建立相应的安全监控模型来分析大坝变形监测资料对保障大坝服役安全意义重大。BP神经网络模型在此方面得到了广泛应用,但采用蚁群算法(ACO)对BP神经网络参数寻优时存在因初期搜索完全随机导致收敛速度慢的问题。将具有快速随机的全局搜索能力的遗传算法(GA)引入蚁群算法中,利用遗传算法指导生成初始信息素分布,再由蚁群算法正反馈寻得最优解来训练BP神经网络,从而得到大坝变形预测值,2种算法优势互补,缩短了蚁群算法的搜索时间并避免陷入局部最优点。在此基础上,为进一步提高预测精度,采用马尔科夫链(MC)对预测结果进行改进,由此建立了应用于大坝变形监控的GACO-BP-MC模型。工程实例分析表明,该模型在参数优化方面具有较快的寻优速率,且具有较高的拟合和预报能力。

Back Propagation (BP) neural network has been widely used to establish monitoring models in analyzing dam deformation data. Nevertheless, when optimizing the BP neural network parameters, Ant Colony Optimization (ACO) algorithm converges slowly in the beginning due to completely random search. In the present study, a dam deformation monitoring model combining genetic ACO, BP, and Markov Chain (MC) is built to tackle this problem. First of all, Genetic Algorithm (GA) which has remarkable ability of global search is introduced to help guide the initial distribution of pheromone, and the optimal solution is obtained by the positive feedback of ACO to train BP neural network to get the predicted values of dam deformation. Since the advantages of the two algorithms are complementary, this improvement greatly reduces the time taken in the initial stage of optimization and avoids falling into the local optimum. Furthermore, to improve the prediction accuracy, MC is employed to correct residual errors of the prediction results. Engineering application case manifests that the model is of good ability of fitting and prediction with fast searching speed in parameter optimization.