CN115841346A - Asset derating prediction method and system for business decisions - Google Patents
Asset derating prediction method and system for business decisions Download PDFInfo
- Publication number
- CN115841346A CN115841346A CN202310153438.2A CN202310153438A CN115841346A CN 115841346 A CN115841346 A CN 115841346A CN 202310153438 A CN202310153438 A CN 202310153438A CN 115841346 A CN115841346 A CN 115841346A
- Authority
- CN
- China
- Prior art keywords
- asset
- monthly
- matrix
- data
- scale
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 52
- 239000013598 vector Substances 0.000 claims description 169
- 239000011159 matrix material Substances 0.000 claims description 164
- 238000012549 training Methods 0.000 claims description 88
- 238000000605 extraction Methods 0.000 claims description 77
- 230000006870 function Effects 0.000 claims description 64
- 230000015654 memory Effects 0.000 claims description 58
- 238000003062 neural network model Methods 0.000 claims description 50
- 230000002457 bidirectional effect Effects 0.000 claims description 42
- 230000003442 weekly effect Effects 0.000 claims description 30
- 238000012545 processing Methods 0.000 claims description 12
- 238000012937 correction Methods 0.000 claims description 3
- 238000013135 deep learning Methods 0.000 abstract description 9
- 238000005516 engineering process Methods 0.000 abstract description 7
- 238000013473 artificial intelligence Methods 0.000 abstract description 6
- 230000008859 change Effects 0.000 description 25
- 238000010586 diagram Methods 0.000 description 16
- 238000009826 distribution Methods 0.000 description 12
- 238000003860 storage Methods 0.000 description 11
- 238000004590 computer program Methods 0.000 description 9
- 230000007787 long-term memory Effects 0.000 description 6
- 230000006403 short-term memory Effects 0.000 description 6
- 238000007792 addition Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 5
- 230000000116 mitigating effect Effects 0.000 description 5
- 230000009467 reduction Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 4
- 235000019580 granularity Nutrition 0.000 description 4
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000002360 preparation method Methods 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000008034 disappearance Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000007405 data analysis Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 230000006798 recombination Effects 0.000 description 1
- 238000005215 recombination Methods 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
- 230000011218 segmentation Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an asset valuation prediction method and an asset valuation prediction system for business decision, which adopt an artificial intelligence technology based on deep learning to carry out prediction through an encoder and a decoder, namely, a multi-scale implicit association characteristic of a historical trading price of a first asset in a second-hand trading market under different time spans is extracted based on the encoder, and decoding regression is carried out based on the decoder to improve the prediction accuracy of the recoverable amount of the first asset. In this way, an accurate predictive estimate of the asset recoverable amount can be made.
Description
Technical Field
The present application relates to the field of data analysis technology, and more particularly, to an asset derating prediction method and system for business decisions.
Background
Asset valuation belongs to the category of profit and loss in accounting. The asset deduction value is the corresponding loss confirmed by the preparation for the asset deduction loss counted by judging that the recoverable amount of the asset is lower than the account value of the asset through the testing of the asset on the balance table day of the asset, namely, the asset deduction loss = the account value of the asset-the recoverable amount of the asset.
Of the most important in making asset derating forecasts is the forecasting and assessment of the amount of recoverable assets. Current practice is time-based wear methods, but this practice ignores the value of the property of the asset's market, resulting in large deviations in predictions and assessments.
Therefore, an optimized asset derating prediction scheme for business decisions is desired.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides an asset derating prediction method and system for business decision, which adopt an artificial intelligence technology based on deep learning to carry out prediction through an encoder and a decoder, namely, a multi-scale implicit association characteristic of a historical trading price of a first asset in a second-hand trading market under different time spans is extracted based on the encoder, and decoding regression is carried out based on the decoder to improve the prediction accuracy of the recoverable amount of the first asset. In this way, an accurate predictive estimate of the asset recoverable amount can be made.
According to one aspect of the present application, there is provided an asset derating prediction method for business decisions, comprising:
a training phase comprising:
acquiring training data, wherein the training data comprises historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable money of the first asset;
dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data;
obtaining a plurality of monthly history feature vectors by passing each monthly history data in the plurality of monthly history data through a multi-scale neighborhood feature extraction module;
passing the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors;
obtaining week dynamic historical feature vectors from the plurality of week historical data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model;
calculating an incidence matrix of the monthly dynamic historical characteristic vector and the weekly dynamic historical characteristic vector to obtain a decoding characteristic matrix;
carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix;
passing the corrected decoding feature matrix through a decoder to obtain a decoding loss function value;
training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder by taking the decoding loss function value as a loss function value; and an inference phase comprising:
obtaining inferred data comprising historical trading price data for the first asset in the second-hand trading market;
inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and determining an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
In the asset derating prediction method for business decision, the passing each monthly history data in the monthly history data through a multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors includes: inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
In the asset derating prediction method for business decision, the inputting the monthly history data into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale monthly history feature vector includes: performing one-dimensional convolution coding on the monthly history data by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first scale monthly history feature vector; wherein the formula is:
wherein,ais a first convolution kernelxA width in the direction,
For the first convolution a kernel parameter vector £ is>
Is a matrix of local vectors operating with a convolution kernel,wis the size of the first convolution kernel,Xrepresenting the monthly history data.
In the asset derating prediction method for business decision, the inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector includes: performing one-dimensional convolution coding on the monthly history data by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale monthly history feature vector; wherein the formula is:
wherein b is a second convolution kernelxA width in the direction,
Is the second convolution kernel parameter vector->
Is a matrix of local vectors operating with a convolution kernel, m is the size of the second convolution kernel,Xrepresenting the monthly history data.
In the asset derating prediction method for business decision, the calculating an association matrix of the monthly dynamic history feature vector and the weekly dynamic history feature vector to obtain a decoding feature matrix includes: calculating an incidence matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
wherein
A transposed vector representing the historical monthly motion feature vector, based on the comparison of the transposed vector and the predicted value>
A characteristic vector representing the week dynamics>
Represents the decoded feature matrix, <' > is selected>
Representing a matrix multiplication.
In the asset derating prediction method for business decision, the performing smooth maximum function approximation modulation on the decoding feature matrix to obtain a corrected decoding feature matrix includes: carrying out smooth maximum function approximation modulation on the decoding characteristic matrix according to the following formula to obtain the corrected decoding characteristic matrix; wherein the formula is:
wherein
Represents the decoding feature matrix, based on the decoding feature matrix>
Is the ^ th or greater than the decoding feature matrix>
The value of the characteristic of the location is,
is a two-norm of a vector, and->
Means that each value of the matrix is multiplied by a predetermined value, is/are>
Indicating a position-wise addition, <' > or>
Representing the corrected decoded feature matrix.
