Financial information data intelligent filling and reporting method

Abstract

The invention relates to the technical field of data filling and reporting, in particular to an intelligent filling and reporting method for financial information data. By constructing a candidate time window and adopting multi-scale entropy to calculate complexity, uncertainty and complexity of LOF distribution in the candidate time window under different time scales are comprehensively captured, short-term noise is smoothed, and long-term trend and structural change are disclosed, so that data description indexes have higher robustness and representativeness; by combining the volatility of LOF values in different overlapped sub-time periods in a candidate time window, the evaluation function can balance the complexity of data description and the stability of a local anomaly detection result at the same time, and an optimal time window is automatically selected, so that when local reachable density and LOF calculation are subsequently carried out in the window, the accuracy of the local anomaly detection result is greatly improved. The local distribution characteristics of the data are reflected more truly, the accuracy of financial data anomaly detection is improved, and the subsequent filling efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data filling, and particularly to an intelligent filling method for financial information data. Background Art

[0002] The filling of financial data covers multi-dimensional indicators such as income, expenditure, budget execution, assets and liabilities. Its data is large in scale, updated frequently, and has strong timeliness. At the same time, there are many uncertainties and noises.

[0003] Therefore, for the accuracy of financial data filling, before filling, it is necessary to detect abnormal financial data, where:

[0004] Traditional detection methods often face challenges brought by data timeliness, periodicity and non-stationarity. Taking monthly, quarterly or annual financial statements as an example, the data has both short-term random fluctuations and long-term trends and structural changes;

[0005] Secondly, if a fixed time window and single-scale entropy value calculation (such as Shannon entropy or coding length) are used to describe the data distribution, it is often difficult to comprehensively capture the dynamic characteristics of financial data at different time scales, resulting in the lack of robustness and accuracy in abnormal judgment. Summary of the Invention

[0006] Aiming at the above-mentioned disadvantages of the prior art, the present invention provides an intelligent filling method for financial information data, which can effectively solve the problem that the accuracy of abnormal data detection of financial data cannot be improved in the prior art.

[0007] To achieve the above object, the present invention is realized through the following technical solutions:

[0008] The present invention provides an intelligent filling method for financial information data, including the following steps:

[0009] Obtain the financial data to be filled;

[0010] Perform abnormal detection on the financial data to be filled, including:

[0011] Convert the financial records in the financial data into nodes v in the graph i , and define the edge e i corresponding to the node v ij ;

[0012] Construct a matrix and obtain the low-dimensional embedding vector H of the node by using a graph convolutional network (L) ;

[0013] Judge the abnormal nodes of the financial data by calculating the local outlier factor, including:

[0014] Introduce the time window T base Define the k-neighborhood with time constraints;

[0015] Calculate the node v i and the neighbor node v j of the reachable distance, calculate the local reachability density based on the reachable distance, combine the reachable distance and the local reachability density to calculate the local outlier factor, and determine the outlier nodes;

[0016] Define the candidate time window, and define the evaluation function based on the data complexity and LOF error to determine the optimal time window to implement the calculation and update of the local outlier factor and determine the outlier nodes;

[0017] Based on the determined outlier nodes, correct the financial data and fill in the form based on the corrected financial data.

[0018] Furthermore, construct the matrix and use the graph convolutional network to obtain the low-dimensional embedding vector H of the nodes (L) The method is as follows:

[0019] The entire graph structure G=(V, E), where V represents the node set and E represents the edge set;

[0020] Define the adjacency matrix A, and its element A ij =e ij ;

[0021] Introduce the identity matrix I to define the self-connection matrix

[0022] Define the matrix as the degree matrix of its diagonal element

[0023]

[0024] n represents the total number of nodes in the graph;

[0025] Use the graph convolutional network to extract the low-dimensional embedding representation of each node under the global and local graph structures, including:

[0026] Integrate the feature vectors of all nodes into the matrix X;

[0027] The graph convolutional network updates the node representation through neighbor information aggregation. Among them, the update formula of a single-layer GCN is:

[0028]

[0029] where H (l) represents the node representation matrix of the l-th layer, and the initial H(0) = X, W (l) represents the weight matrix of the l-th layer, represents the inverse square root of, and σ(·) represents the activation function;

[0030] After iterative processing through multiple layers of GCN, the low-dimensional embedding vector H of the nodes is finally obtained (L) .

