Research On Water Level Prediction Based On Time Series Prediction Model

Urban waterlogging takes a serious threat to people's lives and property,which has troubled urban residents for many years.It is also one of the problems that need to be solved urgently for building a smart city.With the rapid development of sensor technology and the Internet of Things technology,people have gradually shifted their attention to real-time monitoring of the waterlogging trend of flooding points.The monitoring data can provide an important reference for flood warning and formulation of disaster prevention measures.How to conveniently and quickly get the data of the water monitoring sensors and build a prediction model to predict the water level has become a new research hotspot.In order to collect the sensor monitoring data uniformly and get it remotely,a data storage system designed and implemented based on the Internet of Things platform.This system can satisfy the massive and fast-growing storage requirements of data in the platform,and provide efficient data search functions.And also,it can provide a reliable sensor time series data foundation for the construction of prediction models.In order to improve the accuracy of the water time series prediction,a combined CNLSTM time series prediction model is constructed to model and predict the multi-dimensional time series.This model uses Convolutional Neural Network to extract the spatial features between multivariate data,and Long Short-Term Memory Network to extract the temporal features to predict the water level.Compared with the prediction results of a single CNN and LSTM neural network,the validity of the proposed model is verified.Further,the CALNET time-series prediction model is proposed based on the CNLSTM model.It overcome some shortcomings of the CNLSTM model,such as insufficient granularity for data feature extraction,and insensitivity to the linear relationship between data.The CALNET model first uses CNN to extract the feature relationships of the multivariate time series on the spatial dimension and time axis respectively to obtain the feature vector with temporal and spatial correlation,and then uses LSTM to make predictions.At the same time,a linear component is added to increase the linear sensitivity of the model.The experimental results show that the prediction performance of the CALNET model is further improved.Compared with the traditional linear prediction model and BP neural network,the validity and superiority of the proposed model are verified.