In the asset derating prediction method for business decision, the passing the corrected decoded feature matrix through a decoder to obtain a decoding loss function value includes: performing decoding regression on the corrected decoding feature matrix by using the decoder to obtain a training decoding value according to the following formula:
in which>
Is the corrected decoded feature matrix, is->
Is the training decode value, is>
Is rightReset matrix, based on>
Represents a matrix multiplication; and calculating a variance between the training decoded value and a true value of the reclaimable amount of the first asset in the training data as the decoding loss function value.
According to another aspect of the present application, there is provided an asset derating prediction system for business decisions, comprising:
a training module comprising:
the training data acquisition unit is used for acquiring training data, wherein the training data comprises historical trading price data of the first asset in the second-hand trading market and a real value of the recoverable money of the first asset;
the dividing unit is used for dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data;
the monthly multi-scale coding unit is used for enabling each monthly history data in the monthly history data to pass through a multi-scale neighborhood feature extraction module so as to obtain a plurality of monthly history feature vectors;
the month understanding unit is used for enabling the plurality of monthly history feature vectors to pass through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors;
the system comprises a multi-scale week coding and understanding unit, a data processing unit and a data processing unit, wherein the multi-scale neighborhood feature extraction module is used for extracting a bidirectional long-short term memory neural network model from historical data of a plurality of weeks;
the association unit is used for calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector to obtain a decoding characteristic matrix;
the correction unit is used for carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix;
a decoding loss unit, configured to pass the corrected decoding feature matrix through a decoder to obtain a decoding loss function value;
the training unit is used for training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder by taking the decoding loss function value as a loss function value; and an inference module comprising:
an inferred data acquisition unit for acquiring inferred data comprising historical trading price data for the first asset in the second-hand trading market;
a data processing unit for inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and an asset derating loss determination unit to determine an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
In the asset derating prediction system for business decision, the monthly multi-scale coding unit is further configured to: inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
In the asset derating prediction system for business decision, the associating unit is further configured to: calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
wherein
A transposed vector representing the historical monthly motion feature vector, based on the comparison of the transposed vector and the predicted value>
A characteristic vector representing the week's dynamic history>
Represents the decoded feature matrix, <' > is selected>
Representing a matrix multiplication.
In the asset derating prediction system for business decision, the correction unit is further configured to: carrying out smooth maximum function approximation modulation on the decoding characteristic matrix according to the following formula to obtain the corrected decoding characteristic matrix; wherein the formula is:
wherein
Represents the decoded feature matrix, <' > is selected>
Is the ^ th or greater than the decoding feature matrix>
The value of the characteristic of the location is,
is a two-norm of a vector, and->
Means that each value of the matrix is multiplied by a predetermined value, is/are>
Representing a position-wise addition>
Representing the corrected decoded feature matrix.
In the asset derating prediction system for business decision, the decoding loss unit is further configured to: performing decoding regression on the corrected decoding feature matrix by using the decoder to obtain a training decoding value according to the following formula:
wherein->
Is the corrected decoded feature matrix, is->
Is the training decode value, is>
Is a weight matrix, is->
Represents a matrix multiplication; and calculating a variance between the training decoded value and a true value of the reclaimable amount of the first asset in the training data as the decoding loss function value. />
According to still another aspect of the present application, there is provided an electronic apparatus including: a processor; and a memory having stored therein computer program instructions which, when executed by the processor, cause the processor to perform a method of asset derate prediction for business decisions as described above.
According to yet another aspect of the present application, there is provided a computer readable medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform a method of asset derate prediction for business decisions as described above.
Compared with the prior art, the asset deduction prediction method and system for business decision adopt an artificial intelligence technology based on deep learning to perform prediction through an encoder and a decoder, namely, the encoder is used for extracting multi-scale implicit association features of historical trading prices of a first asset in a second-hand trading market under different time spans, and the decoder is used for performing decoding regression to improve the prediction accuracy of the recoverable amount of the first asset. In this way, an accurate predictive estimate of the asset recoverable amount can be made.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
FIG. 1A is a flow diagram of a training phase in an asset derating prediction method for business decisions according to an embodiment of the present application.
FIG. 1B is a flow diagram of an inference phase in a method for asset derate prediction for business decisions, according to an embodiment of the present application.
FIG. 2A is an architecture diagram of a training phase in an asset derating prediction method for business decisions according to an embodiment of the present application.
FIG. 2B is an architecture diagram of an inference stage in a method for asset valuation forecasting for business decisions according to an embodiment of the application.
Fig. 3 is a flowchart of obtaining a plurality of monthly history feature vectors by passing each of the plurality of monthly history data through a multi-scale neighborhood feature extraction module in the asset derating prediction method for business decision according to the embodiment of the present application.
FIG. 4A is a block diagram of a training module in an asset valuation prediction system for business decisions according to an embodiment of the application.
FIG. 4B is a block diagram of an inference module in an asset valuation prediction system for business decisions according to an embodiment of the application.
Fig. 5 is a block diagram of an electronic device according to an embodiment of the application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Summary of the application
As noted above in the background, asset reductions belong to the categories of profit and loss in accounting. The asset deduction value is the corresponding loss confirmed by the preparation for the asset deduction loss counted by judging that the recoverable amount of the asset is lower than the account value of the asset through the testing of the asset on the balance table day of the asset, namely, the asset deduction loss = the account value of the asset-the recoverable amount of the asset.
Of the most important in making asset derating forecasts is the forecasting and assessment of the amount of recoverable assets. Current practice is time-based wear methods, but this practice ignores the value of the property of the asset's market, resulting in large deviations in predictions and assessments. Therefore, an optimized asset derating prediction scheme for business decisions is desired.
At present, deep learning and neural networks have been widely applied in the fields of computer vision, natural language processing, speech signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation and the like.
In recent years, deep learning and the development of neural networks provide new solutions and schemes for asset derating prediction of business decisions.
Accordingly, since the existing forecast and appraisal schemes for the recoverable amount of the property are mostly time-based wear methods, such schemes ignore the market attribute values of the property, resulting in a large deviation in making the forecast and appraisal. Based on this, in the technical scheme of the application, an artificial intelligence technology based on deep learning is adopted, prediction is carried out through an encoder and a decoder, that is, multi-scale implicit association features of historical trading prices of the first asset in the second-hand trading market under different time spans are extracted based on the encoder, and decoding regression is carried out based on the decoder so as to improve the accuracy of prediction of the recoverable amount of the first asset. Therefore, the recoverable amount of the assets can be accurately predicted and estimated, and the asset value reduction loss can be accurately predicted, so that the situation that the value of the assets is reevaluated and the profit is manipulated is avoided, and the artificial fluctuation of the profit is limited.