[0031] Furthermore, the calculation method of the reachable distance is as follows:

[0032] For each node v i , determine its distance from other nodes, and determine the distance to the k-th nearest neighbor, denoted as d k (v i );

[0033] Based on the time window T base , define the time-constrained k-neighborhood as:

[0034]

[0035] where N k (v i ) represents the set of k nearest neighbors of node v i , and t i and t j represent the timestamps corresponding to node v i and node v j respectively;

[0036] Define the reachable distance i between node v j and its neighbor node v

[0037]

[0038] where, h i and h j represent the feature representations corresponding to the nodes respectively, and ||h i - h j || 2 represents the Euclidean distance between node v i and v j , represents the distance from node v base to its k-th nearest neighbor within T j .

[0039] Furthermore, the local reachability density is calculated according to the following relationship:

[0040]

[0041] Furthermore, the local outlier factor is calculated according to the following relational expression:

[0042]

[0043] Furthermore, the evaluation function Q(T) is defined according to the following relational expression:

[0044] Q(T) = L model (T) + λL error (T)

[0045] where L model (T) represents the complexity of the data description under the candidate time window T i , and λ represents the balance parameter.

[0046] Furthermore, the determination method of L model (T) is as follows:

[0047] Define the set of candidate local outlier factors {x 1 , x 2 , ···, x N} corresponding to T i within the set of candidate time windows {T 1 , T 2 , ···, T 3 , ···, T N}, and N represents the total number of nodes within the candidate time window;

[0048] Perform coarse-graining processing, and the specific expression is:

[0049]

[0050] where τ represents the scale factor, M τ represents the sequence length under the scale τ, and y j (τ) represents the average LOF value of the jth group;

[0051] Calculate the sample entropy S(τ):

[0052] S(τ) = SampEn({yj(τ)}, m, r)

[0053] where m represents the embedding dimension, r represents the tolerance, and SampEn(·) represents the sample entropy function;

[0054] Calculate the multi-scale entropy H MES (T):

[0055]

[0056] Among them, τ max represents the selected maximum scale factor;

[0057] Apply H MES (T) to L model (T), and solve for the complexity L model (T).

[0058] Furthermore, if the node v i is recorded as an abnormal node, the correction time of this abnormal node is as follows:

[0059]

[0060] Among them, ΔT corr (v i ) represents the correction time of the abnormal node, represents the local anomaly factor, α represents the sensitivity coefficient, β represents the basic correction time, and E(v i ) represents the environmental impact factor.

[0061] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the steps of any one of the intelligent filling methods for financial information data are implemented.

[0062] A computer-readable storage medium stores a computer program thereon. When the computer program is executed by a processor, the steps of any one of the intelligent filling methods for financial information data are implemented.

[0063] The technical solution provided by the present invention has the following beneficial effects compared with the known prior art:

[0064] By constructing a candidate time window and using multi-scale entropy to calculate L model (T), it can comprehensively capture the uncertainty and complexity of the LOF distribution within the candidate time window at different time scales, not only smoothing short-term noise but also revealing long-term trends and structural changes, making the data description index more robust and representative;

[0065] By combining the volatility of LOF values in different overlapping sub-time periods within the candidate time window, the evaluation function can simultaneously balance the complexity of data description and the stability of local anomaly detection results, automatically selecting the optimal time window, so that when calculating the local reachability density and LOF within this window later, it can more truly reflect the local distribution characteristics of the data and improve the accuracy of financial data anomaly detection;

[0066] In addition, by automatically optimizing the time window, it can be ensured that the most appropriate scale is used for analysis in each time period, thereby providing a more reliable basis for subsequent data correction, filling, and risk warning. This not only improves the accuracy of anomaly detection and correction but also greatly reduces the false alarm risk during the data filling process and improves the filling efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0068] Figure 1 It is a schematic diagram of the overall method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0069] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0070] The present invention will be further described below with reference to the embodiments.

