Low-efficiency asset prediction method and system based on multi-dimensional dynamic weight

Abstract

The invention discloses a low-efficiency asset prediction method and system based on a multi-dimensional dynamic weight, and relates to the technical field of enterprise asset management and prediction analysis. Comprising the steps of collecting original data of assets and performing data cleaning and standardization; constructing and extracting a four-dimensional feature system comprising financial features, state features, time features and business features; calculating an active state score and a health degree score of the assets; dynamically adjusting the weight of each dimension feature based on an asset state evaluation result; based on a feature importance evaluation result, adjusting the weight of each feature in the dimension to which the feature belongs; applying the weight obtained by dynamic calculation to the prediction model, and performing low-efficiency asset identification; and evaluating the prediction performance, and dynamically updating the weight according to a feedback result. By constructing a four-dimensional feature system and designing a dynamic weight calculation mechanism, feature importance real-time evaluation and differentiated data processing are realized, and the accuracy and adaptability of low-efficiency asset identification are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of enterprise asset management and predictive analysis technology, specifically to an inefficient asset prediction method and system based on multi-dimensional dynamic weights. Background Technology

[0002] As companies' asset size continues to expand and their asset structure becomes increasingly complex, efficiently and accurately identifying inefficient assets has become a core challenge in improving corporate asset management and utilization efficiency. Inefficient assets not only tie up significant amounts of corporate funds and space but also lead to escalating maintenance costs, severely impacting the company's operational efficiency.

[0003] Traditional, inefficient asset identification methods rely heavily on the experience of managers for manual judgment. This approach is highly subjective, inefficient, and struggles to uncover underlying patterns within massive amounts of asset data, compromising accuracy. In recent years, with the development of machine learning technology, several data model-based asset classification and prediction methods have emerged, attempting to solve this problem through automation. However, these existing solutions still suffer from many inherent flaws and fail to meet the complex and ever-changing demands of real-world business needs, specifically in the following aspects: Fixed feature weights: Existing models mostly use fixed feature weights, which cannot be dynamically adjusted according to the actual status of the asset (such as health and activity level), lifecycle stage (such as new asset and over-aged asset), and business value (such as ABC classification). This "one-size-fits-all" weight strategy makes it impossible for the model to accurately adapt to the characteristics of different types and statuses of assets, thus limiting the accuracy of predictions.

[0004] Limited Feature Dimensions: Most methods focus only on the financial dimensions of assets (such as net worth and depreciation), lacking in-depth integration of multi-dimensional features such as asset status (e.g., whether it is idle or discontinued), time (e.g., useful life and remaining useful life), and business attributes (e.g., customer type and importance classification). This singular perspective makes it difficult to construct a comprehensive asset profile, resulting in one-sided and inaccurate prediction results.

[0005] Standardized data processing strategies: In the data preprocessing stage, especially in handling missing values, existing technologies typically employ a uniform imputation strategy for all features (e.g., imputing the mean to 0). This approach ignores the data distribution characteristics and business implications of different features, often introducing noise and distorting the original data distribution, thereby reducing the robustness and predictive performance of the model.

[0006] Lack of real-time optimization mechanism: Once the model is trained, its parameters and feature weights remain fixed and cannot self-optimize and evolve based on newly generated asset data, changes in business rules, or prediction feedback results. This makes the model insufficient in long-term effectiveness and adaptability, making it difficult to support the dynamic management needs of enterprise assets.

[0007] Therefore, developing an inefficient asset prediction method that can dynamically adjust feature weights, deeply integrate multi-dimensional features, implement differentiated data processing, and possess real-time optimization capabilities is of urgent need and significant importance in solving the aforementioned technical challenges. Summary of the Invention

[0008] The purpose of this invention is to provide a method and system for predicting inefficient assets based on multi-dimensional dynamic weights, addressing the aforementioned problems. By constructing a four-dimensional feature system, designing a dynamic weight calculation mechanism, and realizing real-time assessment of feature importance and differentiated data processing, the accuracy and adaptability of inefficient asset identification are significantly improved.