Specifically, in the technical solution of the present application, first, training data is obtained, where the training data includes historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable amount of the first asset. Next, it is considered that in the historical deal price data of the first asset in the second-hand trading market, there will be a plurality of monthly history data and a plurality of weeks history data, which differ in the variation characteristics in the period of months and the variation characteristics in the period of weeks, and which also have different dynamic variation characteristics at different time spans in the respective months and at different time spans in the respective weeks. Therefore, in order to sufficiently extract the historical dynamic change characteristics of the historical trading price data of the first asset in the second-hand trading market, the historical trading price data of the first asset in the second-hand trading market is divided by month and week to mine the dynamic change characteristics of the first asset in each month and the dynamic change characteristics of the first asset in each week, and dynamic feature distribution information of the first asset in a month period and dynamic feature distribution information of the first asset in a week period are further mined. That is, specifically, first, the historical deal price data is divided by month and week to obtain a plurality of monthly history data and a plurality of weekly history data, so as to be subjected to dynamic change feature extraction later.
Then, for a plurality of monthly history data of the first asset in the historical trading price data of the second-hand trading market, considering that the first asset has different mode characteristics under different time period spans in each month, encoding each monthly history data of the plurality of monthly history data through a multi-scale neighborhood feature extraction module to extract dynamic multi-scale neighborhood associated features of each monthly history data of the plurality of monthly history data under different time spans, so as to obtain a plurality of monthly history feature vectors, and thus represent the multi-scale dynamic change features of the historical data of each month.
Further, considering that for the multi-scale dynamically changing features of the historical data of the months under different time spans, the relevance extraction of the months is needed to be carried out on the multi-scale dynamically changing features of the historical data of the months so as to mine the changing features of the historical data under the period of the months. Therefore, in the technical solution of the present application, the plurality of monthly history feature vectors are passed through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors. It should be understood that the bidirectional Long-Short Term Memory neural network model (LSTM) enables the weights of the neural network to be updated by adding an input gate, an output gate and a forgetting gate, and the weight scales of different channels can be dynamically changed under the condition that the parameters of the network model are fixed, so that the problem of gradient disappearance or gradient expansion can be avoided. Particularly, the bidirectional long and short term memory neural network model is formed by combining a forward LSTM and a backward LSTM, the forward LSTM can learn the associated feature information of the dynamic variation feature of the historical data of each month in a local area before a current month, and the backward LSTM can learn the associated feature information of the dynamic variation feature of the historical data of each month in a local area after the current month, so that the month dynamic historical feature vector obtained by the bidirectional long and short term memory neural network model learns the implicit associated feature information of medium and short distance dependence of the dynamic variation feature of the historical data of each month in the local area.
Then, for a plurality of weeks of historical data of the first asset in the historical trading price data of the second-hand trading market, in order to extract dynamic change features of the historical data in a week period, similarly, the plurality of weeks of historical data are encoded through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model to extract the change features of the historical data in the week period, so that a week dynamic historical feature vector is obtained.
Then, the correlation matrix of the month dynamic history feature vector and the week dynamic history feature vector is further calculated to represent correlation feature distribution information between the dynamic change features of the history data in the month period and the dynamic change features of the history data in the week period, that is, the overall change feature distribution representation of the history data, and this is used as a decoding feature matrix to perform decoding processing in a decoder, thereby obtaining a decoding value representing the recoverable amount of the first asset. An asset derating loss value is then determined based on the decoded value and an asset accounting value of the first asset. That is, specifically, the asset worth-reducing loss value is determined according to asset worth-reducing loss = asset accounting value of asset-recoverable amount of asset. Thus, the asset recoverable amount can be accurately predicted and estimated, and the asset loss can be accurately predicted.
Particularly, in the technical solution of the present application, here, through a multi-scale neighborhood feature extraction module and a bidirectional long and short term memory neural network model, time sequence associated features of the month historical data and the week historical data at different time granularities can be obtained respectively, so that the association matrix of the month dynamic historical feature vector and the week dynamic historical feature vector is further calculated to obtain the decoding feature matrix, and the decoding feature matrix may include the cross-cycle price associated expression information of the historical deal price data at a rich time scale, thereby improving the accuracy of the decoding value.
However, at the same time, since the decoding feature matrix includes cross-cycle price correlation features at different time granularities, monotonicity of global-local correlation feature semantic expression in the decoding feature matrix is affected, and thus training difficulty of the decoding feature matrix for obtaining the decoded value through a decoder is increased.
Therefore, preferably during training, the decoded feature matrix is subjected to smooth maximum function approximation modulation, which is expressed as:
Here, by using the edge feature matrix
Approximately defines a signed distance function by a smooth maximum function of the row and column dimensions, the decoded feature matrix->
A relatively good union of convex optimizations of high-dimensional manifolds characterized in a high-dimensional feature space and by means of which the decoding feature matrix &>
By modulating the structured feature distribution of (a), the variation of the spatial features from the intrinsic structure of the feature distribution into the feature space can be obtainedNaturally distributed branch enhancing the decoding feature matrix->
The high-dimensional manifold feature expression of (a) is preserved with a convex monotonicity, thereby enhancing the decoded feature matrix->
The decoder obtains the convergence effect of the decoded value, the training difficulty of the decoder is reduced, and the decoding accuracy is further improved. Therefore, the recoverable amount of the assets can be accurately predicted and estimated, and the asset value reduction loss can be accurately predicted, so that the situation that the value of the assets is reevaluated and the profit is manipulated is avoided, and the artificial fluctuation of the profit is limited.
Based on this, the present application proposes an asset derating prediction method for business decisions, which includes: a training phase comprising: acquiring training data, wherein the training data comprises historical trading price data of the first asset in a second-hand trading market and a true value of the recoverable amount of the first asset; dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data; obtaining a plurality of monthly history feature vectors by passing each monthly history data in the plurality of monthly history data through a multi-scale neighborhood feature extraction module; passing the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors; obtaining week dynamic historical feature vectors from the plurality of week historical data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model; calculating an incidence matrix of the monthly dynamic historical characteristic vector and the weekly dynamic historical characteristic vector to obtain a decoding characteristic matrix; carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix; passing the corrected decoding feature matrix through a decoder to obtain a decoding loss function value; training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder by taking the decoding loss function value as a loss function value; and, an inference phase comprising: obtaining inferred data comprising historical trading price data for the first asset in the second-hand trading market; inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and determining an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
Having described the general principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.