[0071] Embodiment 1 (refer to Figure 1 ): An intelligent filling method for financial information data, at least including:

[0072] Obtain the financial data to be filled.

[0073] Verify the current financial data to be filled and determine whether there is an anomaly in the financial data to achieve anomaly detection of the financial data, including:

[0074] Construct a graph structure, specifically:

[0075] Regard each financial record (or the aggregated subjects / departments selected according to business requirements) as a node in the graph, and construct the edges between the nodes according to business rules, thereby converting the tabular data into graph data. Then, there is:

[0076] Convert each financial record in the financial data into a node in the graph and denote it as v i ;

[0077] Construct the connection relationship between nodes according to business rules to reveal the associations between financial records, then there is:

[0078] If two financial records belong to the same department, construct an edge e ij , that is, e ij represents the edge between nodes v i and v j . If the business rules are satisfied, then e ij is 1, otherwise it is 0. (If the two records are close in time, such as within a certain number of days or in the same month, an edge can also be established; if the subject codes of the two financial records are the same or similar, an edge can be formed);

[0079] Define the entire graph structure G=(V, E), where V={v 1 , v 2 , ···, v n} represents the set of nodes. Correspondingly, E represents the set of edges;

[0080] Define the adjacency matrix A, whose element A ij =e ij ;

[0081] To make each node include its own characteristics during information transmission, introduce the identity matrix I (of size n×n) to define the self-connection matrix

[0082] Subsequently, define the matrix as 's degree matrix, in the form of a diagonal matrix, whose diagonal elements

[0083] n represents the total number of nodes in the graph. Therefore, for the degree of node v i , that is, the total number of edges connected to v i , is obtained by adding up all the elements in the i-th row of the adjacency matrix with self-connections .

[0084] Graph embedding learning, specifically:

[0085] Use the graph convolutional network (GCN) to perform feature propagation on the graph and extract the low-dimensional embedding representations of each node under the global and local graph structures, including:

[0086] Integrate the feature vectors of all nodes into a matrix X. Therefore, X i represents the initial feature of node v i ;

[0087] It should be noted that many anomalies are caused by abnormal relationships between nodes and their surrounding nodes. By aggregating the information of neighbors, the representation of each node can be made more comprehensive, making it easier to detect anomalies that are significantly different from the surrounding nodes. Thus, we have:

[0088] The graph convolutional network updates the node representation through neighbor information aggregation. Among them, the update formula for a single-layer GCN is:

[0089]

[0090] Among them, H (l) represents the node representation matrix of the l-th layer, and initially H (0) = X, W (l) represents the weight matrix of the l-th layer, represents the inverse square root of, σ(·) represents the activation function, such as ReLU or Sigmoid, which is used to introduce non-linearity;

[0091] Therefore, after iterating through multiple layers of GCN, the final low-dimensional embedding vector H (L) of the node is obtained, capturing local and global structural information. Through multiple graph convolutional operations, neighbor information is aggregated to generate a low-dimensional representation that can reflect the structural and attribute relationships of the node in the entire graph. It should be noted that GCN not only extracts the features of the node itself but also comprehensively considers the local and global graph structures and can identify imperceptible structural anomalies.

[0092] Anomaly scoring: In the obtained embedding space, the Local Outlier Factor (LOF) (the local outlier factor method not only considers the distance between the node and its neighbors but also measures the density of the region where the node is located. If the density of a node is much lower than that of its surrounding nodes, it may be an isolated outlier) is used to judge the degree of anomaly of the node, including:

[0093] For each node v i , determine its distance from other nodes and determine the distance to the k-th nearest neighbor, denoted as d k (v i );

[0094] Furthermore, considering that financial data often changes over time, a time window T base is introduced to define the time-constrained k-neighborhood as:

[0095]

[0096] where N k (v i ) represents the set of k nearest neighbors of node v i , t i and t jrespectively represent the timestamps corresponding to node v i and node v j ;