[0009] The technical solution of the present invention is as follows: An inefficient asset prediction method based on multi-dimensional dynamic weights includes the following steps: Data acquisition and preprocessing: collecting raw asset data and performing data cleaning and standardization; Multi-dimensional feature extraction: Construct and extract a four-dimensional feature system that includes financial features, status features, time features, and business features; Asset status assessment, calculating the asset's activity level score and health score; The dimensional weights are dynamically adjusted based on the asset status assessment results. The feature-level weights are finely adjusted based on the feature importance assessment results, adjusting the weights of each feature within its respective dimension. Weighting and prediction: The dynamically calculated weights are applied to the prediction model to identify inefficient assets. Performance evaluation and weight update: Evaluate the predicted performance and dynamically update the weights based on the feedback results.

[0010] The above method automates and automates the entire process of inefficient asset prediction. Through a dynamic weight adjustment mechanism, the model can adapt to assets of different types and states, significantly improving the overall accuracy, adaptability, and efficiency of the prediction. Experiments show that this method achieves an accuracy improvement of over 17.5 percentage points compared to traditional fixed-weight methods on a large-scale test set, with a processing speed of up to 1000 data entries per second.

[0011] Furthermore, in the asset status assessment: The asset's activity level score is calculated using the following formula: calculate, , , , It is a binary variable, taking the value 1 or 0; The health score of an asset is determined by the formula: Calculate, where, Net asset value Cost value; For depreciation rate, The coefficients 0.6 and 0.4 were determined using a linear regression model. By quantifying the activity status score and health score, the real-time operating status and value loss of the asset are accurately reflected, providing a reliable and objective basis for subsequent dynamic weight adjustment. This overcomes the subjectivity of human experience judgment and makes the model's perception of the asset status more sensitive and accurate.

[0012] Furthermore, the dynamic adjustment of the dimension weights includes: The adjustment formula for the state feature weights is: ; The formula for adjusting the time feature weights is: ; The formula for adjusting the weights of business features is: ,in Scoring based on historical asset performance; The adjusted weights of each dimension are then normalized to ensure that the sum of the weights is 1.

[0013] The above method enables precise adjustment of model weights "depending on the asset." When an asset's health is low, the weight of state features is automatically increased; when an asset's activity is low, the weight of time features is automatically increased. This targeted adjustment improves the model's prediction accuracy by over 35% on assets with poor health or low activity, significantly enhancing the model's ability to discriminate across different scenarios.

[0014] Furthermore, the adjustment of the time feature weights includes a four-stage weight adjustment strategy: Based on the asset's useful life And whether or not they are over the age limit Set time weighting factor : like ,but ; like ,but ; like ,but ; like ,but .

[0015] The above methods overcome the shortcomings of traditional models in their insensitivity to changes in asset lifecycles. By distinguishing between new assets, middle-aged assets, old assets, and over-aged assets, and assigning them differentiated time weights, the model's prediction accuracy for assets of different lifespans is generally improved by 25%-40%, and it is particularly effective in identifying potentially inefficient assets due to aging.

[0016] Furthermore, the adjustment of business feature weights includes a three-stage weight adjustment strategy: According to the ABC classification of assets Set business weight factors : like ,but ; like ,but ; like ,but .

[0017] Using the methods described above, and based on the Pareto principle, the focus is on the management of high-value assets. By assigning higher weights to the business characteristics of Class A assets, the model's accuracy in identifying inefficiencies in high-value, core assets (Class A) has increased by more than 30%, helping companies prioritize the identification and disposal of inefficient assets that have the greatest impact on their business and optimize resource allocation.

[0018] Furthermore, the feature importance evaluation employs the XGBoost algorithm, and its feature importance score... The calculation formula is: , in, For the number of decision trees, For the number of nodes, Information gain for node splitting. For indicator functions; The formula for fine-tuning the feature-level weights is: .

[0019] By employing the methods described above, feature-level optimization within each dimension is further achieved on top of dimensional weight adjustments, thereby enhancing the model's prediction accuracy once again. Simultaneously, the feature importance assessment based on XGBoost significantly improves the model's interpretability, making the decision-making process transparent. The interpretation accuracy is more than 35% higher than that of black-box models, facilitating understanding and application by business personnel.

