Search

CN-122022698-A - Automatic summarizing and asset valuation checking method for material handling conditions

CN122022698ACN 122022698 ACN122022698 ACN 122022698ACN-122022698-A

Abstract

The invention relates to the technical field of asset valuation, and discloses an automatic summarization and asset valuation verification method for material handling conditions, which is characterized in that a prior parameter distribution is generated by acquiring attribute feature vectors of new warehouse-in materials, screening a reference class set from a mature material class library, merging multiple regression model parameter sets of reference classes, and extracting a limited history of the newly-stored materials to generate a multi-element sample data set, generating posterior regression parameter distribution based on Bayesian fusion, calculating a prediction residual value rate and a prediction uncertainty, generating a prediction estimated value, comparing the prediction estimated value with a declaration estimated value, and outputting a verification result, thereby solving the problem of insufficient multi-factor residual value prediction accuracy under the condition of small samples.

Inventors

  • XIANG QIAN
  • YANG HAI

Assignees

  • 中国铁塔股份有限公司浙江省分公司

Dates

Publication Date
20260512
Application Date
20260414

Claims (10)

  1. 1. The automatic summarization and asset valuation verification method for the material disposal condition is characterized by comprising the following steps of: Acquiring attribute feature vectors of the newly-put materials, wherein the attribute feature vectors comprise material categories, technical parameters, applicable scenes and a list of influence factor types; Acquiring attribute feature vectors of all categories in a mature material category library and a trained multiple regression model parameter set, wherein the multiple regression model parameter set comprises all factor regression coefficients and interaction item parameters; calculating the similarity between the attribute feature vector of the new warehouse-in material and each attribute feature vector in the mature material class library, and screening the class of K bits before the similarity sorting as a reference class set; Weighting and fusing multiple regression model parameter sets of all the classes in the reference class set to generate prior parameter distribution of a new warehouse-in material multiple regression model, wherein the fusion weight is positively correlated with the similarity; acquiring history treatment records of newly-stored materials, extracting multi-factor characteristic values and actual residual value rates of each history treatment record, and generating a multi-element sample data set; constructing a likelihood function based on the multi-element sample data set, and performing Bayesian fusion on the likelihood function and the prior parameter distribution to generate posterior regression parameter distribution; acquiring multi-factor characteristic vectors of materials to be estimated, substituting the multi-factor characteristic vectors into posterior mean parameters of posterior regression parameter distribution, and calculating a prediction residual value rate; and generating a predicted estimated value according to the predicted residual value rate and the purchase price of the material to be estimated, comparing the declared estimated value with the predicted estimated value, and outputting a check result.
  2. 2. The automatic summarizing and asset estimation verification method for material handling situations according to claim 1 is characterized in that an attribute feature vector is constructed by performing single-hot encoding on a category attribute, performing average normalization processing on a numerical attribute based on an extremely bad value, and splicing all encoded attribute values to form a vector with a fixed dimension, wherein the single-hot encoding is realized by constructing a binary vector with a length of M for the category attribute with M possible values, assigning 1 at a position corresponding to an actual value of the category attribute, and assigning 0 at the rest positions.
  3. 3. The automatic summarization and asset estimation verification method for material handling situations according to claim 1, wherein the residual value rate output of the multiple regression model is composed of intercept terms, the sum of products of main effect terms and corresponding regression coefficients of each influence factor, the sum of products of interaction terms and corresponding interaction term parameters between any two of each influence factors, and random error terms, and the mature material category refers to a material category in which the number of historical handling records reaches a preset sample size threshold, and the preset sample size threshold is 5 to 10 times of the total number of parameters in the multiple regression model parameter set.
  4. 4. The automatic summarization and asset estimation verification method for material handling situations according to claim 1 is characterized in that cosine similarity calculation is adopted for similarity calculation, each dimension of attribute feature vectors is weighted when the similarity is calculated, weights are obtained by counting absolute values of regression coefficients of multiple regression models in a mature material class library and averaging the absolute values of the regression coefficients corresponding to the same attribute dimension, and the value range of reference class number K is an integer between 3 and 10 and is determined according to the scale of the mature material class library.