Exemplary method
FIG. 1A is a flow diagram of a training phase of a method for asset derate prediction for business decisions according to an embodiment of the present application. As shown in fig. 1A, an asset derating prediction method for business decision according to an embodiment of the present application includes: a training phase comprising the steps of: s110, acquiring training data, wherein the training data comprises historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable money of the first asset; s120, dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data; s130, enabling each monthly history data in the plurality of monthly history data to pass through a multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors; s140, passing the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors; s150, acquiring week dynamic historical feature vectors from the week historical data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model; s160, calculating the association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector to obtain a decoding characteristic matrix; s170, carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix; s180, enabling the corrected decoding characteristic matrix to pass through a decoder to obtain a decoding loss function value; and S190, taking the decoding loss function value as a loss function value, and training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder.
FIG. 1B is a flow diagram of an inference phase in a method for asset derate prediction for business decisions, according to an embodiment of the present application. As shown in fig. 1B, an asset derating prediction method for business decision according to an embodiment of the present application includes: an inference phase comprising the steps of: s210, obtaining inferred data, wherein the inferred data comprises historical trading price data of the first asset in the second-hand trading market; s220, inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder to obtain a decoded value representing the recoverable amount of the first asset; and S230, determining an asset derating loss value based on the decoded value and the asset account value of the first asset.
FIG. 2A is an architecture diagram of a training phase in an asset derating prediction method for business decisions according to an embodiment of the present application. As shown in fig. 2A, in the training phase, first, in the framework, training data is acquired, which includes historical trading price data of the first asset in the second-hand trading market and the true value of the recoverable amount of the first asset. Then, the historical trading price data is divided according to the months and the weeks to obtain a plurality of monthly history data and a plurality of weekly history data. Then, each monthly history data in the plurality of monthly history data passes through a multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors. And then, passing the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors, and simultaneously, passing the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model to obtain weekly dynamic history feature vectors from the plurality of weekly historical data. Then, calculating the association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector to obtain a decoding characteristic matrix. And then, carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix. Then, the corrected decoding feature matrix is passed through a decoder to obtain a decoding loss function value. And then, taking the decoding loss function value as a loss function value, and training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder.
FIG. 2B is an architecture diagram of an inference stage in an asset derating prediction method for business decisions according to an embodiment of the present application. As shown in fig. 2B, in the inference phase, in the architecture, first, inferred data is obtained that includes historical deal price data for the first asset in the second-hand trading market. Then, the inferred data is input to the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model, and the decoder to obtain a decoded value representing a recoverable amount of the first asset. An asset derating loss value is then determined based on the decoded value and an asset accounting value of the first asset.
In the training phase, in step S110, training data is obtained, the training data including historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable amount of the first asset. As noted above in the background, asset valuation belongs to the profit-and-loss category of subjects in accounting. The asset deduction value is the corresponding loss confirmed by the preparation for the asset deduction loss counted by judging that the recoverable amount of the asset is lower than the account value of the asset through the testing of the asset on the balance table day of the asset, namely, the asset deduction loss = the account value of the asset-the recoverable amount of the asset. Of the most important in making asset derating forecasts is the forecasting and assessment of the amount of recoverable assets. Current practice is time-based wear methods, but this practice ignores the value of the property of the asset's market, resulting in large deviations in predictions and assessments. Therefore, an optimized asset derating prediction scheme for business decisions is desired.
Accordingly, since the existing forecast and appraisal schemes for the recoverable amount of the property are mostly time-based wear methods, such schemes ignore the market attribute values of the property, resulting in a large deviation in making the forecast and appraisal. Based on this, in the technical scheme of the application, an artificial intelligence technology based on deep learning is adopted, prediction is performed through an encoder and a decoder, that is, multi-scale implicit association features of historical trading prices of the first asset in the second-hand trading market under different time spans are extracted based on the encoder, and decoding regression is performed based on the decoder to improve the accuracy of prediction of the recoverable amount of the first asset. Therefore, the asset recoverable amount can be accurately predicted and estimated, and then the asset worth reduction loss can be accurately predicted, so that the asset reestimation value and the operation profit are prevented from being confirmed, and the artificial fluctuation of the profit is limited. Therefore, specifically, in the technical solution of the present application, first, training data is obtained, the training data including historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable amount of the first asset. Here, the training data is existing data.
In the training phase, in step S120, the historical deal price data is divided by month and week to obtain a plurality of monthly history data and a plurality of weekly history data. It is considered that in the historical deal price data of the first asset in the second-hand trading market, there may be a plurality of monthly history data and a plurality of weeks history data, which differ in the variation characteristics of the period of the month and the variation characteristics of the period of the week, and which also have different dynamic variation characteristics at different time spans in the respective months and at different time spans in the respective weeks. Therefore, in order to sufficiently extract the historical dynamic change characteristics of the historical trading price data of the first asset in the second-hand trading market, the historical trading price data of the first asset in the second-hand trading market is divided by month and week to mine the dynamic change characteristics of the first asset in each month and the dynamic change characteristics of the first asset in each week, and dynamic feature distribution information of the first asset in a month period and dynamic feature distribution information of the first asset in a week period are further mined. That is, specifically, first, the historical deal price data is divided by month and week to obtain a plurality of monthly history data and a plurality of weekly history data, so as to be subjected to dynamic change feature extraction later.
In the training phase, in step S130, each of the plurality of monthly history data is passed through a multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors. Considering that the first asset has different mode characteristics under different time period spans in each month in the historical trading price data of the second-hand trading market, encoding each month history data in the plurality of month history data through a multi-scale neighborhood characteristic extraction module to extract dynamic multi-scale neighborhood correlation characteristics of each month history data in the plurality of month history data under different time spans so as to obtain a plurality of month history characteristic vectors, and accordingly, the multi-scale dynamic change characteristics of the history data of each month are represented.
Fig. 3 is a flowchart of obtaining a plurality of monthly history feature vectors by passing each of the plurality of monthly history data through a multi-scale neighborhood feature extraction module in the asset derating prediction method for business decision according to the embodiment of the present application. As shown in fig. 3, the passing each of the plurality of monthly history data through the multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors includes: s310, inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; s320, inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and S330, cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
Specifically, in this embodiment of the present application, the inputting the monthly history data into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale monthly history feature vector includes: performing one-dimensional convolution coding on the monthly history data by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first scale monthly history feature vector; wherein the formula is:
wherein,ais a first convolution kernelxWidth in the direction,
For the first convolution a kernel parameter vector £ is>
Is a matrix of local vectors operating with a convolution kernel,wis the size of the first convolution kernel,Xrepresenting the monthly history data.
