基于双向LSTM神经网络的站点周边水位预测系统设计

Design of water level prediction system around station based on bidirectional LSTM neural network

摘要:

水位记录数据与原始水位数据之间的差值过大,是导致站点主机无法准确预测周边水系特点的主要原因,为解决上述问题,设计基于双向LSTM神经网络的站点周边水位预测系统。站点周边水位预测系统硬件部分设计了周边水系查询体系与水位记录装置；在此基础上,根据初始参数定义结果,建立LSTM神经网络布局模型,并完善水位预测系统双向LSTM解码器的连接闭环,实现站点周边水位预测系统的总体执行方案设计。采集水位数据,并实施针对性的清洗处理,利用完成清洗后的数据对象,建立一维水动力模型,再根据水系糙率计算结果,确定流量与延时时间的数值关系,实现对站点及其周边水系特点的分析,结合相关软、硬件结构,完成基于双向LSTM神经网络的站点周边水位预测系统的设计。实验结果表明,上述系统的应用可以保证水位记录数据与原始水位数据之间的无误差拟合,不会因为水位数据差值过大而导致非精准预测水系特点的问题。

Abstract:

The large difference between the recorded water level data and the original water level data is the main reason why the station host cannot accurately predict the characteristics of the surrounding water system. To solve the above problem, a bidirectional LSTM neural network based station surrounding water level prediction system is designed. The hardware part of the water level prediction system around the station is designed with a peripheral water system query system and water level recording device; On this basis, based on the initial parameter definition results, an LSTM neural network layout model is established, and the connection loop of the bidirectional LSTM decoder in the water level prediction system is improved to achieve the overall execution plan design of the water level prediction system around the station. Collect water level data and implement targeted cleaning treatment. Using the cleaned data object, establish a one-dimensional hydrodynamic model. Then, based on the calculation results of water system roughness, determine the numerical relationship between flow rate and delay time, and analyze the characteristics of the station and its surrounding water system. Combined with relevant software and hardware structures, complete the design of a bidirectional LSTM neural network based station surrounding water level prediction system. The experimental results indicate that the application of the above system can ensure error free fitting between the water level recorded data and the original water level data, and will not lead to inaccurate prediction of water system characteristics due to the large difference in water level data.