摘要
针对当前犯罪预测模型中LSTM网络的神经元数量和时间步长等超参数难以确定的问题,提出基于误差采样改进的多元宇宙算法与LSTM网络相结合的预测模型。该模型在优化过程中采用随机权重样本的误差分布来评估网络结构的性能,无需训练就可以比较出不同候选结构的优劣,从而大幅缩短寻找模型最优超参数的时间。
Aiming at the problem that it is difficult to determine the number of neurons and time step of LSTM network in the current crime prediction model,a prediction model based on improved multiverse algorithm with error sampling and LSTM network is proposed.In the optimization process,the model uses the error distribution of random weight samples to evaluate the performance of network structure and can compare the advantages and disadvantages of different candidate structures without training,therefore,the time of searching the optimal hyperparameters of the model is shortened greatly.
作者
杨铭
Yang Ming(Network Security Team of Public Security Department of Shanxi Provincial Public Security Department,Taiyuan Shanxi 030006,China)
出处
《山西电子技术》
2023年第2期98-100,共3页
Shanxi Electronic Technology
关键词
数据挖掘
LSTM网络
多元宇宙算法
data mining
LSTM network
multiverse optimization algorithm
作者简介
杨铭(1981-),男,河北泊头人,本科学历,现从事网络安全,大数据研究。