Specifically, in this embodiment of the present application, the inputting the monthly history data into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector includes: performing one-dimensional convolution coding on the monthly history data by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale monthly history feature vector; wherein the formula is:
wherein b is a second convolution kernelxWidth in the direction,
Is the second convolution kernel parameter vector->
As local direction of operation with convolution kernelA quantity matrix, m being the size of the second convolution kernel,Xrepresenting the monthly history data.
In the training phase, in step S140, the plurality of monthly history feature vectors are passed through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors. Considering that for the multi-scale dynamic change features of the historical data of the months under different time spans, correlation extraction of the months is needed to excavate the change features of the historical data under the period of the months. Therefore, in the technical scheme of the application, the plurality of monthly history feature vectors are processed by a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors.
It should be understood that the bidirectional Long-Short Term Memory neural network model (LSTM) enables the weights of the neural network to be updated by adding an input gate, an output gate and a forgetting gate, and the weight scales of different channels can be dynamically changed under the condition that the parameters of the network model are fixed, so that the problem of gradient disappearance or gradient expansion can be avoided. Particularly, the bidirectional long and short term memory neural network model is formed by combining a forward LSTM and a backward LSTM, the forward LSTM can learn the associated feature information of the dynamic change features of the historical data of each month in a local area before the current month, and the backward LSTM can learn the associated feature information of the dynamic change features of the historical data of each month in a subsequent local area after the current month, so that the monthly dynamic historical feature vectors obtained by the bidirectional long and short term memory neural network model learn the implicit associated feature information of medium and short distance dependence of the dynamic change features of the historical data of each month in the local area.
In the training phase, in step S150, a week dynamic history feature vector is obtained from the plurality of week history data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model. For the plurality of weeks of historical data of the first asset in the historical trading price data of the second-hand trading market, in order to extract the dynamic change features of the historical data in the week period, similarly, the plurality of weeks of historical data are encoded through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model to extract the change features of the historical data in the week period, so that week dynamic historical feature vectors are obtained.
In the training phase, in step S160, the correlation matrix of the monthly dynamics history feature vector and the weekly dynamics history feature vector is calculated to obtain a decoding feature matrix. That is, the correlation matrix of the month dynamic history feature vector and the week dynamic history feature vector is calculated to represent correlation feature distribution information between the dynamic change feature of the history data in the month period and the dynamic change feature of the history data in the week period, that is, the overall change feature distribution representation of the history data, and this is used as the decoding feature matrix.
Specifically, in this embodiment of the present application, the calculating an association matrix of the monthly dynamic history feature vector and the weekly dynamic history feature vector to obtain a decoding feature matrix includes: calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
wherein
A transposed vector representing the monthly dynamic history feature vector, based upon the predicted historical data, and based upon the predicted historical data>
A characteristic vector representing the week's dynamic history>
Represents the decoded feature matrix, <' > is selected>
Representing a matrix multiplication.
In the training phase, in step S170, the decoding feature matrix is subjected to smooth maximum function approximation modulation to obtain a corrected decoding feature matrix. Particularly, in the technical solution of the present application, here, through a multi-scale neighborhood feature extraction module and a bidirectional long and short term memory neural network model, time sequence associated features of the month historical data and the week historical data at different time granularities can be obtained respectively, so that the association matrix of the month dynamic historical feature vector and the week dynamic historical feature vector is further calculated to obtain the decoding feature matrix, and the decoding feature matrix may include the cross-cycle price associated expression information of the historical deal price data at a rich time scale, thereby improving the accuracy of the decoding value.
However, at the same time, since the decoding feature matrix includes cross-cycle price correlation features at different time granularities, monotonicity of global-local correlation feature semantic expression in the decoding feature matrix is affected, and thus training difficulty of the decoding feature matrix for obtaining the decoded value through a decoder is increased. Therefore, preferably, in the training process, the decoding feature matrix is subjected to smooth maximum function approximation modulation.
Specifically, in this embodiment of the present application, the performing smooth maximum function approximation modulation on the decoded feature matrix to obtain a corrected decoded feature matrix includes: carrying out smooth maximum function approximation modulation on the decoding characteristic matrix according to the following formula to obtain the corrected decoding characteristic matrix; wherein the formula is:
wherein
Represents the decoded feature matrix, <' > is selected>
Is the ^ th or greater than the decoding feature matrix>
The value of the characteristic of the location is,
is a two-norm of a vector, and->
Means that each value of the matrix is multiplied by a predetermined value, is/are>
Representing a position-wise addition>
Representing the corrected decoded feature matrix.
Here, by using the edge feature matrix
Approximately defines a signed distance function by a smooth maximum function of the row and column dimensions, the decoded feature matrix->
A relatively good union of the convex optimizations of the high-dimensional manifold characterized in the high-dimensional feature space and by which the decoded feature matrix ≦ or ≦ the decoded feature matrix>
Is modulated, a natural distribution transfer of the intrinsic structure of the feature distribution into a spatial feature variation in the feature space can be achieved, which enhances the decoded feature matrix->
The high-dimensional manifold feature expression of (a) is preserved with a convex monotonicity, thereby enhancing the decoded feature matrix->
The decoder obtains the convergence effect of the decoded value, the training difficulty of the decoder is reduced, and the decoding accuracy is further improved. />
In the training phase, in step S180, the corrected decoded feature matrix is passed through a decoder to obtain the decoding loss function value. Specifically, in this embodiment of the present application, the passing the corrected decoded feature matrix through a decoder to obtain a decoding loss function value includes: performing decoding regression on the corrected decoding feature matrix by using the decoder to obtain a training decoding value according to the following formula:
in which>
Is the corrected decoded feature matrix, is->
Is the training decode value, is>
Is a weight matrix, is->
Represents a matrix multiplication; and calculating a variance between the training decoded value and a true value of the reclaimable amount of the first asset in the training data as the decoding loss function value.
In the training phase, in step S190, the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder are trained with the decoding loss function value as a loss function value.
After training is completed, the inference phase is entered. That is, the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model, and the decoder trained in the training stage can be obtained according to the method described above, and then the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model, and the decoder trained in the training stage are used in actual inference.
