CN-121998449-A - Low-efficiency asset prediction method and system based on multidimensional dynamic weight

CN121998449ACN 121998449 ACN121998449 ACN 121998449ACN-121998449-A

Abstract

The invention discloses an inefficient asset prediction method and system based on multidimensional dynamic weights, and relates to the technical field of enterprise asset management and predictive analysis. The method comprises the steps of collecting original data of assets, conducting data cleaning and standardization, constructing and extracting a four-dimensional feature system comprising financial features, state features, time features and business features, calculating active state scores and health scores of the assets, dynamically adjusting weights of the features of each dimension based on an asset state evaluation result, adjusting weights of the features in the dimensions of the features based on a feature importance evaluation result, applying the dynamically calculated weights to a prediction model, conducting low-efficiency asset identification, evaluating prediction performance, and dynamically updating the weights according to a feedback result. By constructing a four-dimensional feature system, designing a dynamic weight computing mechanism, realizing real-time feature importance assessment and differential data processing, the accuracy and adaptability of the low-efficiency asset identification are obviously improved.

Inventors

MENG ZUJUN
Wan Mengling
WU JIANCHUN
CHEN WEI
XIA MIN
PENG JUN
GAO JIANWEI
HE WENYUN
LIU JUAN
LUO LIANGUI

Assignees

中通服网络信息技术有限公司

Dates

Publication Date: 20260508
Application Date: 20251224

Claims (10)

1. An inefficient asset prediction method based on multi-dimensional dynamic weights is characterized by comprising the following steps: collecting and preprocessing data, collecting the original data of the assets, and cleaning and standardizing the data; multi-dimensional feature extraction, namely constructing and extracting a four-dimensional feature system comprising financial features, state features, time features and business features; asset status assessment, calculating an activity status score and a health score for an asset; dynamically adjusting the weight of each dimension characteristic based on the asset state evaluation result; fine adjustment of feature level weights, namely adjusting the weights of the features in the affiliated dimensions based on feature importance evaluation results; weight application and prediction, wherein the weight obtained by dynamic calculation is applied to a prediction model to perform low-efficiency asset identification; And (3) evaluating the performance, updating the weight, evaluating the prediction performance, and dynamically updating the weight according to the feedback result.
2. The method for predicting an asset based on multi-dimensional dynamic weights of claim 1, wherein in the asset status assessment: The liveness score of an asset is calculated by the formula: The calculation is performed such that, 、、、 Is a binary variable, and takes the value of 1 or 0; the health score of an asset is calculated by the formula: And (3) calculating, wherein, As a net worth of the asset, Is a cost value; In order to be of a depreciation rate, Coefficients 0.6 and 0.4 were determined by a linear regression model.
3. The method for predicting an asset based on multi-dimensional dynamic weights as recited in claim 2, the method is characterized in that the dimension weight dynamic adjustment comprises the following steps: the adjustment formula of the state characteristic weight is as follows: ; the adjustment formula of the time characteristic weight is as follows: ; the adjustment formula of the business feature weight is as follows: Wherein Historical performance scores for the assets; and carrying out normalization processing on the adjusted weight of each dimension to ensure that the sum of the weights is 1.
4. The method of claim 1, wherein the adjusting of the temporal feature weights further comprises a four-stage weight adjustment strategy: according to the service life of the asset And whether or not to exceed the age Setting time weight factor : If it is Then ; If it is Then ; If it is Then ; If it is Then 。
5. The method of claim 1, wherein the adjusting of the business feature weights further comprises a three-stage weight adjustment strategy: ABC classification by asset Setting a service weight factor : If it is Then ; If it is Then ; If it is Then 。
6. The method for predicting low-efficiency assets based on multi-dimensional dynamic weights as claimed in claim 1, wherein said feature importance assessment uses XGBoost algorithm, characterized by an importance score The calculation formula of (2) is as follows: , Wherein, the In order to make a decision about the number of trees, For the number of nodes to be the number of nodes, The gain of the information for node splitting is that, Is an indication function; the formula for fine adjustment of the feature level weight is as follows: 。
7. the method for predicting an asset based on multi-dimensional dynamic weights of claim 1, the method is characterized in that the preprocessing step comprises a differential missing value processing strategy: for remaining life in temporal characteristics A missing value, filled to-1.3; for age in temporal characteristics A missing value, filling 9.3; and uniformly filling the missing values of other numerical characteristics into-1.
8. The method for predicting the low-efficiency asset based on the multi-dimensional dynamic weight according to claim 1, wherein the four-dimensional feature system specifically comprises: Financial characteristics: equity value of assets Amount of depreciation Cost value Preparation of reduced value ; Status feature of whether or not to deactivate Whether or not to be idle Whether or not to wait for scrapping Whether or not to exceed the age Whether or not to be fully depreciated ; Time characteristics of service life Residual life of ; Service characteristics customer type Network attributes ABC classification , , 。
9. The method for predicting an asset based on multi-dimensional dynamic weights as recited in claim 1, wherein the initial base weight assignment for each dimensional feature is: the financial feature basic weight is 0.3, the state feature basic weight is 0.3, the time feature basic weight is 0.2, and the business feature basic weight is 0.2.
10. An inefficient asset prediction system based on multi-dimensional dynamic weights, characterized in that it implements a method for predicting an inefficient asset based on multi-dimensional dynamic weights as claimed in any one of claims 1 to 9, comprising: the data acquisition and preprocessing module is used for collecting the original data of the assets and carrying out data cleaning and standardization; the multi-dimensional feature extraction module is used for constructing and extracting a four-dimensional feature system comprising financial features, state features, time features and business features; an asset status assessment module that calculates an activity status score and a health score for an asset; the dimension weight dynamic adjustment module is used for dynamically adjusting the weight of each dimension characteristic based on the asset state evaluation result; The feature level weight refinement adjustment module is used for adjusting the weight of each feature in the dimension of the feature based on the feature importance evaluation result; The weight application and prediction module is used for applying the weight obtained by dynamic calculation to a prediction model and carrying out low-efficiency asset identification; And the performance evaluation and weight updating module evaluates the prediction performance and dynamically updates the weight according to the feedback result.

