Water level prediction method based on extended wavelet neural-network models

Abstract

The invention discloses a water level prediction method based on an extended wavelet neural network model. The method determines the number of input items by analyzing the correlation between input and output, thereby improving the prediction accuracy. The method includes the following steps: Step 1: Calculate the correlation between the input xt, xt-1, xt-2, . . . , xt-n sequence and the output x. Step 2: According to the correlation, each sequence xt, xt-1, xt-2,...,xt-n, n≥1 is decomposed by 3 layers of wavelets using db5 as the mother wavelet, and each sequence is decomposed by wavelets to obtain the wavelet decomposition Coefficient normalization processing is used as the input of the network, and different extended wavelet neural network models are established through different values ​​of n; Step 3: Select the hyperbolic tangent function and linear function as the activation functions of the hidden layer and the output layer respectively, according to the mean square The size of the error (MSE) is used to select the appropriate number of hidden layer nodes, and the Levenberg-Marquardt (LM) optimization algorithm is used to train the network; Step 4: To test the different extended networks obtained, and select the most appropriate extended network neural network, Improve forecast accuracy.

Description

technical field [0001] The invention relates to a water level prediction method based on an extended wavelet neural network model, which belongs to the analysis and research of hydrological processes in hydrology. Background technique [0002] Runoff prediction and river water level prediction are the most common uncertain prediction problems in hydrology. Accurate prediction plays a decisive role in the construction of flood control projects and the allocation and management of water resources. However, because the time series of runoff and water level are affected by many factors such as geographical location and climate change, there are both uncertain and certain factors, so their changes have no regularity and are very complicated. , has certain non-stationary and nonlinear characteristics. For nonlinear time series, the traditional time series analysis method is to establish an ARIMA model, but the biggest disadvantage of this method is that the trend item and cycle i...

AI Technical Summary

Problems solved by technology

However, because the time series of runoff and water level are affected by many factors such as geographical location and climate change, there are both uncertain and certain factors, so their changes have no regularity and are very complicated. , has certain non-stationary and nonlinear characteristics

For nonlinear time series, the traditional time series analysis method is to establish an ARIMA model, but the biggest disadvantage of this method is that the trend item and cycle item are lost, and the characteristics required for modeling are difficult to determine, and it is difficult to identify non-stationary states

Images