CN-121997000-A - Method and device for predicting performance curve

CN121997000ACN 121997000 ACN121997000 ACN 121997000ACN-121997000-A

Abstract

The disclosure relates to a performance curve prediction method and device, relates to the technical field of artificial intelligence, and aims to improve accuracy of a predicted curve. The method comprises the steps of obtaining first data used for representing the change trend of a historical performance curve of a target object, obtaining second data used for representing the attribute of the target object, wherein the attribute is overlapped with a time range corresponding to the historical performance curve, the attribute affects the change trend of the historical performance curve, obtaining fusion characteristics through a prediction model according to the first data, the fusion characteristics are represented by the characteristics of the historical performance curve in a hidden space after the attribute corresponding to a plurality of historical time periods are fused, decoding the fusion characteristics through the prediction model, and obtaining a prediction curve of the target object, wherein the prediction curve is used for predicting the change trend of the performance of the target object.

Inventors

TANG JINGXIANG
LI ZIYONG
ZHANG YU
PAN ZHICHENG

Assignees

深圳前海微众银行股份有限公司
新加坡南洋理工大学

Dates

Publication Date: 20260508
Application Date: 20260120

Claims (10)

1. A method of predicting a performance curve, the method comprising: Acquiring first data used for representing the change trend of a historical performance curve of a target object, wherein the change trend of the historical performance curve continuously changes along with time; Acquiring second data for representing attributes of a target object, wherein the attributes overlap with a time range corresponding to the historical performance curve, and the attributes influence the change trend of the historical performance curve; Encoding the first data according to the second data through a prediction model to obtain fusion characteristics, wherein the fusion characteristics are characteristic representations of the historical performance curves in a hidden space after the attributes are fused; And decoding the fusion characteristic through the prediction model to obtain a prediction curve of the target object, wherein the prediction curve is used for predicting the change trend of the performance of the target object.
2. The method of claim 1, wherein the predictive model includes an encoder, wherein the encoding the first data from the second data results in a fusion feature, comprising: And carrying out feature fusion of the historical performance curve relative to the attribute by the encoder according to the first data and the second data, and obtaining the fusion feature.
3. The method of claim 2, wherein said obtaining, by the encoder, the fusion of the historical performance curve with respect to the attribute features from the first data and the second data, comprises: The encoder is used for carrying out attention crossing of a historical performance curve relative to attributes on the first data and the second data to obtain a first intermediate feature, wherein the first intermediate feature is a feature representation of the first data in the hidden space after the attention crossing is carried out on the first intermediate feature and the second data; And performing self-attention processing according to the first intermediate feature by the encoder to obtain the fusion feature.
4. The method of claim 3, wherein the fusion feature is obtained by a number N of encoding passes, N being a positive integer greater than 1, and wherein in an ith pass: Performing attention crossing of a historical performance curve relative to attributes through the first characteristic of the ith round of encoding process and the second characteristic of the ith round of encoding process to obtain a first intermediate characteristic of the ith round of encoding process; performing attention crossing of attributes relative to a historical performance curve through the first intermediate feature of the ith round of encoding process and the second feature of the ith round of encoding process to obtain the second intermediate feature of the ith round of encoding process; Performing self-attention processing through the first intermediate feature of the ith round of coding process to obtain a first fusion feature of the ith round of coding process; The first characteristic of the 1 st round of encoding process is the first data, the second characteristic of the 1 st round of encoding process is the second data, the first fusion characteristic of the i st round of encoding process is used as the first characteristic of the i+1 th round of encoding process, the second intermediate characteristic of the i st round of encoding process is used as the second characteristic of the i+1 th round of encoding process, and the first fusion characteristic of the N th round of encoding process is the fusion characteristic.
5. The method of claim 4, wherein the encoder comprises a curve attribute attention block, a self-attention block, and an attribute curve attention block, The curve attribute attention block is used for performing attention crossing of the historical performance curve relative to the attribute; The attribute curve attention block is used for performing attention crossing of attributes relative to a historical performance curve; The self-attention block is for performing self-attention processing of the first intermediate feature.
6. The method of claim 4, wherein said crossing the attention of the historical performance curve with respect to the attribute by the first feature of the ith round of encoding process and the second feature of the ith round of encoding process, resulting in a first intermediate feature of the ith round of encoding process, comprises: Calculating a first Q matrix through a first feature of the ith round of encoding process, calculating a second K matrix and a second V matrix through a second feature of the ith round of encoding process, and calculating a first intermediate feature of the ith round of encoding process according to the first Q matrix, the second K matrix and the second V matrix; The performing the attention crossing of the attribute relative to the historical performance curve through the first intermediate feature of the ith round of encoding process and the second feature of the ith round of encoding process to obtain the second intermediate feature of the ith round of encoding process, including: Calculating a second Q matrix according to the second characteristic of the ith round of coding process, calculating a first K matrix and a first V matrix according to the first intermediate characteristic of the ith round of coding process, and calculating the second intermediate characteristic of the ith round of coding process according to the second Q matrix, the first K matrix and the first V matrix; The self-attention processing is carried out through the first intermediate feature of the ith round of coding process, so as to obtain a first fusion feature of the ith round of coding process, which comprises the following steps: And calculating a third Q matrix, a third K matrix and a third V matrix according to the first intermediate feature of the ith round of coding process, and calculating a first fusion feature of the ith round of coding process according to the third Q matrix, the third K matrix and the third V matrix.
7. The method according to any one of claims 1-6, wherein said decoding the fusion feature by the predictive model to obtain a predicted curve of the target object comprises: Decoding the fusion characteristics to obtain predicted performance values of the target object at a plurality of predicted time points; Determining a prediction curve of the target object in a prediction time period according to the prediction performance values of the plurality of prediction time points; The values of a plurality of prediction time points in the prediction curve are equal to the prediction performance values of the plurality of prediction time points; the predicted performance values of the plurality of predicted time points are absolute predicted values or the predicted performance values of the plurality of predicted time points are relative predicted values with respect to the historical performance values of the corresponding historical time points in the historical performance curve.
8. The method of any one of claims 1-6, wherein the obtaining first data characterizing a trend of change in a historical performance curve of the target object comprises: Respectively sampling historical performance curves of a plurality of historical time periods to obtain a plurality of sampling points respectively corresponding to the historical performance curves of the historical time periods; And obtaining the first data according to a plurality of sampling points respectively corresponding to the historical performance curves of the plurality of historical time periods, wherein the plurality of sampling points respectively corresponding to the historical performance curves of the plurality of historical time periods are arranged in the first data according to a time sequence.
9. The method of any of claims 1-6, wherein the obtaining second data characterizing the properties of the target object comprises: acquiring attributes corresponding to the target object in a plurality of historical time periods; Extracting the characteristics of the attributes corresponding to each historical time period to obtain the attribute characteristics of each historical time period, wherein the attribute characteristics of the historical time period are represented by the attributes of the corresponding historical time period in a hidden space; And obtaining the second data according to the attribute characteristics of the plurality of historical time periods, wherein the attribute characteristics of different time periods in the second data correspond to different position codes.
10. The method according to any one of claims 1 to 6, wherein, The prediction model is obtained based on a sample history performance curve of a sample object and sample attribute training of the sample object; In the training process, the prediction model generates a prediction curve according to first sample data corresponding to the sample historical performance curve and second sample data corresponding to the sample attribute; The weight parameters in the prediction model are adjusted based on training loss, and the training loss is determined according to the application effect of the prediction curve.