Description

Low-efficiency asset prediction method and system based on multidimensional dynamic weight Technical Field The invention relates to the technical field of enterprise asset management and predictive analysis, in particular to an inefficient asset prediction method and system based on multidimensional dynamic weights. Background With the increasing size of enterprise assets, how to efficiently and accurately identify inefficient assets has become a core challenge in improving enterprise asset management levels and asset utilization efficiency. The low-efficiency assets occupy a large amount of funds and space of the enterprise, and can also cause maintenance cost to rise, so that the operation benefit of the enterprise is seriously affected. The traditional low-efficiency asset identification method mainly relies on experience of management personnel to carry out manual judgment. The method is high in subjectivity and low in efficiency, potential rules are difficult to find in huge asset data, and accuracy cannot be guaranteed. In recent years, with the development of machine learning technology, some asset classification and prediction methods based on data models have emerged, and attempts have been made to solve the problem by an automated means. However, these prior art solutions still have a number of inherent drawbacks, which make it difficult to meet the actual business requirements that are complex and variable, particularly in terms of: The characteristic weight is fixed, namely the existing model mostly adopts fixed characteristic weight, and cannot be dynamically adjusted according to the actual state (such as health degree and liveness degree) of the asset, the life cycle stage (such as new asset and overage asset) and the business value (such as ABC classification). The weight strategy of 'one-tool cut' causes that the model cannot be accurately adapted to the characteristics of different types of assets in different states, and the prediction accuracy is limited. Feature dimension singleness-most methods focus only on the financial dimension features (e.g., equity, depreciation) of the asset, lack of deep fusion of multi-dimensional features such as asset status (e.g., idle, disabled), time (e.g., age, life remaining), and business attributes (e.g., customer type, importance classification). This single view makes it difficult to construct a comprehensive representation of the asset, rendering the predicted outcome one-sided, inaccurate. Unifying data processing strategies, namely in the data preprocessing stage, especially in the missing value processing link, the prior art generally adopts a unified filling strategy (for example, filling with the average value of 0) for all the characteristics. The processing mode ignores the data distribution characteristics and business meanings of different characteristics, noise is often introduced, and the original distribution of data is distorted, so that the robustness and the prediction performance of a model are reduced. And the model is lack of a real-time optimization mechanism, once training is completed, parameters and characteristic weights of the model are fixed, and self-optimization and evolution cannot be performed according to newly generated asset data, business rule changes or prediction feedback results. This makes the model inadequate for long term effectiveness and adaptability, and difficult to support the dynamic management needs of enterprise assets. Therefore, the development of an inefficient asset prediction method which can dynamically adjust the feature weight, deeply fuse multidimensional features, implement differential data processing and have real-time optimization capability has urgent needs and great significance for solving the technical problems. Disclosure of Invention Aiming at the problems existing at present, the invention provides the low-efficiency asset prediction method and the system based on the multi-dimensional dynamic weight, which remarkably improve the accuracy and the adaptability of the low-efficiency asset identification by constructing a four-dimensional characteristic system, designing a dynamic weight calculation mechanism, realizing the real-time evaluation of the characteristic importance and the differential data processing. The technical scheme of the invention is as follows: an inefficient asset prediction method based on multi-dimensional dynamic weights, comprising the steps of: collecting and preprocessing data, collecting the original data of the assets, and cleaning and standardizing the data; multi-dimensional feature extraction, namely constructing and extracting a four-dimensional feature system comprising financial features, state features, time features and business features; asset status assessment, calculating an activity status score and a health score for an asset; dynamically adjusting the weight of each dimension characteristic based on the asset state evaluation result; fine ad