[0020] Furthermore, the preprocessing step includes a differential missing value handling strategy: For the remaining lifetime in the time characteristics Missing values ​​were filled with -1.3; Regarding the service life in the time characteristics Missing value, filled with 9.3; For missing values ​​of other numerical features, fill them with -1.

[0021] By using experimentally optimized special values ​​for imputation based on different feature types, the model effectively avoids the disruption of data distribution caused by traditional uniform imputation methods, significantly improving the robustness of the model when facing incomplete data. Even with a missing data rate of 20%, the model accuracy remains at 89.5%, a 21.3% improvement over traditional methods.

[0022] Furthermore, the four-dimensional feature system specifically includes: Financial characteristics: Net assets Depreciation amount Cost value Impairment Provision ; Status characteristics: Whether it is disabled Is it idle? Whether to wait for scrapping Are they over the age limit? Is it fully depreciated? ; Time characteristics: service life Remaining lifespan ; Business characteristics: Customer types Network attributes ABC classification , , .

[0023] Using the methods described above, a comprehensive and multi-dimensional asset profile was constructed, comprehensively evaluating assets from four dimensions: financial, status, time, and business, overcoming the limitations of single-dimensional features. Experimental results show that this four-dimensional feature system improves prediction accuracy by more than 35% compared to a single financial dimension, providing a solid data foundation for accurate prediction.

[0024] Furthermore, the initial basic weight allocation for each dimension of the feature is as follows: The basic weight for financial features is 0.3, the basic weight for status features is 0.3, the basic weight for time features is 0.2, and the basic weight for business features is 0.2.

[0025] The above method provides a scientific and reasonable initial benchmark for dynamic weight adjustment. This benchmark is derived from the analysis of massive historical data and A / B testing, avoiding the randomness and subjectivity of weight initialization, ensuring the stability and convergence speed of model training, and laying a solid foundation for all subsequent dynamic adjustment steps.

[0026] This application also includes an inefficient asset prediction system based on multi-dimensional dynamic weights, implementing an inefficient asset prediction method based on multi-dimensional dynamic weights, including: The data acquisition and preprocessing module collects raw asset data and performs data cleaning and standardization. The multi-dimensional feature extraction module constructs and extracts a four-dimensional feature system that includes financial features, status features, time features, and business features; The asset status assessment module calculates the asset's activity status score and health score. The dimension weight dynamic adjustment module dynamically adjusts the weights of each dimension feature based on the asset status assessment results. The feature-level weight fine-tuning module adjusts the weight of each feature within its respective dimension based on the feature importance assessment results. The weighting and prediction module applies dynamically calculated weights to the prediction model to identify inefficient assets. The performance evaluation and weight update module evaluates the predicted performance and dynamically updates the weights based on the feedback results.

[0027] Compared with existing technologies, the advantages of this invention are: 1. Significantly Improved Prediction Accuracy: Through a dynamic weight adjustment mechanism and a multi-dimensional feature fusion strategy, the model can better adapt to asset data of different types and states. On a test dataset containing 10,000+ assets, the prediction accuracy reached 92.3%, an improvement of 22.2 percentage points compared to the 70.1% of the traditional fixed-weight method; 2. Enhanced Adaptability: The system can dynamically adjust feature weights based on the actual state and changing trends of assets, maintaining high accuracy in identifying new assets, middle-aged assets, elderly assets, and assets past their prime. Experimental data shows an accuracy rate of 88.7% for new assets (used for <1 year), 93.5% for middle-aged assets (1 ≤ used for <5 years), 94.2% for elderly assets (used for ≥5 years), and 91.8% for assets past their prime, with an overall fluctuation of no more than 6 percentage points, demonstrating excellent adaptability. 3. Improved Robustness: A differentiated missing value handling strategy ensures stable performance even with incomplete or anomalous data. On a test set with a 20% missing data rate, the model accuracy was 89.5%, only 2.8 percentage points lower than the complete dataset, demonstrating a 21.3% improvement in robustness compared to traditional methods. 4. Improved efficiency: The entire process from feature extraction and weight calculation to model prediction is automated, with a processing speed of 1,000 items per second. This is more than 500 times more efficient than manual review, greatly reducing human intervention and improving the efficiency of identifying inefficient assets. 5. Enhanced Interpretability: By assessing feature importance, the model's decision-making process becomes more transparent, and the prediction results are easier to understand and apply. Experiments show that after processing with the method of this invention, the model's decision interpretation accuracy reaches 94.6%, which is more than 35% higher than that of traditional black-box models. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the method described in this application.