  5. 5. The automatic summarization and asset estimation verification method for material handling situations according to claim 1, wherein the prior parameter distribution is in a multi-element normal distribution form, the prior mean value is obtained by weighting and averaging multiple regression model parameter sets of all classes in a reference class set according to fusion weights, the fusion weights are obtained by normalizing similarity between all the reference classes and new warehouse-in materials, the prior covariance matrix is obtained by weighting and summing the outer products of differences between the multiple regression model parameter sets of all the reference classes and the prior mean value according to the fusion weights, and a regularization term is superimposed, and the regularization term is the product of a regularization coefficient and an identity matrix.
  6. 6. The method for automatically summarizing and verifying asset valuation according to claim 1, wherein the multi-factor characteristic values comprise service life, use intensity index, maintenance score and environmental factor, and the multi-factor characteristic vector is constructed by splicing values of all influence factors subjected to normalization processing according to a preset sequence to be used as a main effect item, calculating a product between any two of the influence factors to be used as an interaction item to splice, and adding a constant item at the head of the multi-factor characteristic vector to correspond to an intercept parameter.
  7. 7. The automatic summarization and asset estimation verification method according to claim 1, wherein posterior regression parameters are distributed in a multi-element normal distribution, a posterior covariance matrix is obtained by adding and inverting a matrix obtained by dividing the sum of the external products of multi-factor eigenvectors in a priori covariance matrix and multi-element sample data set by residual variances, and a posterior mean is obtained by multiplying the product of the posterior covariance matrix and the prior covariance matrix by the prior mean and the product of multiplying the posterior covariance matrix and the multi-factor eigenvectors by the corresponding actual residual ratios.
  8. 8. The method for automatically summarizing and checking asset valuation according to claim 1, wherein the uncertainty of prediction is calculated based on a posterior covariance matrix of posterior regression parameter distribution, the uncertainty of prediction is square root of products of multi-factor eigenvectors, posterior covariance matrix and multi-factor eigenvectors of the asset to be evaluated, an uncertainty interval is determined according to products of the uncertainty of prediction and purchase price, a check result is passed if the declaration valuation falls within the uncertainty interval, and a check result is to be checked if the declaration valuation exceeds the uncertainty interval.
  9. 9. The automatic summarization and asset estimation verification method for material handling situations according to claim 1, wherein contribution degrees of all influence factors to a predicted residual value rate are synchronously output when verification results are output, and for a j-th influence factor, the main effect contribution degree is a product of a regression coefficient corresponding to the j-th influence factor in a posterior mean vector and a j-th influence factor normalization value of a material to be estimated, the interaction item contribution degree is a sum of all interaction item parameters participated by the j-th influence factor and a sum of corresponding two influence factor normalization value multiplication products, and the total contribution degree is a sum of the main effect contribution degree and the interaction item contribution degree.
  10. 10. A material handling situation automatic summarization and asset valuation verification system for performing the material handling situation automatic summarization and asset valuation verification method of any one of claims 1 to 9, comprising: The attribute feature acquisition module is used for acquiring attribute feature vectors of the new warehouse-in materials; The mature category library management module is used for acquiring attribute feature vectors of various categories in the mature material category library and a trained multiple regression model parameter set; The similarity calculation and screening module is used for calculating the similarity between the attribute feature vector of the new warehouse-in material and each category attribute feature vector in the mature material category library, and screening the reference category set; The prior parameter fusion module is used for carrying out weighted fusion on multiple regression model parameter sets of all the classes in the reference class set to generate prior parameter distribution; the sample data set generation module is used for acquiring the history handling record of the newly-put materials and generating a multi-element sample data set; The Bayesian fusion module is used for constructing a likelihood function based on the multi-element sample data set, carrying out Bayesian fusion on the likelihood function and the prior parameter distribution, and generating posterior regression parameter distribution; The residual value prediction module is used for substituting the multi-factor characteristic vector of the material to be estimated into posterior mean value parameters of posterior regression parameter distribution to calculate a predicted residual value rate; And the estimated value verification module is used for generating a predicted estimated value according to the predicted residual value rate and the purchase price, comparing the declared estimated value with the predicted estimated value and outputting a verification result.

