(1. 中南大學(xué) 資源與安全工程學(xué)院，長(zhǎng)沙 410083；
2. 中南大學(xué) 數(shù)字礦山研究中心，長(zhǎng)沙 410083)
摘 要: 地質(zhì)模型在礦產(chǎn)勘探與開(kāi)發(fā)中具有重要作用，但在礦山生產(chǎn)實(shí)踐中，由于成本和技術(shù)等諸多因素影響，很難獲得整個(gè)區(qū)塊的地質(zhì)數(shù)據(jù)，而且傳統(tǒng)插值方法依靠經(jīng)驗(yàn)確定參數(shù)有很大局限性。提出將粒子群優(yōu)化算法(PSO)和自適應(yīng)神經(jīng)模糊推理系統(tǒng)(ANFIS)應(yīng)用到礦體品位插值中，利用粒子群優(yōu)化算法的快速搜索能力，神經(jīng)網(wǎng)絡(luò)的學(xué)習(xí)機(jī)制和模糊系統(tǒng)的語(yǔ)言推理能力等優(yōu)勢(shì)構(gòu)建PSO-ANFIS品位插值模型，并借助MATLAB生成571組樣本數(shù)據(jù)作為輸入空間對(duì)模型進(jìn)行訓(xùn)練，其中每一個(gè)訓(xùn)練樣本由待估點(diǎn)三維坐標(biāo)及真實(shí)值和其周?chē)?個(gè)樣品點(diǎn)組成，最后用訓(xùn)練后的PSO-ANFIS模型對(duì)待估點(diǎn)進(jìn)行品位插值，并與距離冪次反比插值法進(jìn)行對(duì)比，其均方根誤差(RMSE)提高了近15%，驗(yàn)證了該模型的可行性和有效性。
關(guān)鍵字: 礦石品位；空間插值；粒子群優(yōu)化算法；自適應(yīng)模糊神經(jīng)推理系統(tǒng)；優(yōu)化
（1. School of Resources and Safety Engineering, Central South University, Changsha 410083, China;
2. Center of Digital Mine Research, Central South University, Changsha 410083, China）
Abstract:Geological model plays an important role in mineral exploration and development, but in the practice of mine production, because of the influence of cost and technology, it is difficult to obtain the geological data of the whole block, and the spatial interpolation is an important means to solve this problem. The particle swarm optimization (PSO) and adaptive neuro-fuzzy inference system (ANFIS) were applied to the grade interpolation of orebody, which overcomes the limitation of traditional interpolation method based on empirical determination of parameters, PSO-ANFIS grade interpolation model was constructed by using the fast searching ability of particle swarm optimization, the learning mechanism of neural network and the language reasoning ability of fuzzy system. Selecting 571 groups of sample points as training data to train the model with the cross verification method in MATALB, each of these training samples consists of three-dimensional coordinates and true values of the estimated points and eight surrounding sample points, finally, the PSO-ANFIS model was used to evaluate the evaluation point and the mean square root error (RMSE) was improved by comparing with the distance power-time inverse interpolation method, which is nearly 15%. The feasibility and effectiveness of the model were validated.
Key words: ore grade; spatial interpolation; particle swarm optimization algorithm; adaptive neuron-fuzzy inference system; optimization
