基于堆栈式自编码器的尾矿库安全评价

Safety evaluation of the tailing pond based on stacked auto-encoder

为有效评估尾矿库的安全状况,针对尾矿库数据的随机波动性、非线性和多数据源的特点,采用堆栈式自编码器算法对尾矿库进行安全评价.基于多层结构、稀疏性限制,该算法采用贪心逐层训练策略对网络权值进行优化,进而对尾矿库进行安全评价.结合淳安某尾矿库的数据进行了安全评价的仿真实验,结果表明:堆栈式自编码器算法能克服多层网络结构权值易陷入局部最小值的缺陷,有效刻画数据的非线性和随机波动性,具备良好的评价准确率.

For the purpose of evaluating the safety status of tailing pond, a prediction model is established by adopting stacked auto-encoder algorithm according to the characteristics of stochastic fluctuation, non-linear and multiple data sources. Based on the multi-level architectures and sparsity limitation, this algorithm uses Greedy Layer-wise Algorithm to train parameters in order to optimize the network weights. The applied safety evaluation on Chunan tailing pond shows that stacked auto-encoder could overcome the defect that the optimized network weights will be easy to fall into local minimum in the multi-level architectures. It can effectively describe characterization of nonlinear and stochastic volatility of the data with a good evaluation accuracy.