(1. 中南大學(xué) 有色金屬成礦預(yù)測(cè)教育部重點(diǎn)實(shí)驗(yàn)室，長(zhǎng)沙 410083；
2. 中南大學(xué) 地球科學(xué)與信息物理學(xué)院，長(zhǎng)沙 410083)
摘 要: 粒子群優(yōu)化算法是一種啟發(fā)式的全局優(yōu)化算法，將其與BP神經(jīng)網(wǎng)絡(luò)結(jié)合，能夠有效地改善BP神經(jīng)網(wǎng)絡(luò)在進(jìn)行電阻率層析反演中的收斂速度和求解質(zhì)量。提出一種基于混沌振蕩的粒子群算法，使用混沌振蕩曲線來(lái)自適應(yīng)調(diào)整慣性權(quán)重w以提高PSO算法的全局尋優(yōu)能力，并使用其訓(xùn)練和優(yōu)化BP神經(jīng)網(wǎng)絡(luò)的權(quán)值和閾值。比較不同隱含層節(jié)點(diǎn)數(shù)目和慣性權(quán)重w值對(duì)反演結(jié)果的影響，并給出混沌振蕩PSO-BP算法非線性反演的具體實(shí)現(xiàn)方案。對(duì)均勻半空間中異常體理論模型進(jìn)行反演，實(shí)驗(yàn)結(jié)果表明：混沌振蕩PSO-BP不依賴初始模型，在穩(wěn)定性和準(zhǔn)確性上優(yōu)于BP反演和標(biāo)準(zhǔn)PSO-BP反演，成像質(zhì)量?jī)?yōu)于最小二乘法反演的。
關(guān)鍵字: 電阻率層析成像；非線性反演；粒子群優(yōu)化；反向傳播網(wǎng)絡(luò)；混沌序列
（1. Key Laboratory of Metallogenic Prediction of Nonferrous Metals, Ministry of Education, Central South University, Changsha 410083, China;
2. School of Geosciences and Info-Physics, Central South University, Changsha 410083, China）
Abstract:The particle swarm optimization (PSO) is a heuristic global optimization method, which can effectively improve the convergence speed and the results quality with the BP neural network in resistivity tomography 2-D nonlinear inversion. A chaotic oscillation PSO algorithm was presented, and the chaos oscillation curve was used to adjust the inertia weight adaptively and improve the global optimum capability of PSO. And this algorithm was used to train and optimize the weights and threshold values of the BP neural network. The impacts of different numbers of the hidden layer nodes and types of the inertia weight to the inversion result were compared, and an implementation of chaotic oscillation PSO-BP algorithm was given. The half space abnormity synthetic model was inversed. The results show that the chaotic oscillation PSO-BP algorithm that is independent of the initial model has better performance than BP and standard PSO-BP algorithm in stability and accuracy, and has higher imaging quality than least square inversion.
Key words: electrical resistivity tomography; nonlinear inversion; particle swarm optimization; back propagation network; chaotic sequences