Description

Automatic summarizing and asset valuation checking method for material handling conditions Technical Field The invention relates to the technical field of asset valuation, in particular to an automatic summarization and asset valuation verification method for material handling conditions. Background In an enterprise asset disposal scenario, the residual values of lead-acid batteries and lithium batteries decay while being influenced by multiple factors such as time of use, charge-discharge cycle times, maintenance conditions, ambient temperature, and the like. In the prior art, two residual value prediction methods exist, namely, a first method finds a reference category through battery attribute feature similarity matching, and shifts fluctuation features of the reference category as priori knowledge to be fused with limited historical data in a Bayesian way, the method can solve the problem of small sample estimation, but only can shift single fluctuation features, and cannot embody the differential influence of multiple factors on the residual value, and a second method predicts the residual value rate through a multiple nonlinear regression model by taking multiple factors such as charge and discharge cycle times, capacity attenuation rate, environmental factors and the like as input variables, and the method can capture the interaction influence of multiple factors, but needs sufficient historical samples to train multiple parameters of the multiple nonlinear regression model. The newly-put battery materials face multiple parameters of which the historical data quantity is insufficient to reliably train the multiple nonlinear regression models, and single fluctuation characteristics of the simple migration similar materials cannot embody two challenges of differentiating influence of multiple factors on residual values, so that accurate multi-factor residual value prediction cannot be realized under the condition of small samples by using the two methods independently. Disclosure of Invention The invention provides an automatic summarizing and asset estimation verification method for material handling situations, which solves the technical problems that the residual value prediction precision is low, the prior knowledge of mature categories cannot be fully utilized and a prediction uncertainty quantization mechanism is lacked due to insufficient historical data of new warehouse-in materials in the related technology. The invention provides a method for automatically summarizing material handling situations and checking asset valuation, which comprises the following steps: Acquiring attribute feature vectors of the newly-put materials, wherein the attribute feature vectors comprise material categories, technical parameters, applicable scenes and a list of influence factor types; Acquiring attribute feature vectors of all categories in a mature material category library and a trained multiple regression model parameter set, wherein the multiple regression model parameter set comprises all factor regression coefficients and interaction item parameters; calculating the similarity between the attribute feature vector of the new warehouse-in material and each attribute feature vector in the mature material class library, and screening the class of K bits before the similarity sorting as a reference class set; Weighting and fusing multiple regression model parameter sets of all the classes in the reference class set to generate prior parameter distribution of a new warehouse-in material multiple regression model, wherein the fusion weight is positively correlated with the similarity; acquiring history treatment records of newly-stored materials, extracting multi-factor characteristic values and actual residual value rates of each history treatment record, and generating a multi-element sample data set; constructing a likelihood function based on the multi-element sample data set, and performing Bayesian fusion on the likelihood function and the prior parameter distribution to generate posterior regression parameter distribution; acquiring multi-factor characteristic vectors of materials to be estimated, substituting the multi-factor characteristic vectors into posterior mean parameters of posterior regression parameter distribution, and calculating a prediction residual value rate; and generating a predicted estimated value according to the predicted residual value rate and the purchase price of the material to be estimated, comparing the declared estimated value with the predicted estimated value, and outputting a check result. The construction method of the attribute feature vector comprises the steps of carrying out single-heat coding on the category type attribute, adopting average normalization processing based on extremely bad value on the numerical value type attribute, and splicing all the coded attribute values to form a vector with fixed dimension, wherein the single-heat coding is carried out in th