[0097] Define the reachable distance between node v i and its neighbor node v j ;

[0098]

[0099] where, h i and h j respectively represent the feature representations corresponding to the nodes, and ||h i -h j || 2 represents the Euclidean distance between node v i and v j ; represents within T base , the distance from node v j to its k-th nearest neighbor, reflecting the local scale of the node under the current time constraint;

[0100] Thus, calculate the local reachability density

[0101]

[0102] Define the LOF value of node v i , that is, the local outlier factor

[0103]

[0104] It should be noted that financial data often has obvious time series and periodicity (such as monthly, quarterly, and annual financial statements). By restricting the time range of neighbors (that is, introducing the time window T base ), ensure that only records that are close in time participate in the calculation of local reachability density and local outlier factor, so that the comparison is more meaningful, avoiding masking the true local outliers due to too large a time span. And, by introducing time constraints, construct time-weighted so that the calculated local reachability density and local outlier factor can better reflect the actual distribution of financial data in the current period, improving the robustness and accuracy of financial data anomaly detection.

[0105] Therefore, if the local outlier factor is greater than the anomaly threshold, it means that the local density of v i is much lower than that of its neighbors, indicating that v iIt is an abnormal node. Based on the determined abnormal nodes, anomaly detection of financial records in financial data is implemented to facilitate the correction of financial data and its subsequent filling into the financial system.

[0106] Among the above, for the calculated local anomaly factor, if the of a certain node far exceeds the normal range, for example, if the income and profit of a certain month decrease abnormally, it is abnormal data, and these abnormal data usually correspond to a row of financial records, as shown in the following table:

[0107]

[0108] Furthermore, in the anomaly detection of financial data, the size of the time window T base directly affects the calculation of the scores within the local neighborhood:

[0109] If the time window T base is too small, it may cause the neighborhood data to be too scattered, and the scores will fluctuate greatly;

[0110] If the time window T base is too large, it may mask short-term anomaly information and introduce interference from long-term trends;

[0111] Therefore, an evaluation function Q(T) is introduced to balance the two aspects of influence, including:

[0112] Define the set of candidate time windows {T 1 , T 2 , ···, T N}, for example, the candidate time windows T i can be 7 days, 14 days, 30 days, etc. respectively, covering different time scales and being able to compare the performance of the scores under different time scales;

[0113] For each candidate time window T i calculate according to the above expression

[0114] Define the error Lerror(T):

[0115]

[0116] where N represents the total number of nodes within the candidate time window, and Var(·) represents the variance of the set (under a certain candidate time window T i , for each node v i , the ​The variance of the set formed by the scores, that is, the set is included in the same candidate time window T i under different time periods (e.g., under a sliding window) values), reflecting the volatility of the scores in different time periods. The lower the error value, the more stable the scores of the normal data under the candidate time window T i are, and they are less affected by noise;

[0117] Subsequently, construct the evaluation function Q(T):

[0118] Q(T)=L model (T)+λL error (T)

[0119] where L model (T) represents the complexity of the data description under the candidate time window T i under, and λ represents the balance parameter, which is used to control the weight between complexity and error stability, and comprehensively consider the complexity and the stability of the scores, so as to find a balance point under different candidate time windows, so that the selected candidate time window can not only simply describe the data, but also ensure the stability of local anomaly detection.

[0120] Among the above, it should be noted that the multi-scale sample entropy method is introduced to calculate the complexity L model (T), specifically as follows:

[0121] Define the set of candidate time windows {T 1 ,T 2 ,···,T N} within the corresponding set of candidate local anomaly factors {x i ,x 1 ,x 2 ,···,x 3 ,···,x N}(x i all represent the here for the convenience of subsequent writing and representation, replace with x i );

[0122] Perform coarse-grained processing, specifically, divide {x 1 ,x 2 ,x 3 ,···,x N} into several segments according to a fixed scale, take the average value of each segment, and obtain a new sequence {y 1 (τ),y 2 (τ),y 3 (τ),···,y Mτ ​(τ)}, where the specific expression is:

[0123]

