【摘要】 为了提高中小流域流量预测的精度，提出一种混合遗传算法(SP_GA)优化长短时记忆神经网络(LSTM)的流量预测模型(SP_GA-LSTM)。由于LSTM模型参数难以确定，因此采用SP_GA进行参数寻优。SP_GA算法在遗传算法中引入粒子群算法公式作为变异算子，并且在种群进化后期进行模拟退火操作，以此提高算法的收敛速度和全局搜索能力。通过对龙山流域2010年1月到2014年7月39998个小时流量数据进行仿真，结果表明SP_GA算法优化LSTM的方法能够有效提高中小流域流量预测的稳定性和精度。更多还原

【Abstract】 In order to improve the accuracy of flow prediction in small and medium-sized basins, a flow prediction model(sp_ga LSTM) based on long-term memory neural network(LSTM) optimized by hybrid genetic algorithm(sp_ga) is proposed. Since the parameters of LSTM model are difficult to determine, SP_GA was adopted to optimize the parameters. The SP_ GA introduced particle swarm optimization(PSO) formula as mutation operator in genetic algorithm, and carried out simulated annealing operation in the later stage of population evolution, so as to improve the convergence speed and global search ability of the algorithm. Through the simulation experiment of 39998 hours flow data of Longshan basin from January 2010 to July 2014,the results show that the method of optimizing LSTM by SP_GA algorithm can effectively improve the stability and accuracy of flow prediction in small and medium-sized basins.更多还原