4.2BP網(wǎng)絡(luò)存在的問題及算法改進………………………………………………… 18
4.2.1BP網(wǎng)絡(luò)存在的問題………………………………………………………………… 18
4.2.2BP算法改進………………………………………………………………… 19
5.數(shù)據(jù)預(yù)處理與BP網(wǎng)絡(luò)設(shè)計………………………………………………… 20
5.1數(shù)據(jù)預(yù)處理……………………………………………………………………………… 20
5.2確定網(wǎng)絡(luò)的隱層數(shù)和各層神經(jīng)元數(shù)……………………………………………… 20
5.2.1隱層數(shù)的確定………………………………………………………………………… 20
5.2.2隱層神經(jīng)元的選擇…………………………………………………………… 20
5.3訓(xùn)練參數(shù)及訓(xùn)練測試…………………………………………………………… 21
5.3.1訓(xùn)練網(wǎng)絡(luò)程序代碼…………………………………………………………………… 21
5.3.2測試網(wǎng)絡(luò)及結(jié)論………………………………………………………………23
6.圖形界面設(shè)計結(jié)構(gòu)……………………………………………………………25
6.1圖形界面設(shè)計結(jié)果……………………………………………………………………… 25
6.2網(wǎng)絡(luò)程序源代碼…………………………………………………………………27
7.總結(jié)………………………………………………………………………………30
參考文獻……………………………………………………………………………31
致謝………………………………………………………………………………… 32