Power enterprise financial management classification optimization method

Abstract

The invention discloses a power enterprise financial management classification optimization method. The method comprises the following steps: S1, collecting and storing power enterprise financial data; s2, the collected financial data are preprocessed; s3, performing clustering analysis on the financial data based on an improved K-means clustering algorithm, clustering similar financial records together, and mining and discovering item sets which frequently appear in the financial data and association relationships thereof based on association rules; s4, selecting features highly related to the predicted target through correlation analysis, and extracting sequence features based on a convolutional neural network; s5, constructing a prediction model based on the bidirectional long-short-term memory neural network and the attention mechanism model, and predicting the time sequence of the financial data by taking the feature vector extracted by the convolutional neural network as input; the financial data of the power enterprise can be analyzed and predicted, financial analysis and wind direction early warning are provided for the enterprise, and financial management efficiency and decision accuracy are improved.

Description

Technical Field

[0001] This invention belongs to the field of financial management technology for power enterprises, and specifically relates to a method for optimizing the classification of financial management in power enterprises. Background Technology

[0002] With the advancement of a series of measures such as transmission and distribution price reform, retail electricity reform, and electricity market reform, the relationship between power grid investment and revenue, costs, electricity volume, and electricity price will undergo significant changes. This will inevitably have a major impact on the content and methods of cash flow management for power companies. Their business methods, such as electricity bill payment, fund settlement, and accounting, will be directly or indirectly affected, and will undergo new changes, either proactively or passively. At the same time, the electricity market will also face more intense competition, and the movement of corporate cash flow, the composition of cash flow, and the timing of cash flow will also face new changes. Power companies' financial management faces many challenges.

[0003] However, traditional financial management methods often rely on experience and subjective judgment, lacking scientific rigor and accuracy. They also suffer from low data processing efficiency, are unable to identify potential financial risks in a timely manner, make inaccurate predictions, and accurately control revenue and costs, thus affecting the operational efficiency and risk control of power companies.

[0004] Therefore, in order to solve the above problems, it is necessary to develop a classification optimization method for financial management of power enterprises. Summary of the Invention

[0005] The purpose of this invention is to overcome the shortcomings of the prior art and provide a classification optimization method for financial management of power enterprises. This method analyzes the financial data of power enterprises, identifies abnormal patterns and potential risks in the data, and predicts the financial data to accurately control revenue and costs. It provides enterprises with financial analysis and risk warning, thereby improving the efficiency of financial management and the accuracy of decision-making.

[0006] The objective of this invention is achieved as follows: a method for optimizing the classification of financial management in power enterprises, comprising the following steps:

[0007] S1. Collect and store financial data of power companies;

[0008] S2. Preprocess the collected financial data, including data cleaning, standardization and normalization;

[0009] S3. Based on the improved K-means clustering algorithm, cluster analysis is performed on financial data to group similar financial records together, and the frequent itemsets and their relationships in the financial data are discovered based on association rule mining.

[0010] S4. Select features that are highly correlated with the prediction target through correlation analysis, and extract sequence features based on convolutional neural networks;

[0011] S5. A prediction model is constructed based on a bidirectional long short-term memory neural network and an attention mechanism model. The feature vector extracted by the convolutional neural network is used as input to predict the time series of financial data.

[0012] Furthermore, in step S2, data cleaning is used to handle missing and outlier values ​​to ensure data integrity and accuracy; data standardization and normalization are used to scale the data to a uniform scale, wherein data standardization converts the data into a normal distribution with a mean of 0 and a standard deviation of 1. In the formula: For standardized data, The original data, The mean of the original data. Standard deviation; data normalization is used to scale the data to... Interval.

[0013] Furthermore, step S3, which involves clustering financial data using the improved K-means clustering algorithm, specifically includes the following steps: (1) Determining the initial cluster centers, including the following steps: ① Using any point in the dataset as the center, calculate the density parameter of each sample data point; ② Based on the density parameter, find the sample point with the largest value as the initial cluster center; ③ Calculate the weighted average Euclidean distance of the sample point, and use this as the radius to form the first cluster set of points within this region. ④ Order In the remaining dataset, the sample point with the largest density parameter is selected as the cluster center, and the second cluster set is obtained with the weighted average Euclidean distance of the sample points as the radius. ⑤ Repeat steps ①-④ until found (2) Construct the initial partition, including the following steps: ① Calculate the distance from each sample point to the selected cluster center; The distance between the initial centers is used to divide the sample points into the nearest clusters, forming the initial partition; ② Calculate the mean of each class in the initial partition as the new center of that class; ③ Calculate the sum of squared errors of the clustering results; (3) Iteratively update the clustering, including the following steps: ① Calculate the sum of squared errors of the new clusters; ② Compare the sum of squared errors of the current partition with the sum of squared errors of the previous partition, if the absolute value of the difference between the two does not exceed the minimum value of the sum of squared errors of the previous partition, the sum of squared errors of the current partition is calculated as follows: Stop the iteration and output the final clustering result; otherwise, repeat the above clustering steps until the clustering result converges.

