A time series prediction method and device based on indefinite length fuzzy information granules

Abstract

The invention provides a time series prediction method and device based on indefinite length fuzzy information granules. The method adopts a K-line graph construction method to process the original time series, and the unequal-length partition universe method based on the K-line pen-splitting idea is used to partition the fuzzy time series. Based on the partitioned fuzzy time series, a band-shapedGaussian fuzzy information granule is constructed to form an indefinite-length fuzzy information granule population of fuzzy time series. The loop fuzzy neural network is constructed, and the structure learning and the parameter learning is carried out. The long-term prediction of fuzzy time series is carried out by using cyclic fuzzy neural network, and the prediction results are de-fuzzified. The invention realizes the long-term prediction of the time series based on the information particle and the circulating fuzzy neural network, and the predicted plurality of values can be completed inone step, instead of predicting each value separately and iteratively, that is, the long-term prediction can be realized.

Description

Technical field [0001] The present disclosure relates to a novel time series prediction method and device based on fuzzy information particles of indefinite length. Background technique [0002] Time series refers to the collection of data collected at different points in time that reflect changes in a certain thing, phenomenon, etc. over time, which generally appear in many fields such as economy, finance, social nature, etc. An important purpose of time series analysis is to predict the time series, that is, use statistical methods and techniques to find out the internal evolution of the time series from the observation data, establish a mathematical model, and estimate the changing trend of the predictor variables. Time series forecasting has been widely used in many fields such as meteorology, agricultural production, tourist numbers and energy, especially in the field of control and financial markets. [0003] The prediction models of time series are mainly divided into three...

AI Technical Summary

Problems solved by technology

If there are too few or too many sub-intervals, the fuzzy semantics of the sub-intervals will be too fuzzy or the fuzzy semantics will be too close to be distinguished.

Regarding the method of dividing the domain of discourse, the research results can be roughly divided into three categories, each of which has its own shortcomings: the first category is to divide the domain of discourse evenly, this type of model is relatively simple, but the prediction accuracy is not high, and the division The fuzzy sets obtained after the domain of discourse are less explanatory and contain less semantic information; the second type is to divide the domain of discourse according to the distribution of data. Although it has strong explanatory power, the result of domain division can be better understood by people However, the complexity of the algorithm is generally high, and it is difficult to achieve dynamic real-time division; the third category is to use optimization theory to divide the domain of discourse. This type of method uses some optimization algorithms, such as genetic algorithm (GA), particle swarm algorithm (PSO), etc., to find the optimal division point, this type of method improves the prediction accuracy compared with the previous two methods, but the interval obtained after division is not easy to describe with people's natural language

However, most of the loop structures in RFNN proposed today are missing loops in the time dimension, and cannot achieve true loops in the time dimension.

[0006] To sum up, the current prediction model has strong requirements for data integrity, weak interpretability, ignoring the trend of data changes, and the intervals obtained after division are not easy to describe with people's natural language and the cyclic fuzzy neural network. Most of the loop structures are lacking in the loop in the time dimension, and the problem of looping in the time dimension in the true sense cannot be realized, and there is still a lack of effective solutions

Images