An evaluation index weighting method based on initial value embedding and residual learning

By employing initial value embedding and residual learning methods in engineering projects, combined with AHP, entropy weighting, and BP neural networks, the problem of integrating expert knowledge and data patterns was solved, achieving high-precision and adaptive evaluation index weighting, and improving the accuracy and stability of the evaluation.

CN122390535APending Publication Date: 2026-07-14KUNMING UNIV OF SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
KUNMING UNIV OF SCI & TECH
Filing Date
2026-04-17
Publication Date
2026-07-14

AI Technical Summary

Technical Problem

Existing technologies cannot effectively integrate expert knowledge and data patterns in engineering projects, cannot achieve non-linear weight correction, and are unstable in learning under small sample conditions, resulting in inaccurate evaluation results.

Method used

We employ an initial value embedding and residual learning approach, using AHP and entropy weighting to calculate initial weights and BP neural network to learn residuals, constructing an evaluation index weighting system, and combining expert knowledge and data-driven methods to optimize the weights.

Benefits of technology

It achieves high-precision and adaptive indicator weighting under small sample conditions, improving the accuracy and robustness of the evaluation, adapting to different types of projects and evaluation scenarios, and the output evaluation report is easy to understand and accept.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122390535A_ABST
    Figure CN122390535A_ABST
Patent Text Reader

Abstract

The application discloses an evaluation index weighting method based on initial value embedding and residual learning, and belongs to the cross technical field of comprehensive evaluation and machine learning. In view of the technical problems that the existing weighting method cannot fuse expert knowledge and data law, the linear weighting has great limitations, and the small sample adaptability is poor, the application proposes a structure that decomposes the weight into an initial value and a residual: the initial value is obtained by linear combination of AHP and entropy weight method, and is embedded into the BP network as prior knowledge; the BP network learns the residual between the initial weight and the optimal weight, and realizes nonlinear weight correction. The application effectively fuses expert experience and data law, reduces the network learning difficulty, still has good generalization ability under the condition of small samples, can realize high-precision and high-interpretability index weighting and comprehensive evaluation, and adapts to the needs of multiple evaluation scenes such as engineering projects.
Need to check novelty before this filing date? Find Prior Art