[0014] Furthermore, in step S3, association rule analysis is performed on the clustered data based on the Apriori algorithm, specifically including the following steps: (1) Scanning the dataset Generate candidate selection set from dataset ( Starting from 1), express Itemsets, each itemset contains (2) Calculate the elements; The support of each itemset is used to exclude itemsets below a certain threshold, and the resulting itemsets are considered frequent itemsets. ( (Starting from 1); (3) Frequent itemsets Combine the elements in to obtain candidates Itemset (4) Repeat steps (2) and (3) to continuously search for frequent itemsets in the dataset during the iteration process until the iteration stops. (5) After generating frequent itemsets, split each frequent itemset into two non-empty subsets and use these two subsets to form association rules. (6) Calculate the confidence of each association rule and retain the association rules that meet the minimum confidence. (7) Calculate the lift of each association rule and retain the association rules with a lift greater than 1.

[0015] Furthermore, the formula for calculating support in step (2) is expressed as follows: , representing itemsets In the dataset The frequency of occurrence in; the confidence level in step (6) uses itemsets. The number of times the combination occurs divided by the itemset The number of occurrences is used to calculate the frequency of occurrence, and the formula is as follows: , indicating in association rules In Under the premise of occurrence The probability of occurrence; the association rules in step (7) The formula for calculating the lift is expressed as: , indicating in Under the premise of occurrence The probability of occurrence and The result is a comparison of the probabilities of each occurrence occurring individually.

[0016] Furthermore, the iteration stopping condition in step (4) includes two conditions: ① the current frequent iterations. Itemset Unable to generate new candidates through further combination Itemset ① The iteration stops when all possible frequent itemset combinations have been found; ② In a certain iteration, all generated candidate itemsets... Itemset Since the support of these candidate itemsets is below the preset minimum support threshold, they will not be considered frequent itemsets and cannot generate new frequent itemsets. Itemset, stop iteration.

[0017] Furthermore, in step S4, the Pearson correlation coefficient is used to calculate the correlation, specifically expressed as follows: In the formula: This represents the Pearson correlation coefficient. Indicates the first Feature values ​​of each sample , Representation of features The mean, Indicates the first The target value for each sample Representation of features The mean.

[0018] Further, in step S4, the convolutional neural network includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the calculation formula for the convolution of the convolutional layer is expressed as: In the formula: and They represent the first Layer, First The feature output of the layer, Represents a weight vector. This represents the convolution operation. Indicates the first Layer offset.

[0019] Furthermore, the structure of the long short-term memory neural network in step S5 includes a forget gate, an input gate, an output gate, and internal memory units; wherein, the forget gate is responsible for deleting useless data information, and its calculation process is represented as follows: The input gate updates the unit state based on the input data and historical information; its calculation process is represented as follows: Internal memory units select and add candidate memory state information. A portion of the information received new memory state information. The calculation process is expressed as follows: , The output gate determines its output based on the input and memory unit; its calculation process is expressed as follows: , ;in: and This represents the activation function. , , , and , , , These represent the weight matrices for the forget gate, input gate, cell state, and output gate, respectively. This represents the state matrix of the previous time step. express The input vector at time t, , , , These represent the offsets of the forget gate, input gate, cell state, and output gate, respectively. This indicates the hidden state within a unit.

[0020] Furthermore, in step S5, the attention mechanism is used to provide the decoder with information about the hidden state of each encoder, and when the encoder encodes the elements of the input sequence, it obtains multiple semantic codes that are combinations of different sequence elements and different weight parameters. Its calculation formula is expressed as: , , Indicates the sequence number One element, Indicates the sequence length. For encoding functions, Represents element right The importance of Let be a probability calculated using softmax. This indicates the degree of matching between the element to be encoded and other elements.

[0021] Due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0022] (1) By performing cluster analysis on financial data based on the improved K-means clustering algorithm, similar financial records are clustered together, and the frequent itemsets and their relationships in the financial data are discovered based on association rule mining, abnormal patterns in the financial data are discovered in a timely manner, potential financial risks are identified, and financial analysis and risk warnings are provided for enterprises.

[0023] (2) By selecting features that are highly correlated with the prediction target through correlation analysis, and extracting sequence features based on convolutional neural networks, the feature selection and extraction can be accurately achieved, thereby effectively improving the prediction accuracy of subsequent models;

[0024] (3) By constructing a prediction model based on a bidirectional long short-term memory neural network and an attention mechanism model, the time series of financial data can be predicted, the revenue and cost of enterprises can be accurately controlled, and the efficiency of financial management and the accuracy of decision-making can be effectively improved. Attached Figure Description

[0025] Figure 1 This is a flowchart of the present invention.