Specifically, in the inference phase, first, inferred data is obtained that includes historical trading price data for the first asset in the second-hand trading market. Then, the inferred data is input to the trained multi-scale neighborhood feature extraction module, the two-way long-short term memory neural network model, and the decoder to derive a decoded value representing a reclaimable amount of the first asset. An asset derating loss value is then determined based on the decoded value and an asset accounting value of the first asset. That is, specifically, the asset worth-reducing loss value is determined according to asset worth-reducing loss = asset accounting value of asset-recoverable amount of asset. Therefore, the recoverable amount of the assets can be accurately predicted and estimated, the asset value reduction loss can be accurately estimated, the situation that the asset value re-estimation is increased and the operation profit is controlled is avoided, and the artificial fluctuation of the profit is limited.
In summary, an asset deduction prediction method for business decision based on the embodiment of the present application is illustrated, which adopts an artificial intelligence technology based on deep learning to perform prediction through an encoder and a decoder, that is, a multi-scale implicit association feature of a historical trading price of a first asset in a second-hand trading market at different time spans is extracted based on the encoder, and decoding regression is performed based on the decoder to improve the accuracy of prediction of the recoverable amount of the first asset. In this way, an accurate predictive estimate of the asset recoverable amount can be made.
Exemplary System
FIG. 4A is a block diagram of a training module in an asset derating prediction system for business decisions according to an embodiment of the present application. As shown in FIG. 4A, an asset derating prediction system 100 for business decisions according to embodiments of the present application includes a training module 110; wherein the training module 110 includes: a training data obtaining unit 111 configured to obtain training data including historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable amount of the first asset; a dividing unit 112, configured to divide the historical transaction price data according to months and weeks to obtain multiple monthly calendar history data and multiple week history data; the month multi-scale coding unit 113 is configured to pass each month history data in the plurality of month history data through a multi-scale neighborhood feature extraction module to obtain a plurality of month history feature vectors; a month understanding unit 114, configured to pass the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors; a week multi-scale encoding and understanding unit 115, configured to obtain a week dynamic history feature vector from the week history data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model; an association unit 116, configured to calculate an association matrix of the monthly dynamic history feature vector and the weekly dynamic history feature vector to obtain a decoding feature matrix; a correcting unit 117, configured to perform smooth maximum function approximation modulation on the decoded feature matrix to obtain a corrected decoded feature matrix; a decoding loss unit 118, configured to pass the corrected decoding feature matrix through a decoder to obtain a decoding loss function value; and a training unit 119, configured to train the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model, and the decoder with the decoding loss function value as a loss function value.
FIG. 4B is a block diagram of an inference module in an asset derating prediction system for business decisions according to an embodiment of the present application. As shown in FIG. 4B, the asset derating prediction system 100 for business decisions according to embodiments of the present application includes an inference module 120; wherein the inference module 120 comprises: an inferred data acquisition unit 121 for acquiring inferred data including historical trading price data of the first asset in the second-hand trading market; a data processing unit 122 for inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and an asset derating loss determination unit 123 for determining an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
In an example, in the asset derating prediction system for business decision 100, the monthly multi-scale encoding unit 113 is further configured to: inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length; inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
In an example, in the asset derating prediction system 100 for business decision described above, the associating unit 116 is further configured to: calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix; wherein the formula is:
wherein
A transposed vector representing the historical monthly motion feature vector, based on the comparison of the transposed vector and the predicted value>
A characteristic vector representing the week's dynamic history>
Representing the decoded bitsSign matrix, ->
Representing a matrix multiplication.
In an example, in the asset derating prediction system 100 for business decisions described above, the correcting unit 117 is further configured to: carrying out smooth maximum function approximation modulation on the decoding characteristic matrix according to the following formula to obtain the corrected decoding characteristic matrix; wherein the formula is:
wherein
Represents the decoded feature matrix, <' > is selected>
Is the ^ th or greater than the decoding feature matrix>
The value of the characteristic of the location is,
is a two-norm of a vector, and->
Means that each value of the matrix is multiplied by a predetermined value, is/are>
Indicating a position-wise addition, <' > or>
Representing the corrected decoded feature matrix.
In one example, in the asset derating prediction system 100 for business decisions described above, the decoding loss unit 118 is further configured to: performing decoding regression on the corrected decoding feature matrix by using the decoder to obtain a training decoding value according to the following formula:
wherein->
Is the corrected decoded feature matrix, is->
Is said training decode value, is>
Is a weight matrix, is->
Represents a matrix multiplication; and calculating a variance between the training decoded value and a true value of the reclaimable amount of the first asset in the training data as the decoding loss function value.
Here, it will be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the asset deduction prediction system for business decision 100 described above have been described in detail in the description of the asset deduction prediction method for business decision with reference to fig. 1 to 3, and thus, a repetitive description thereof will be omitted.
As described above, the asset derating prediction system 100 for business decisions according to the embodiment of the present application may be implemented in various terminal devices, such as a server for asset derating prediction for business decisions and the like. In one example, the asset derating prediction system 100 for business decisions according to embodiments of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the asset mitigation prediction system 100 for business decisions may be a software module in the operating system of the terminal device or may be an application developed for the terminal device; of course, the asset derating prediction system 100 for business decisions may also be one of many hardware modules of the terminal device.
Alternatively, in another example, the asset mitigation prediction system for business decision 100 and the terminal device may be separate devices, and the asset mitigation prediction system for business decision 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interaction information in an agreed data format.
Exemplary electronic device
Next, an electronic apparatus according to an embodiment of the present application is described with reference to fig. 5. FIG. 5 illustrates a block diagram of an electronic device in accordance with an embodiment of the present application. As shown in fig. 5, the electronic device 10 includes one or more processors 11 and memory 12.
The processor 11 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.
In one example, the electronic device 10 may further include: an input device 13 and an output device 14, which are interconnected by a bus system and/or other form of connection mechanism (not shown).
The input device 13 may include, for example, a keyboard, a mouse, and the like.
The output device 14 can output various information including a decoded value and the like to the outside. The output devices 14 may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.
Of course, for simplicity, only some of the components of the electronic device 10 relevant to the present application are shown in fig. 5, and components such as buses, input/output interfaces, and the like are omitted. In addition, the electronic device 10 may include any other suitable components depending on the particular application.
Exemplary computer program product and computer-readable storage Medium
In addition to the above-described methods and apparatus, embodiments of the present application may also be a computer program product comprising computer program instructions that, when executed by a processor, cause the processor to perform the steps in the functions of the asset mitigation prediction method for business decisions according to the various embodiments of the present application described in the "exemplary methods" section of this specification above.
The computer program product may be written with program code for performing the operations of embodiments of the present application in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.