[0029] Figure 2 A schematic diagram of the dynamic weight calculation mechanism. Detailed Implementation

[0030] It should be noted that relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0031] The features and performance of the present invention will be further described in detail below with reference to embodiments.

[0032] Please see Figure 1 and Figure 2 An inefficient asset prediction method based on multi-dimensional dynamic weights includes the following steps: Dynamic weight calculation process: Data Acquisition and Preprocessing: Collect raw asset data and perform preprocessing operations such as data cleaning and standardization; Multi-dimensional feature extraction: Features are extracted from four dimensions: financial, status, time, and business. Asset status assessment: Calculate the asset's activity status score and health score; Dynamic adjustment of dimensional weights: The weights of each dimension are adjusted based on the asset status assessment results; Feature-level weight fine-tuning: Adjust the weights of each feature based on the feature importance assessment results; Weighting Application and Prediction: Dynamically calculated weights are applied to the prediction model to identify inefficient assets; Performance evaluation and weight update: Evaluate the predicted performance and dynamically update the weights based on the feedback results.

[0033] Multi-dimensional Feature System Construction: A four-dimensional feature system is constructed, encompassing financial, status, time, and business characteristics. This system comprehensively covers all aspects of asset attributes, providing a robust data foundation for identifying inefficient assets. The basic table of dimensional attributes is shown below: Table 1. Basic Table of Dimensional Attributes

[0034] Dynamic weighting mechanism: Based on the current state and historical performance of the asset, the weights of each dimension of features are dynamically calculated to achieve adaptive adjustment of feature weights. Specifically, this includes: Basic Weight Allocation: Initial weights are set for each dimension of features. These initial weights are derived from empirical data obtained through statistical analysis of seven years of historical asset data from a specific company in the telecom industry. The evaluation data volume exceeds 6,000,000+ historical asset data points, and feature importance analysis and A / B testing were conducted. This provides a scientifically sound initial benchmark for subsequent dynamic weight adjustments, ensuring that each dimension of feature receives an initial weight consistent with its basic importance in the prediction model, avoiding arbitrariness and subjectivity in weight allocation.

[0035] Table 2 Basic Weight Allocation

[0036] The basic weight allocation was derived through feature importance analysis and A / B testing of over 10,000 historical asset data points. The basic weights for each dimension were set as follows: financial features 0.3, status features 0.3, time features 0.2, and business features 0.2. These weight values ​​were not subjectively set, but rather represent the optimal allocation scheme verified through extensive experimental data.

[0037] Asset Status Assessment: This involves calculating the asset's Activity Score and Health Score. These scores are derived through correlation analysis and regression modeling of historical performance data from over 1,000,000 assets in different states. They accurately reflect the asset's actual operating status and degree of value loss. Accurately quantifying the asset's current operating status and value loss provides a core basis for subsequent dynamic weight adjustments, enabling the model to intelligently adjust the importance of each dimension's features based on the asset's actual state.

[0038] Active state score calculation formula: , in, , , , The variable is a binary variable, taking the value 1 or 0, where 1 indicates yes and 0 indicates no. The coefficients 0.3, 0.5, 0.8, and 0.7 are determined by the Pearson correlation coefficient, which shows the degree of influence of different states on asset inefficiency, based on the relationship between the asset's idle / discontinued / pending scrapping / disabled status and the asset's inefficiency.

[0039] Health score calculation formula: , in, Net asset value Cost value; For depreciation rate, The coefficients 0.6 and 0.4 were determined through a linear regression model, reflecting the contribution weights of the net asset value ratio and the residual value rate to the asset health status.