[0124] where τ represents the scale factor, which describes the number of data points contained in each group during the coarse-graining process. By changing τ, the behavior of the data at different time scales can be observed. M τ represents the length of the sequence obtained after coarse-graining at scale τ, that is, the integer part of N divided by τ y j (τ) represents the average LOF value of the j-th group, which is a simplified description of N at scale τ. Coarse-graining can smooth out short-term noise and reveal the overall structure of the data at larger time scales, enabling the subsequent entropy value calculation to capture the multi-scale dynamic characteristics of the data;

[0125] Calculate the sample entropy S(τ) (used to measure the unpredictability or complexity of the coarse-grained sequence):

[0126] S(τ) = SampEn({yj(τ)}, m, r)

[0127] where m represents the embedding dimension, r represents the tolerance, and SampEn(·) represents the sample entropy function. It calculates the probability of similar patterns in the sequence under the given embedding dimension m and tolerance r. The sample entropy can quantitatively describe the complexity and uncertainty of the sequence. The higher S(τ) is, the more unpredictable and complex the data is at scale τ; conversely, the more regular it is. This can reflect the dynamic changes of the LOF distribution of the candidate time window set at different scales;

[0128] Furthermore, calculate the multi-scale entropy H MES (T):

[0129]

[0130] where τmax represents the selected maximum scale factor;

[0131] Therefore, apply H MES (T) to L model (T) to solve for the complexity L model (T). It should be noted that the comprehensive multi-scale entropy H MES (T) can comprehensively capture the complexity of the data within the candidate time window set at short-term and long-term scales, avoid information that may be ignored by a single scale, reflect the multi-level structure and uncertainty of the data at different time scales, so as to define the evaluation function Q(T), thereby providing more robustness for the subsequent selection of the optimal time window T″.

[0132] Finally, the optimal time window T″ can be solved:

[0133]

[0134] to update the optimal time window T to be solved to the time window T base and substitute it into the solution process, and then use the evaluation function Q(T) to measure the stability of the score (LOF) and the data description complexity under each candidate time window Then, the method of minimizing the objective function is used to automatically solve the optimal time window T″. This process can adaptively select the optimal time window T″ that is most suitable for the time series characteristics of the current financial data, thereby improving the accuracy, stability, and robustness of financial data anomaly detection, and avoiding the limitations brought by directly presetting the time window.

[0135] It should be noted that the above determines the abnormal nodes in the financial data to determine the abnormal financial data. In order to facilitate the subsequent filling of the corrected financial data, therefore, estimate the time required to correct the current financial data to form a plan for filling the financial data, including:

[0136] If the node v i is recorded as an abnormal node, then the correction time for this abnormal node is:

[0137]

[0138] where ΔT corr (v i ) represents the correction time of the abnormal node, represents the local anomaly factor, α represents the sensitivity coefficient, represents the impact of unit change on the correction time, β represents the basic correction time, represents the basic operation time that needs to be consumed even if the financial data is close to normal, and E(v i ) represents the environmental impact factor, which is obtained by weighting the reflection of the operation load (the current workload of the operating user, the correction time may be extended when the load is high, and it can be obtained by weighting the number of tasks to be processed by the operating user and the system CPU utilization rate) and the financial data complexity (reflecting the data volume or data structure complexity of this financial record, the higher the complexity, the longer the correction time required, where the financial data complexity is calculated according to the field types included in the financial record corresponding to the abnormal node, such as numerical type, date type, text type, and the corresponding numbers are summed up);

[0139] Therefore, obtain ΔT corr (v i ) corresponding to all abnormal nodes and sum them to obtain the total correction time required to correct the current financial data. Add the current actual time to the total correction time to obtain the correction completion time of the financial data, and perform the filling operation of the financial data based on the correction completion time.

