Nonparametric regression method

Abstract

The invention discloses a nonparametric regression method, which relates to the field of forecast methods. The method of the invention comprises the following steps: determining a forecast quantity according to a forecast object; acquiring the properties P1-Pn of the forecast quantity from the forecast object, and using the properties P1-Pn as each component of the forecast object state, wherein n is the number of properties; collecting patterns; constructing a pattern library by a KD tree data structure according to the collected patterns; collecting parameters of the state of the forecast object, composing the current state vectors of the forecast object by the parameters, searching for k numbered patterns similar to the current state vectors in the pattern library according to a predetermined criterion, and acquiring the values y1-yn of the quantities to be forecasted corresponding to the k numbered patterns; substituting the acquired values y1-yn of the quantities to be forecasted into a forecast function to obtain the forecast value yforcast; after a time T, collecting the real value yreal of the quantities to be forecasted; calculating the forecast error e according to the forecast value yforcast and the real value yreal; and adjusting the weight in the predetermined criterion and the structure of the pattern library according to the forecast error e. The method improves the calculation speed and the forecast precision of nonparametric regression forecast, and meets the requirement in practical application.

Description

technical field [0001] The invention relates to the field of prediction methods, in particular to a non-parametric regression method. Background technique [0002] Non-parametric regression (NPR) is a data-driven prediction technique, and it is a non-parametric estimation method suitable for uncertain, nonlinear dynamic systems that has emerged in recent years. The principle of non-parametric regression prediction is as follows: Assume that there is a relationship y=m(x) between the state x of the predicted object and a variable y of concern, if the current state of the predicted object is If the method of non-parametric regression is used, the following process can be used to predict when the state of the predicted object is Time value of: [0003] 1. Collect enough known data (x 1 ,y 1 )~(x n ,y n ) and stored as a dataset. [0004] 2. Find a subset X of the data set n , satisfying for where R is a "similarity" relation defined by the method user, i.e. x and ...

AI Technical Summary

Problems solved by technology

[0007] In terms of calculation speed, most of the existing methods use linear structures to build pattern libraries. Although they are simple and easy to implement search algorithms, their structure determines that the implementation of search algorithms can only be performed in one-dimensional order, so the search operation The time required is too long, especially when the scale of the sample database is relatively large, and it cannot meet the real-time requirements

In terms of prediction accuracy, most of the existing methods are open-loop systems. Once the model library and system parameters are set, they will not change and cannot be adjusted during operation, thus affecting the actual prediction effect.

Images