[0040] The coefficients in the formula are derived through correlation analysis and regression modeling of historical performance data from over 5000 assets in different states. The magnitude of the coefficients reflects the degree of impact of different states on asset inefficiency. This accurately reflects the actual operating status and value loss of assets, providing reliable data support for dynamic weight adjustments.

[0041] Dynamic Dimension Weight Adjustment: Based on asset status assessment results, the weights of each dimension feature are readjusted according to three dimensions: status characteristics, time characteristics, and business characteristics. Through optimization through more than 10 rounds of cross-validation experiments, the model can intelligently adjust the importance of each dimension feature based on the actual state of the asset. The importance of each dimension feature is dynamically adjusted according to the different states, service lives, and business value of the asset, ensuring that the model can better adapt to asset data of different types and states. Specifically, this includes: The formula for adjusting the weights of status features adjusts the weights of status features in real time based on the asset's health score: , Adjusted state feature weight = base weight * (1.5 - health score of healthy assets (value between 0 and 1)). When the asset health is lower, the weight of the state feature is increased because the state index of unhealthy assets is more critical for judging their inefficiency. Through experiments, it is verified that a coefficient of 1.5 can improve the prediction accuracy of the model by more than 35% on assets with a health score below 0.3.

[0042] The time feature weighting adjustment formula adjusts the weights of status features in real time based on the asset's activity status score: , The adjusted state feature weight = base weight * (1.2 + active state score (value between 0 and 1)). When the asset activity is lower, the weight of the time feature is increased because the aging speed of inactive assets has a greater impact on its inefficiency. The coefficient 1.2 was determined through experiments and can balance the relationship between time features and active state.

[0043] The business feature weight adjustment formula adjusts the weights of status features in real time based on the asset's historical scores: , in, For historical asset performance, the weight of business characteristics is increased as the historical performance of an asset worsens, because the business attributes of assets with poor business performance are more valuable for judging inefficiency.

[0044] After normalization, the final weight value = dynamically adjusted dimension weight value / total weights: Ensure that the sum of the weights of each dimension is 1.

[0045] The aforementioned adjustment mechanism, optimized through more than 10 rounds of cross-validation experiments, significantly improves the model's prediction accuracy across assets in different states. Increasing the weight of state features as asset health decreases improves prediction accuracy by over 35% for assets with health scores below 0.3. A four-stage weight adjustment based on asset age improves prediction accuracy by 25%-40% across assets of different ages. A three-stage weight adjustment based on ABC asset classification improves prediction accuracy by over 30% for Class A assets.

[0046] Four-stage weight adjustment strategy for time features: Based on the asset's useful life, a four-stage time characteristic weighting adjustment strategy was designed, explaining the purpose of setting different weighting factors according to the asset's useful life, and providing support for more accurately identifying the inefficiency of assets at different life cycle stages.

[0047] ， in For service life, For assets deemed over-aged, the time factor has a relatively small impact when the asset's useful life is less than one year, with a weighting factor of 0.5; for middle-aged assets with a useful life of 1-5 years, the time factor has a moderate impact, with a weighting factor of 1.0; for elderly assets with a useful life of 5 years or more, the aging effect is significant, with a weighting factor of 1.5; for over-aged assets, the time factor becomes the main cause of inefficiency, with a weighting factor of 2.0. Experimental results show that the four-stage weight adjustment improves the prediction accuracy of assets of different ages by 25%-40% compared to the traditional fixed weight approach.

[0048] Three-stage weighting strategy for business characteristics: Based on the ABC classification of assets, design a three-stage business characteristic weight adjustment strategy: , Among them, Class A assets have high business value, and their business characteristics are more important for inefficiency judgment, with a weight factor of 2.0; Class B assets have medium business value, with a weight factor of 1.0; and Class C assets have low business value, with a weight factor of 0.5. Experimental data shows that the three-stage weight adjustment improved the model's prediction accuracy for Class A assets by more than 30%.

[0049] Real-time assessment of feature importance: This invention employs the feature importance assessment function of the XGBoost algorithm to analyze the contribution of each feature to the prediction results in real time, providing data support for model optimization. Although the XGBoost algorithm itself is well-known, this invention creatively combines it with a dynamic weight adjustment mechanism in the preprocessing stage. By using the feature importance assessment results to guide the optimization of the weight adjustment strategy, a unique data processing workflow is formed.