[0140] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements will not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for intelligently reporting financial information data, characterized in that: The steps include: Obtain the financial data to be reported; Perform anomaly detection on the financial data to be reported, including: Convert the financial records in the financial data into nodes v in the graph i , and define node v i The corresponding edge e ij ; Constructing the Matrix And use the graph convolutional network to obtain the low-dimensional embedding vector H(L) of the node; The abnormal nodes of financial data are determined by calculating the local abnormal factors, including: Introducing the time window T base Define the k-neighborhood of the time constraint; Compute Node v i Node v with neighbors j The reachable distance is calculated based on the local reachable density, and the local anomaly factor is calculated by combining the reachable distance and the local reachable density to determine the abnormal node; Define candidate time windows, and define evaluation functions based on data complexity and LOF error, determine the optimal time window to calculate and update local anomaly factors and determine abnormal nodes; The financial data is corrected based on the identified abnormal nodes, and the report is filled out based on the corrected financial data.

2. According to claim 1, a method for intelligently reporting financial information data is characterized in that: The construction matrix And use the graph convolutional network to get the low-dimensional embedding vector H of the node (L) The method is: The entire graph structure G = (V, E), where V represents the node set and E represents the edge set; Define the adjacency matrix A, whose elements A ij =e ij ; Introduce the identity matrix I to define the self-connection matrix Defining the Matrix for The degree matrix of Its diagonal elements n represents the total number of nodes in the graph; Graph convolutional networks are used to extract low-dimensional embedding representations of each node in the global and local graph structures, including: Integrate the feature vectors of all nodes into matrix X; The graph convolutional network updates the node representation by aggregating neighbor information, where the update formula for a single-layer GCN is: Among them, H (l) represents the l-th layer node representation matrix, and the initial H (0) =X,W (l) represents the weight matrix of the lth layer, express The inverse square root of , σ(·) represents the activation function; After multiple layers of GCN iteration, we finally get the low-dimensional embedding vector H of the node. (L) .

3. According to claim 2, a method for intelligently reporting financial information data is characterized in that: The calculation method of the reachable distance is: For each node v i , determine its distance to other nodes, and determine the distance to the kth nearest neighbor, denoted as d k (v i ); Based on the time window T base , defining the k-neighborhood of the time constraint for: Where N k (v i ) represents node v i The k nearest neighbor set, t i and t j Respectively represent the node v i and node v j The corresponding timestamp; Define node v i Node v with neighbors j The reachable distance Among them, h i and h j They represent the feature representations corresponding to the nodes, respectively, ||h i -h j ||2 represents node v i With v j The Euclidean distance between Indicates that in T base Inside, node v j The distance to its kth nearest neighbor.

4. The method for intelligently reporting financial information data according to claim 3 is characterized in that: The local reachable density The calculation is performed according to the following relationship:

5. The method for intelligently reporting financial information data according to claim 4 is characterized in that: The local abnormal factor The calculation is performed according to the following relationship:

6. A method for intelligently reporting financial information data according to claim 5, characterized in that: The evaluation function Q(T) is defined according to the following relationship: Q(T)=L mod el (T)+λL error (T) Among them, L mod el (T) represents the candidate time window T i The complexity of the data described below, λ represents the balance parameter.

7. The method for intelligently reporting financial information data according to claim 6 is characterized in that: The L mod el (T) is determined as follows: Define a candidate time window set {T1,T2, ···,T N T in} i The corresponding candidate local anomaly factor set {x1,x2,x3,···,x N }, N represents the total number of nodes in the candidate time window; The specific expression is: Among them, τ represents the scale factor, M τ represents the sequence length at scale τ, y j (τ) represents the average LOF value of the jth group; Calculate the sample entropy S(τ): S(τ)=SampEn({yj(τ)},m,r) Where m represents the embedding dimension, r represents the tolerance, and SampEn(·) represents the sample entropy function; Calculate the multiscale entropy H MES (T): Among them, τ max represents the selected maximum scale factor; H MES (T) applied to L mod el (T), realize the complexity L mod el (T) solution.

8. The method for intelligently reporting financial information data according to claim 5 is characterized in that: If node vi is point V i It is recorded as an abnormal node, and the correction time of the abnormal node is: Where, ΔT corr (v i ) represents the correction time of abnormal nodes, represents the local abnormal factor, α represents the sensitivity coefficient, β represents the basic correction time, E(v i ) represents environmental impact factors.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images