Neural network model construction method, time sequence prediction method and device

Abstract

The invention provides an EWT-BO-BiLSTM neural network model construction method based on information entropy, and a time sequence prediction method and device. The construction method comprises the following steps: decomposing sample time sequence data into N subsequences by using an EWT decomposition algorithm; calculating the information entropy of each subsequence; performing complexity analysis on each sub-sequence according to the size of the information entropy, combining and recombining each sub-sequence based on the approximation degree of the information entropy to obtain a new recombined sub-sequence, and sequentially marking the new recombined sub-sequence as a trend component sequence, a periodic component sequence and a random component sequence from low to high according to the information entropy; and constructing BiLSTM networks for the trend component sequence, the periodic component sequence and the random component sequence obtained by recombining, adjusting BiLSTM network parameters by using a Bayesian optimization method to obtain a BO-BiLSTM model, and performing model training. The prediction method comprises the steps of performing prediction and reconstruction based on the constructed model. According to the scheme, the prediction precision of the high-frequency time sequence with multiple influence factors can be improved.

Description

technical field [0001] The present invention relates to the technical field of time series forecasting, in particular to a method for constructing a neural network model based on information entropy EWT-BO-BiLSTM, a time series forecasting method and a device. Background technique [0002] With the extensive application of sensor technology, automatic identification and prompting of existing or upcoming hidden dangers in equipment and systems through data collection and analysis technologies has gradually become an important means to achieve efficient operation and maintenance under low labor cost conditions. How to improve the accuracy of time series forecasting based on equipment historical monitoring data is an important research topic in data-driven forecasting models. [0003] In actual production and engineering applications, since the equipment monitoring value is often affected by many factors, its sequence change trend presents non-stationary, nonlinear characterist...

AI Technical Summary

Problems solved by technology

However, the traditional time-domain and frequency-domain methods cannot simultaneously take into account the localization characteristics of signals in the time-frequency domain, and in the existing research on time series prediction of equipment monitoring data with complex characteristics, it is often difficult to extract and separate effective information in the data. Based on The prediction model of deep learning is faced with improper adjustment of hyperparameters, which fails to effectively utilize the advantages of model prediction and the pressure of model training time.

Images