[0050] Feature importance calculation: The contribution of each feature is evaluated in real time using feature importance scores automatically calculated during XGBoost model training. The formula for calculating the feature importance score is: , in, For the number of decision trees, The number of nodes in each decision tree, The information gain brought about by node splitting. For indicator functions, The feature importance score is calculated for the XGBoost model, representing the contribution of the feature to the prediction of inefficient assets.

[0051] Dynamic adjustment of model parameters: Based on the feature importance assessment results, the model parameters are dynamically adjusted to optimize prediction performance.

[0052] Differential missing value handling strategy: Differentiated missing value handling methods are designed for different types of features to ensure data quality and model robustness. This strategy is not simply missing value imputation, but rather based on in-depth analysis of feature distribution characteristics and business scenarios. Extensive experiments have verified that specific imputation values ​​can maximize model performance.

[0053] Handling missing values ​​in time features: Filling them with special marker values; Remaining lifespan ( Missing data: Filled with -1.3 (experimental verification shows that this value is outside the normal remaining lifetime range, and compared with commonly used filling values ​​such as -1, it can improve the accuracy of the model in handling missing data by more than 15%). Service life ( Missing data: Filled with 9.3 (also based on experimental data, this value is significantly different from the average lifespan of assets and can help the model better identify special cases of missing data).

[0054] Other missing numerical features are handled by using a uniform default value of -1 for filling. This filling strategy was determined through comparative experiments, which ensures data integrity while avoiding excessive interference with model training.

[0055] Feature-level weight refinement further details the importance of specific features within each dimension, ensuring that the weight of each feature matches its actual contribution to the prediction, thus improving prediction accuracy. This ensures that important features receive higher weights and less important features receive lower weights, further enhancing prediction accuracy. The adjustment formula is: .

[0056] This application also includes an inefficient asset prediction system based on multi-dimensional dynamic weights, comprising: The data acquisition and preprocessing module collects raw asset data and performs data cleaning and standardization. The multi-dimensional feature extraction module constructs and extracts a four-dimensional feature system that includes financial features, status features, time features, and business features; The asset status assessment module calculates the asset's activity status score and health score. The dimension weight dynamic adjustment module dynamically adjusts the weights of each dimension feature based on the asset status assessment results. The feature-level weight fine-tuning module adjusts the weight of each feature within its respective dimension based on the feature importance assessment results. The weighting and prediction module applies dynamically calculated weights to the prediction model to identify inefficient assets. The performance evaluation and weight update module evaluates the predicted performance and dynamically updates the weights based on the feedback results.

[0057] Comparative experiment: Experimental objective: To verify the performance advantages of dynamic weight adjustment mechanism compared with traditional fixed weight method.

[0058] Experimental environment: The experimental environment was built using Python data analysis frameworks and machine learning libraries.

[0059] Experimental dataset: Small-scale test set: Contains 10,000+ historical asset data, covering assets of different types and states.

[0060] Large-scale test set: Contains 6,000,000+ asset data points to verify the model's performance on massive datasets.

[0061] Comparison method: Experimental group: Employed the dynamic weight adjustment mechanism of this invention. Control group: Employed the traditional fixed weight method.

[0062] Detailed explanation of the experimental verification process: Phase 1: Data Preparation and Preprocessing; 1. Data Collection: Collect historical asset data of the enterprise, including basic asset information, financial data, status data, time data, and business data.

[0063] 2. Data Cleaning: Handling outliers and missing values ​​to ensure data quality. A differentiated missing value handling strategy is adopted: missing remaining lifetime is filled with -1.3, and missing other numerical features are filled with -1.

[0064] 3. Data partitioning: The dataset is divided into training and test sets in a 7:3 ratio to ensure the reliability of the experiment.

[0065] Phase Two: Model Training and Optimization; 1. Feature Engineering: Construct a four-dimensional feature system: financial features, status features, time features, and business features.

[0066] Calculate derived characteristics such as useful life, remaining useful life, and depreciation rate.

[0067] 2. Determining the basic weights: Perform feature importance analysis on 10,000+ historical asset data.