Furthermore, embodiments of the present application may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform steps in functions in a method for asset mitigation prediction for business decisions according to various embodiments of the present application described in the "exemplary methods" section above in this specification.
The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The basic principles of the present application have been described above with reference to specific embodiments, but it should be noted that advantages, effects, etc. mentioned in the present application are only examples and are not limiting, and the advantages, effects, etc. must not be considered to be possessed by various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. As used herein, the words "or" and "refer to, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.
Claims (10)
1. An asset derating prediction method for business decisions, comprising:
a training phase comprising:
acquiring training data, wherein the training data comprises historical trading price data of the first asset in the second-hand trading market and a true value of the recoverable money of the first asset;
dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data;
obtaining a plurality of monthly history feature vectors by passing each monthly history data in the plurality of monthly history data through a multi-scale neighborhood feature extraction module;
passing the plurality of monthly history feature vectors through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors;
obtaining week dynamic historical feature vectors from the plurality of week historical data through the multi-scale neighborhood feature extraction module and the bidirectional long-short term memory neural network model;
calculating an incidence matrix of the monthly dynamic historical characteristic vector and the weekly dynamic historical characteristic vector to obtain a decoding characteristic matrix;
carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix;
passing the corrected decoding feature matrix through a decoder to obtain a decoding loss function value;
training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder by taking the decoding loss function value as a loss function value; and an inference phase comprising:
obtaining inferred data comprising historical trading price data for the first asset in the second-hand trading market;
inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and determining an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
2. The asset valuation prediction method for business decisions as claimed in claim 1, wherein said passing each of said plurality of monthly history data through a multi-scale neighborhood feature extraction module to obtain a plurality of monthly history feature vectors comprises:
inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length;
inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
3. The asset derating prediction method for business decisions as claimed in claim 2, wherein the inputting the monthly history data into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector comprises:
performing one-dimensional convolution coding on the monthly history data by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first scale monthly history feature vector;
wherein the formula is:
wherein,ais a first convolution kernelxA width in the direction,
Is a first convolution kernel parameter vector,
Is a matrix of local vectors operating with a convolution kernel,wis the size of the first convolution kernel,Xrepresenting the monthly history data.
4. The asset derating prediction method for business decisions as claimed in claim 3, wherein the inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale monthly history feature vector comprises:
performing one-dimensional convolution coding on the monthly history data by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale monthly history feature vector;
wherein the formula is:
5. The asset derating prediction method for business decisions as claimed in claim 4, wherein the computing the correlation matrix of the monthly dynamic historical eigenvectors and the weekly dynamic historical eigenvectors to obtain a decoded eigenvector matrix comprises:
calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix;
wherein the formula is:
6. The asset derating prediction method for business decisions as claimed in claim 5, wherein said smooth maximum function approximation modulation of the decoded feature matrix to obtain a corrected decoded feature matrix comprises:
carrying out smooth maximum function approximation modulation on the decoding characteristic matrix according to the following formula to obtain the corrected decoding characteristic matrix;
wherein the formula is:
7. The asset valuation prediction method for business decisions of claim 6 wherein said passing said corrected decoded features matrix through a decoder to obtain a decoding loss function value comprises:
performing decoding regression on the corrected decoding feature matrix by using the decoder to obtain a training decoding value according to the following formula:
wherein
Is the corrected decoded feature matrix and,
is the value of the training decoded value or values,
is a matrix of the weights that is,
represents a matrix multiplication; and calculating a variance between the training decoded value and a true value of the reclaimable amount of the first asset in the training data as the decoding loss function value.
8. An asset devaluation prediction system for business decisions, comprising:
a training module comprising:
the training data acquisition unit is used for acquiring training data, wherein the training data comprises historical trading price data of the first asset in the second-hand trading market and a real value of the recoverable money of the first asset;
the dividing unit is used for dividing the historical transaction price data according to months and weeks to obtain a plurality of monthly history data and a plurality of weekly history data;
the monthly multi-scale coding unit is used for enabling each monthly history data in the monthly history data to pass through a multi-scale neighborhood feature extraction module so as to obtain a plurality of monthly history feature vectors;
the month understanding unit is used for enabling the plurality of monthly history feature vectors to pass through a bidirectional long-short term memory neural network model to obtain monthly dynamic history feature vectors;
the system comprises a multi-scale week coding and understanding unit, a data processing unit and a data processing unit, wherein the multi-scale neighborhood feature extraction module is used for extracting a bidirectional long-short term memory neural network model from historical data of a plurality of weeks;
the association unit is used for calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector to obtain a decoding characteristic matrix;
the correction unit is used for carrying out smooth maximum function approximation modulation on the decoding characteristic matrix to obtain a corrected decoding characteristic matrix;
a decoding loss unit, configured to pass the corrected decoding feature matrix through a decoder to obtain a decoding loss function value;
the training unit is used for training the multi-scale neighborhood feature extraction module, the bidirectional long-short term memory neural network model and the decoder by taking the decoding loss function value as a loss function value; and an inference module comprising:
an inferred data acquisition unit for acquiring inferred data comprising historical trading price data for the first asset in the second-hand trading market;
a data processing unit for inputting the inferred data into the trained multi-scale neighborhood feature extraction module, the bi-directional long-short term memory neural network model and the decoder to obtain a decoded value representing a recoverable amount of the first asset; and an asset derating loss determination unit to determine an asset derating loss value based on the decoded value and an asset accounting value of the first asset.
9. The asset derating prediction system for business decisions as claimed in claim 8, wherein the monthly multi-scale coding unit is further configured to:
inputting the monthly history data into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale monthly history feature vector, wherein the first convolution layer has a first one-dimensional convolution kernel with a first length;
inputting the monthly history data into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale monthly history feature vector, wherein the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and
and cascading the first scale monthly history feature vector and the second scale monthly history feature vector to obtain the monthly history feature vector.