Description

Method and device for predicting performance curve Technical Field The application relates to the technical field of artificial intelligence, in particular to a performance curve prediction method and device. Background With the development of artificial intelligence technology, data prediction using an artificial intelligence model is widely used in a plurality of fields. In the related art, an artificial intelligence model is used to predict a performance curve of an object. For example, a performance curve is predicted based on a curve-curve prediction approach. For example, historical performance curves are collected and input into the artificial intelligence model such that the artificial intelligence model outputs a predicted performance curve, or outputs key points in the predicted performance curve, based on the historical performance curves. However, because of the uncertainty of fluctuations in the historical performance curve, the artificial intelligence model is based on the historical performance curve, and the accuracy of the resulting predicted curve is not high. Disclosure of Invention The disclosure provides a performance curve prediction method and device, and aims to improve accuracy of a predicted performance curve. The technical scheme of the present disclosure is as follows: according to an aspect of an embodiment of the present application, there is provided a method for predicting a performance curve, including the steps of: Acquiring first data used for representing the change trend of a historical performance curve of a target object, wherein the change trend of the historical performance curve continuously changes along with time; acquiring second data for representing the attribute of the target object, wherein the attribute overlaps with a time range corresponding to the historical performance curve, and the attribute influences the change trend of the historical performance curve; Encoding the first data according to the second data through a prediction model to obtain fusion characteristics, wherein the fusion characteristics are characteristic representations of the historical performance curves in the hidden space after the fusion of the attributes corresponding to the historical time periods; and decoding the fusion characteristics through the prediction model to obtain a prediction curve of the target object, wherein the prediction curve is used for predicting the change trend of the performance of the target object. In one possible implementation, the prediction model includes an encoder that encodes the first data based on the second data to obtain a fusion feature, including: And carrying out feature fusion of the historical performance curve relative to the attribute through the encoder according to the first data and the second data, and obtaining fusion features. In one possible implementation, the method for obtaining the fusion feature by the encoder according to the feature fusion of the historical performance curve relative to the attribute according to the first data and the second data includes: the method comprises the steps of performing attention crossing of a historical performance curve relative to attributes on first data and second data through an encoder to obtain first intermediate features, wherein the first intermediate features are represented by the features of the first data in a hidden space after the attention crossing is performed on the first intermediate features and the second data; And performing self-attention processing according to the first intermediate feature by an encoder to obtain a fusion feature. In one possible implementation, the fusion feature is obtained by N rounds of encoding, where N is a positive integer greater than 1, and in the ith round of encoding: The method comprises the steps of performing attention crossing of a historical performance curve relative to attributes through a first characteristic of an ith round of encoding process and a second characteristic of the ith round of encoding process to obtain a first intermediate characteristic of the ith round of encoding process; Performing attention crossing of the attribute relative to the historical performance curve through the first intermediate feature of the ith round of encoding process and the second feature of the ith round of encoding process to obtain the second intermediate feature of the ith round of encoding process; performing self-attention processing through the first intermediate feature of the ith round of coding process to obtain a first fusion feature of the ith round of coding process; The first characteristic of the 1 st round of coding process is first data, the second characteristic of the 1 st round of coding process is second data, the first fusion characteristic of the i st round of coding process is used as the first characteristic of the i+1 th round of coding process, the second intermediate characteristic of the i st round of coding pro