[0068] Design multiple A / B tests to verify the effects of different combinations of basic weights.

[0069] Determine the optimal basic weight allocation: financial characteristics 0.3, state characteristics 0.3, time characteristics 0.2, business characteristics 0.2.

[0070] 3. Optimization of dynamic weight adjustment coefficients: Correlation analysis and regression modeling were performed on historical performance data of 5000+ assets in different states.

[0071] Design more than 10 rounds of cross-validation experiments and optimize the coefficients in each adjustment formula.

[0072] Determine the optimal combination of coefficients, such as coefficient 1.5 in the state feature weight adjustment and coefficient 1.2 in the time feature weight adjustment.

[0073] Phase 3: Model Performance Evaluation; 1. Evaluation of the experimental group: The model is trained using a dynamic weight adjustment mechanism.

[0074] Evaluate model performance on the test set and record metrics such as prediction accuracy, precision, and recall.

[0075] 2. Control group assessment: The model is trained using the traditional fixed-weight method.

[0076] Evaluate model performance on the same test set and record the same metrics.

[0077] 3. Detailed scenario assessment: The predictive performance of assets with different health levels is evaluated.

[0078] The predictive performance of assets with different useful lives is evaluated.

[0079] Evaluate the predictive performance of assets with different business values ​​(ABC classification).

[0080] The robustness of the model in scenarios with missing data is evaluated.

[0081] Analysis of experimental results: Overall performance comparison: On a small test set with 10,000+ assets: Experimental group (dynamic weights): prediction accuracy reached 92.3%; control group (fixed weights): prediction accuracy was 70.1%. Improvement: 22.2 percentage points.

[0082] On a large-scale test set with 6,000,000+ assets: Experimental group (dynamic weights): prediction accuracy reached 80%; control group (fixed weights): prediction accuracy was 62.5%. Improvement: 17.5 percentage points.

[0083] Performance comparison in specific scenarios: Different health assets: For assets with a health score below 0.3, the dynamic weighting method improves prediction accuracy by more than 35%. The lower the health score, the more obvious the advantage of the dynamic weighting method becomes.

[0084] Assets with different useful lives: New assets (<1 year): Prediction accuracy improved by 25%; Mid-life assets (1-5 years): Prediction accuracy improved by 30%; Old age assets (≥5 years): Prediction accuracy improved by 35%; Over-aged assets: Prediction accuracy improved by 40%; Different business value assets: Class A assets: Prediction accuracy improved by more than 30%.

[0085] Asset Class B: Prediction accuracy improved by approximately 20%.

[0086] Asset Class C: Prediction accuracy improved by approximately 15%.

[0087] Missing data scenarios: On a test set with a missing data rate of 20%, the dynamic weighting method achieved an accuracy of 89.5%, only 2.8 percentage points lower than the complete dataset. Robustness was improved by 21.3% compared to traditional methods.

[0088] Experimental conclusion: The dynamic weight adjustment mechanism is significantly better than the traditional fixed weight method: the prediction accuracy of the dynamic weight method is significantly improved in various scenarios.

[0089] Dynamic weighting methods are more adaptable: they can automatically adjust feature importance based on different asset states, years of use, and business value, adapting to diverse data scenarios.

[0090] Dynamic weighting methods are more robust: they maintain high prediction accuracy even when faced with missing data.

[0091] Dynamic weighting methods have good scalability: they still maintain a significant performance advantage on large-scale datasets.

[0092] The scientific rigor of experimental verification: Large-sample validation: The experiments used large-scale datasets containing 10,000+ and 6,000,000+ assets to ensure the reliability of the results.

[0093] Multiple rounds of cross-validation: Through more than 10 rounds of cross-validation experiments, the model parameters are optimized to avoid overfitting.

[0094] Multi-dimensional evaluation: The model performance is evaluated from multiple dimensions such as accuracy, precision, recall, and robustness.

[0095] Detailed scenario testing: Conduct detailed tests on assets of different types and states to comprehensively verify model performance.

[0096] Objective comparative analysis: Fair comparison with traditional methods, using the same dataset and evaluation metrics to ensure the objectivity of the comparison results.

[0097] The embodiments described above merely illustrate specific implementation methods of this application, and while the descriptions are detailed and specific, they should not be construed as limiting the scope of protection of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the technical solution of this application, and these modifications and improvements all fall within the scope of protection of this application.

Claims

1. An inefficient asset prediction method based on multi-dimensional dynamic weights, characterized in that, Includes the following steps: Data acquisition and preprocessing: collecting raw asset data and performing data cleaning and standardization; Multi-dimensional feature extraction: Construct and extract a four-dimensional feature system that includes financial features, status features, time features, and business features; Asset status assessment, calculating the asset's activity level score and health score; The dimensional weights are dynamically adjusted based on the asset status assessment results. The feature-level weights are finely adjusted based on the feature importance assessment results, adjusting the weights of each feature within its respective dimension. Weighting and prediction: The dynamically calculated weights are applied to the prediction model to identify inefficient assets. Performance evaluation and weight update: Evaluate the predicted performance and dynamically update the weights based on the feedback results.

2. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, In the asset status assessment: The asset's activity level score is calculated using the following formula: calculate, , , , It is a binary variable, taking the value 1 or 0; The health score of an asset is determined by the formula: Calculate, where, Net asset value Cost value; For depreciation rate, The coefficients 0.6 and 0.4 were determined using a linear regression model.

3. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 2, characterized in that, The dynamic adjustment of the dimension weights includes: The adjustment formula for the state feature weights is: ; The formula for adjusting the time feature weights is: ; The formula for adjusting the weights of business features is: ,in Scoring based on historical asset performance; The adjusted weights of each dimension are then normalized to ensure that the sum of the weights is 1.

4. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The adjustment of the weights for time features further includes a four-stage weight adjustment strategy: Based on the asset's useful life and whether or not they are over the age limit Set time weighting factor : like ,but ; like ,but ; like ,but ; like ,but .

5. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The adjustment of business feature weights further includes a three-stage weight adjustment strategy: According to the ABC classification of assets Set business weight factors : like ,but ; like ,but ; like ,but .

6. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The feature importance evaluation uses the XGBoost algorithm. Its feature importance score The calculation formula is: ， in, For the number of decision trees, For the number of nodes, Information gain for node splitting. For indicator functions; The formula for fine-tuning the feature-level weights is: .

7. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The preprocessing step includes a differential missing value handling strategy: For the remaining lifetime in the time characteristics Missing values ​​were filled with -1.3; Regarding the service life in the time characteristics Missing value, filled with 9.3; For missing values ​​of other numerical features, fill them with -1.

8. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The four-dimensional feature system specifically includes: Financial characteristics: Net assets Depreciation amount Cost value Impairment Provision ; Status characteristics: Whether it is disabled Is it idle? Whether to wait for scrapping Are they over the age limit? Is it fully depreciated? ; Time characteristics: service life Remaining lifespan ; Business characteristics: Customer types Network attributes ABC classification , , .

9. The inefficient asset prediction method based on multi-dimensional dynamic weights according to claim 1, characterized in that, The initial basic weight allocation for each dimension of the feature is as follows: The basic weight for financial features is 0.3, the basic weight for status features is 0.3, the basic weight for time features is 0.2, and the basic weight for business features is 0.

2.

10. An inefficient asset prediction system based on multi-dimensional dynamic weights, characterized in that, Implementing an inefficient asset prediction method based on multi-dimensional dynamic weights as described in any one of claims 1-9, comprising: The data acquisition and preprocessing module collects raw asset data and performs data cleaning and standardization. The multi-dimensional feature extraction module constructs and extracts a four-dimensional feature system that includes financial features, status features, time features, and business features; The asset status assessment module calculates the asset's activity status score and health score. The dimension weight dynamic adjustment module dynamically adjusts the weights of each dimension feature based on the asset status assessment results. The feature-level weight fine-tuning module adjusts the weight of each feature within its respective dimension based on the feature importance assessment results. The weighting and prediction module applies dynamically calculated weights to the prediction model to identify inefficient assets. The performance evaluation and weight update module evaluates the predicted performance and dynamically updates the weights based on the feedback results.

Images