10. The asset derating prediction system for business decisions as claimed in claim 9, wherein the correlation unit is further configured to:
calculating an association matrix of the monthly dynamic history characteristic vector and the weekly dynamic history characteristic vector according to the following formula to obtain a decoding characteristic matrix;
wherein the formula is:
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310153438.2A CN115841346A (en) | 2023-02-23 | 2023-02-23 | Asset derating prediction method and system for business decisions |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310153438.2A CN115841346A (en) | 2023-02-23 | 2023-02-23 | Asset derating prediction method and system for business decisions |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115841346A true CN115841346A (en) | 2023-03-24 |
Family
ID=85580089
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310153438.2A Pending CN115841346A (en) | 2023-02-23 | 2023-02-23 | Asset derating prediction method and system for business decisions |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115841346A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116362503A (en) * | 2023-03-30 | 2023-06-30 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
| CN116681428A (en) * | 2023-08-03 | 2023-09-01 | 天津奇立软件技术有限公司 | Intelligent recycling management system and method for electronic equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738762A (en) * | 2020-06-19 | 2020-10-02 | 中国建设银行股份有限公司 | Method, device, equipment and storage medium for determining recovery price of poor assets |
| CN113506039A (en) * | 2021-08-03 | 2021-10-15 | 杭银消费金融股份有限公司 | Risk prediction method and device based on intelligent AI machine learning |
| CN113724017A (en) * | 2021-09-09 | 2021-11-30 | 平安银行股份有限公司 | Pricing method and device based on neural network, electronic equipment and storage medium |
| CN114091903A (en) * | 2021-11-22 | 2022-02-25 | 支付宝(杭州)信息技术有限公司 | Training method and device of loss assessment model, and loss assessment method and device |
| CN114169985A (en) * | 2021-12-10 | 2022-03-11 | 建信金融科技有限责任公司 | Asset value-reducing stage determination method and device |
| CN114519610A (en) * | 2022-02-16 | 2022-05-20 | 支付宝(杭州)信息技术有限公司 | Information prediction method and device |
| CN115293910A (en) * | 2022-08-04 | 2022-11-04 | 杭州衡泰技术股份有限公司 | Intelligent enterprise cash flow rating system based on financial big data |
| CN115511187A (en) * | 2022-09-29 | 2022-12-23 | 中国建设银行股份有限公司 | Asset recovery prediction method, device, equipment, medium and computer program product |
| US20230028266A1 (en) * | 2021-07-23 | 2023-01-26 | Dell Products L.P. | Product recommendation to promote asset recycling |
-
2023
- 2023-02-23 CN CN202310153438.2A patent/CN115841346A/en active Pending
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738762A (en) * | 2020-06-19 | 2020-10-02 | 中国建设银行股份有限公司 | Method, device, equipment and storage medium for determining recovery price of poor assets |
| US20230028266A1 (en) * | 2021-07-23 | 2023-01-26 | Dell Products L.P. | Product recommendation to promote asset recycling |
| CN113506039A (en) * | 2021-08-03 | 2021-10-15 | 杭银消费金融股份有限公司 | Risk prediction method and device based on intelligent AI machine learning |
| CN113724017A (en) * | 2021-09-09 | 2021-11-30 | 平安银行股份有限公司 | Pricing method and device based on neural network, electronic equipment and storage medium |
| CN114091903A (en) * | 2021-11-22 | 2022-02-25 | 支付宝(杭州)信息技术有限公司 | Training method and device of loss assessment model, and loss assessment method and device |
| CN114169985A (en) * | 2021-12-10 | 2022-03-11 | 建信金融科技有限责任公司 | Asset value-reducing stage determination method and device |
| CN114519610A (en) * | 2022-02-16 | 2022-05-20 | 支付宝(杭州)信息技术有限公司 | Information prediction method and device |
| CN115293910A (en) * | 2022-08-04 | 2022-11-04 | 杭州衡泰技术股份有限公司 | Intelligent enterprise cash flow rating system based on financial big data |
| CN115511187A (en) * | 2022-09-29 | 2022-12-23 | 中国建设银行股份有限公司 | Asset recovery prediction method, device, equipment, medium and computer program product |
Non-Patent Citations (3)
| Title |
|---|
| ZIHAO CHEN: "Asset Allocation Strategy with Monte-Carlo Simulation for Forecasting Stock Price by ARIMA Model" * |
| 汪怔江;张洪伟;雷彬;: "基于蚁群算法的神经网络在企业资信评估中的应用" * |
| 沈沛龙;崔婕;: "内部评级法中违约损失率的度量方法研究" * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116362503A (en) * | 2023-03-30 | 2023-06-30 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
| CN116362503B (en) * | 2023-03-30 | 2023-11-07 | 国网河南省电力公司安阳供电公司 | Electric power regulating method and system based on artificial intelligence |
| CN116681428A (en) * | 2023-08-03 | 2023-09-01 | 天津奇立软件技术有限公司 | Intelligent recycling management system and method for electronic equipment |
| CN116681428B (en) * | 2023-08-03 | 2023-09-29 | 天津奇立软件技术有限公司 | Intelligent recycling management system and method for electronic equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Tsantekidis et al. | Using deep learning for price prediction by exploiting stationary limit order book features | |
| Shah et al. | A comprehensive review on multiple hybrid deep learning approaches for stock prediction | |
| CN115841346A (en) | Asset derating prediction method and system for business decisions | |
| CN111242752A (en) | Method and system for determining recommended object based on multi-task prediction | |
| CN115082920B (en) | Deep learning model training method, image processing method and device | |
| CN111275514A (en) | Intelligent purchasing method and system, storage medium and electronic device | |
| CN112988840A (en) | Time series prediction method, device, equipment and storage medium | |
| US20230342797A1 (en) | Object processing method based on time and value factors | |
| CN111260189A (en) | Risk control method, risk control device, computer system and readable storage medium | |
| CN116739653A (en) | Sales data acquisition and analysis system and method thereof | |
| CN115271886A (en) | Financial product recommendation method and device, storage medium and electronic equipment | |
| Dael et al. | Stock Market Prediction Using Generative Adversarial Networks (GANs): Hybrid Intelligent Model. | |
| CN110033113A (en) | The learning method of information processing system and information processing system | |
| CN110213239B (en) | Suspicious transaction message generation method and device and server | |
| Xu et al. | Research on Multistage Dynamic Trading Model Based on Gray Model and Auto‐Regressive Integrated Moving Average Model | |
| CN113515920B (en) | Method, electronic device and computer readable medium for extracting formulas from tables | |
| CN115860889A (en) | Financial loan big data management method and system based on artificial intelligence | |
| CN113806622A (en) | Recommendation method, device and equipment | |
| Homma | Competition on the Cost Frontier and Intertemporal Regular Linkages: Theoretical Implications of the Efficient Structure and Quiet-Life Hypotheses | |
| Wood et al. | Addressing probationary period within a competing risks survival model for retail mortgage loss given default | |
| CN112862602B (en) | User request determining method, storage medium and electronic device | |
| CN111881747B (en) | Information prediction method and device and electronic equipment | |
| Abdool et al. | Islamic finance in Canada powered by big data: A case study | |
| CN113255891B (en) | Method, neural network model and device for processing event characteristics | |
| CN117540745A (en) | Named entity recognition model training method and system based on countermeasure training |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |