發(fā)布時間：2018-05-07 13:53

  本文選題：差分進(jìn)化 + 自適應(yīng)　； 參考：《燕山大學(xué)》2016年博士論文

【摘要】：差分進(jìn)化算法是一種基于個體差異的并行隨機(jī)搜索進(jìn)化算法,具有結(jié)構(gòu)簡單、控制參數(shù)少、全局搜索能力強(qiáng)等優(yōu)點,已被應(yīng)用于諸多領(lǐng)域。但是差分進(jìn)化算法仍存在諸如容易陷入局部最優(yōu)、進(jìn)化停滯、不能求解多目標(biāo)優(yōu)化問題等缺點,限制了其性能的發(fā)揮,阻礙了其應(yīng)用的推廣。因此,對差分進(jìn)化算法的改進(jìn)研究具有重要的理論研究意義與實際應(yīng)用價值。本文在對差分進(jìn)化算法進(jìn)行深入研究的基礎(chǔ)上,針對其存在的缺點提出了三種改進(jìn)算法,并將其應(yīng)用于河南某鋁廠“1+4”鋁熱連軋線的軋制力預(yù)報和負(fù)荷分配優(yōu)化中。本文的主要研究內(nèi)容如下:(1)針對標(biāo)準(zhǔn)差分進(jìn)化算法變異策略和控制參數(shù)固定的問題,提出了一種基于指數(shù)平滑法和混沌映射的自適應(yīng)差分進(jìn)化算法(ECADE)。ECADE算法根據(jù)策略候選池中每個變異策略在當(dāng)前產(chǎn)生更好個體的成功率,使用指數(shù)平滑法預(yù)測下一代的成功概率,并使用輪盤賭選擇法為下一代個體選擇變異策略。此外,ECADE算法使用能平衡算法開發(fā)和探索能力的函數(shù)和Logistic映射生成控制參數(shù)值,從而實現(xiàn)控制參數(shù)自適應(yīng)。經(jīng)標(biāo)準(zhǔn)測試函數(shù)驗證,ECADE算法具有收斂速度快、收斂精度高、探索和開發(fā)能力均衡等優(yōu)點。(2)針對如何提高初始種群多樣性的問題,提出了一種基于對稱拉丁超立方體設(shè)計的自適應(yīng)差分進(jìn)化算法(SLADE)。SLADE算法采用對稱拉丁超立方體設(shè)計(SLHD)技術(shù)初始化種群,并根據(jù)一個較大概率從先前產(chǎn)生更好個體的策略列表或策略候選池中為個體隨機(jī)選擇變異策略。此外,SLADE算法引入柯西分布和正態(tài)分布生成控制參數(shù)值,并且根據(jù)產(chǎn)生更好個體的控制參數(shù)值進(jìn)行自適應(yīng)。實驗結(jié)果表明SLADE算法較其他算法具有更強(qiáng)的尋優(yōu)能力,并且SLHD技術(shù)的引入提高了SLADE算法的性能。(3)針對如何實現(xiàn)差分進(jìn)化算法求解多目標(biāo)優(yōu)化問題,提出了一種基于角度鄰域的多目標(biāo)差分進(jìn)化算法(ANMODE)。ANMODE算法在選擇操作中引入了弱支配的概念,實現(xiàn)了對多目標(biāo)優(yōu)化問題的求解。角度鄰域的引入使得變異操作可以在鄰域內(nèi)進(jìn)行,保證了個體的進(jìn)化方向。此外,ANMODE算法的外部存檔維護(hù)機(jī)制對于改善Pareto前沿近似解集的分布性也起到了關(guān)鍵作用。實驗結(jié)果表明ANMODE算法求解到的Pareto前沿近似解集具有良好的收斂性和分布性,性能明顯優(yōu)于對比算法。(4)針對如何減少傳統(tǒng)軋制力模型誤差的問題,提出了一種基于SLADE的BP神經(jīng)網(wǎng)絡(luò)軋制力預(yù)報模型(S-BP)。該模型使用SLADE算法優(yōu)化BP神經(jīng)網(wǎng)絡(luò),從而提高軋制力的預(yù)報精度。此外,為了增強(qiáng)S-BP模型的抗干擾能力,在S-BP模型的基礎(chǔ)上引入了模型自學(xué)習(xí),提高了軋制力預(yù)報的穩(wěn)定性。實驗結(jié)果表明S-BP模型的軋制力預(yù)報精度明顯高于傳統(tǒng)軋制力模型和BP神經(jīng)網(wǎng)絡(luò),模型自學(xué)習(xí)也有效地改善了S-BP模型的魯棒性。(5)針對如何避免精軋機(jī)組負(fù)荷分配不合理及打滑的問題,使用ECADE算法、SLADE算法和ANMODE算法優(yōu)化鋁熱連軋的負(fù)荷分配。將負(fù)荷分配優(yōu)化分別作為單目標(biāo)和多目標(biāo)優(yōu)化問題,使用上述改進(jìn)算法針對不同目標(biāo)函數(shù)進(jìn)行優(yōu)化。實驗結(jié)果表明優(yōu)化后的負(fù)荷分配方案減輕了精軋機(jī)組的打滑問題,并且平衡了各機(jī)架負(fù)荷,改善了軋件板形,對實際生產(chǎn)有著重要的指導(dǎo)作用。
[Abstract]:Differential evolution algorithm is a parallel random search evolutionary algorithm based on individual differences. It has the advantages of simple structure, less control parameters and strong global search ability, and has been applied to many fields. However, the differential evolution algorithm still exists such shortcomings as easy to fall into local optimal, evolutionary stagnation, and can not solve multi-objective optimization problems. The development of its application has hindered the popularization of its application. Therefore, the improvement of the differential evolution algorithm has important theoretical significance and practical application value. Based on the in-depth study of the differential evolution algorithm, three improved algorithms are proposed for the shortcomings of the differential evolution algorithm, and are applied to a certain aluminum plant in Henan, 1+ The main contents of this paper are as follows: (1) an adaptive differential evolution algorithm (ECADE) based on exponential smoothing and chaotic mapping (ECADE).ECADE algorithm based on the strategy candidate pool is proposed. The mutation strategy produces the success rate of the better individual at present, uses the exponential smoothing method to predict the success probability of the next generation, and uses the roulette selection method as the next generation of individual mutation strategy. In addition, the ECADE algorithm uses the function of the ability to develop and explore the ability of the energy balance algorithm and the Logistic mapping to generate the control parameter values, thus realizing the control. Parameter self-adaptive. The ECADE algorithm has the advantages of fast convergence, high convergence precision and balance of exploration and development through standard test function. (2) in view of how to improve the diversity of initial population, a symmetric Latin hypercube based self adaptive differential evolution algorithm (SLADE).SLADE algorithm is proposed. The Ding Chao cube design (SLHD) technology initializes the population, and according to a larger probability from the previous generation of better individual strategy list or strategy candidate pool for individual random selection mutation strategy. In addition, the SLADE algorithm introduces the Cauchy distribution and normal distribution to generate control parameters, and according to the generation of better individual control parameter values. The experimental results show that the SLADE algorithm has a stronger optimization ability than other algorithms, and the introduction of SLHD technology improves the performance of the SLADE algorithm. (3) a multi-objective differential evolution algorithm (ANMODE).ANMODE algorithm based on the angle neighborhood is proposed to solve the multi-objective optimization problem of differential evolution algorithm. In the optional operation, the concept of weak domination is introduced to solve the multi-objective optimization problem. The introduction of the angle neighborhood makes the mutation operation carry out in the neighborhood and ensure the evolution direction of the individual. In addition, the external archiving maintenance mechanism of the ANMODE algorithm also plays a key role in improving the distribution of the approximate solution set of the Pareto frontier. The experimental results show that the Pareto frontier approximate solution set by ANMODE algorithm has good convergence and distribution, and the performance is obviously superior to the contrast algorithm. (4) a rolling force pre report model (S-BP) based on the BP neural network based on SLADE is proposed to reduce the traditional rolling force model error. This model uses SLADE algorithm to optimize BP. In addition, in order to improve the prediction accuracy of rolling force, in addition, in order to enhance the anti-interference ability of the S-BP model, the model self-learning is introduced on the basis of the S-BP model to improve the stability of the rolling force prediction. The experimental results show that the rolling force prediction accuracy of the S-BP model is higher than that of the traditional rolling force model and the BP neural network. Learning also effectively improves the robustness of the S-BP model. (5) in order to avoid the problem of unreasonable load distribution and skidding of the finishing mill, ECADE algorithm, SLADE algorithm and ANMODE algorithm are used to optimize the load distribution of the aluminum hot continuous rolling mill. Different target functions are optimized. The experimental results show that the optimized load distribution scheme reduces the skidding problem of the finishing mill, balances the load of each frame, improves the shape of the rolled piece, and has an important guiding role for the actual production.

【學(xué)位授予單位】：燕山大學(xué)
【學(xué)位級別】：博士
【學(xué)位授予年份】：2016
【分類號】：TG339;TP18

【參考文獻(xiàn)】

相關(guān)期刊論文 前2條

1 姚峰;楊衛(wèi)東;張明;李仲德;;改進(jìn)自適應(yīng)變空間差分進(jìn)化算法[J];控制理論與應(yīng)用;2010年01期

2 張進(jìn)之;熱連軋機(jī)負(fù)荷分配方法的分析和綜述[J];寬厚板;2004年03期

，

本文編號：1857178