[0026] Figure 2This is a diagram of the prediction model structure in step S5 of the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be further described in detail below through embodiments and in conjunction with the accompanying drawings.

[0028] like Figure 1 , Figure 2 As shown, a method for optimizing the classification of financial management in power enterprises includes the following steps:

[0029] S1. Collect and store financial data of power companies; preferably, use a reasonable financial data table when storing the data, create the table and insert the data through SQL statements to ensure that the organization and management of the data comply with the standards and facilitate subsequent query and analysis.

[0030] S2. Preprocess the collected financial data, including data cleaning, standardization, and normalization.

[0031] Preferably, in step S2, data cleaning is used to handle missing and outlier values ​​to ensure data integrity and accuracy; data standardization and normalization are used to scale the data to a uniform scale to improve model training performance. Specifically, data standardization converts the data into a normal distribution with a mean of 0 and a standard deviation of 1. In the formula: For standardized data, The original data, The mean of the original data. Standard deviation; data normalization is used to scale the data to... Interval.

[0032] S3. Based on the improved K-means clustering algorithm, cluster analysis is performed on financial data to group similar financial records together, and the frequent itemsets and their relationships in the financial data are discovered based on association rule mining.

[0033] Preferably, step S3, which involves clustering financial data using an improved K-means clustering algorithm, specifically includes the following steps: Let the dataset to be clustered be... It contains n samples, and each sample is... Dimension, represented as , The initial cluster centers are respectively .

[0034] (1) Determine the initial cluster centers, including the following steps:

[0035] ① Using any point in the dataset Centered on the weighted average Euclidean distance of the sample data points The number of points within a region of radius is points. Based on distance The density parameter is calculated for each sample data point. In the formula: , Two samples and The weighted average Euclidean distance between them. Two samples and The weighted Euclidean distance between them Formulas for calculating the weights of data points in different dimensions;

[0036] ② Based on the density parameter, find the sample point with the largest value. As the initial cluster center;

[0037] ③ Calculate sample points The weighted average Euclidean distance is used as the radius, and points within this region form the first cluster set. ;

[0038] ④ Order Find the sample point with the largest density parameter in the remaining dataset. Using the weighted average Euclidean distance of the sample points as the radius, a second cluster set is obtained, with the cluster center as the cluster center. ;

[0039] ⑤ Repeat steps ①-④ until found Until there are 1 cluster center.

[0040] (2) Construct the initial partition, including the following steps:

[0041] ① Calculate the distance from each sample point to the selected point. The distance between the initial centers is used to assign the sample points to the nearest cluster, thus forming the initial partition;

[0042] ② Calculate the mean of each class in the initial division, and use it as the new center of that class;

[0043] ③ Calculate the sum of squared errors of the clustering results In the formula: express Therefore Centered on Points within a region of radius .

[0044] (3) Iteratively update the clusters, including the following steps:

[0045] ① Calculate the sum of squared errors for the new cluster;

[0046] ② Compare the sum of squared errors of this partition with the sum of squared errors of the previous partition. If the absolute value of the difference between the two does not exceed... Stop the iteration and output the final clustering result; otherwise, repeat the above clustering steps until the clustering result converges.

[0047] Preferably, step S3, which involves performing association rule analysis on the clustered data based on the Apriori algorithm, specifically includes the following steps:

[0048] (1) Scan the dataset Generate candidate selection set from dataset ( Starting from 1), express Itemsets, each itemset contains Each element.

[0049] (2) Calculation The support of each itemset is used to exclude itemsets below a certain threshold, and the resulting itemsets are considered frequent itemsets. ( Starting from 1); preferably, the formula for calculating support is expressed as: , representing itemsets In the dataset The frequency of occurrence.

[0050] (3) Frequent itemsets Combine the elements in to obtain candidates Itemset .

[0051] (4) Repeat steps (2) and (3) to continuously search for frequent itemsets in the dataset during the iteration process until the iteration stopping condition is met; preferably, the iteration stopping condition includes two: ① the current frequent itemsets Itemset Unable to generate new candidates through further combination Itemset ① The iteration stops when all possible frequent itemset combinations have been found; ② In a certain iteration, all generated candidate itemsets... Itemset Since the support of these candidate itemsets is below the preset minimum support threshold, they will not be considered frequent itemsets and cannot generate new frequent itemsets. Itemset, stop iteration.

[0052] (5) After generating frequent itemsets, split each frequent itemset into two non-empty subsets and use these two subsets to form association rules.

[0053] (6) Calculate the confidence score of each association rule and retain the association rules that meet the minimum confidence score; preferably, the confidence score is calculated using itemsets. The number of times the combination occurs divided by the itemset The number of occurrences is used to calculate the frequency of occurrence, and the formula is as follows: , indicating in association rules In Under the premise of occurrence The probability of occurrence.

[0054] (7) Calculate the lift of each association rule and retain association rules with a lift greater than 1; association rules The formula for calculating the lift is expressed as: , indicating in Under the premise of occurrence The probability of occurrence and The result is a comparison of the probabilities of each occurrence occurring individually.

[0055] S4. Select features that are highly correlated with the prediction target through correlation analysis, and extract sequence features based on convolutional neural networks.

[0056] Preferably, in step S4, the Pearson correlation coefficient is used to calculate the correlation, specifically expressed as follows: In the formula: This represents the Pearson correlation coefficient. Indicates the first Feature values ​​of each sample , Representation of features The mean, Indicates the first The target value for each sample Representation of features The mean.

[0057] Preferably, the convolutional neural network in step S4 includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the calculation formula for the convolution of the convolutional layer is expressed as: In the formula: and They represent the first Layer, First The feature output of the layer, Represents a weight vector. This represents the convolution operation. Indicates the first The layer bias; preferably, in step S4, the ReLU activation function is used to extract sequence features based on the convolutional neural network, expressed as: The range is .

[0058] S5. A prediction model is constructed based on a bidirectional long short-term memory neural network and an attention mechanism model. The feature vector extracted by the convolutional neural network is used as input to predict the time series of financial data.

[0059] Preferably, the structure of the long short-term memory neural network in step S5 includes a forget gate, an input gate, an output gate, and internal memory units; wherein, the forget gate is responsible for deleting useless data information, and its calculation process is represented as follows: The input gate updates the unit state based on the input data and historical information; its calculation process is represented as follows: Internal memory units select and add candidate memory state information. A portion of the information received new memory state information. The calculation process is expressed as follows: , The output gate determines its output based on the input and memory unit; its calculation process is expressed as follows: , ;in: and This represents the activation function. , , , and , , , These represent the weight matrices for the forget gate, input gate, cell state, and output gate, respectively. This represents the state matrix of the previous time step. express The input vector at time t, , , , These represent the offsets of the forget gate, input gate, cell state, and output gate, respectively. This indicates the hidden state within a unit.

[0060] Preferably, in step S5, the bidirectional long short-term memory neural network and attention mechanism employ the Sigmoid activation function, expressed as: The range is The loss function is the root mean square error function, and the optimizer is the Adam optimization algorithm.

[0061] Preferably, in step S5, the attention mechanism is used to provide the decoder with information about the hidden state of each encoder, and when the encoder encodes the elements of the input sequence, it obtains multiple semantic codes composed of different sequence elements and different weight parameters. Its calculation formula is expressed as: , , Indicates the sequence number One element, Indicates the sequence length. For encoding functions, Represents element right The importance of Let be a probability calculated using softmax. This indicates the degree of matching between the element to be encoded and other elements.

[0062] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing the classification of financial management in power enterprises, characterized in that: Includes the following steps: S1. Collect and store financial data of power companies; S2. Preprocess the collected financial data, including data cleaning, standardization and normalization; S3. Based on the improved K-means clustering algorithm, cluster analysis is performed on financial data to group similar financial records together, and the frequent itemsets and their relationships in the financial data are discovered based on association rule mining. S4. Select features that are highly correlated with the prediction target through correlation analysis, and extract sequence features based on convolutional neural networks; S5. A prediction model is constructed based on a bidirectional long short-term memory neural network and an attention mechanism model. The feature vector extracted by the convolutional neural network is used as input to predict the time series of financial data.

2. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: In step S2, data cleaning is used to handle missing and outlier values ​​to ensure data integrity and accuracy; data standardization and normalization are used to scale the data to a uniform scale, wherein data standardization converts the data into a normal distribution with a mean of 0 and a standard deviation of 1. In the formula: For standardized data, The original data, The mean of the original data. Standard deviation; data normalization is used to scale the data to... Interval.

3. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: The step S3, which involves clustering financial data using the improved K-means clustering algorithm, specifically includes the following steps: (1) Determining the initial cluster centers, including the following steps: ① Taking any point in the dataset as the center, calculate the density parameter of each sample data point; ② Based on the density parameter, find the sample point with the largest value as the initial cluster center; ③ Calculate the weighted average Euclidean distance of the sample point, and use this as the radius to form the first cluster set of points within this region. ④ Order In the remaining dataset, the sample point with the largest density parameter is selected as the cluster center, and the second cluster set is obtained with the weighted average Euclidean distance of the sample points as the radius. ⑤ Repeat steps ①-④ until found (2) Construct the initial partition, including the following steps: ① Calculate the distance from each sample point to the selected cluster center; The distance between the initial centers is used to divide the sample points into the nearest clusters, forming the initial partition; ② Calculate the mean of each class in the initial partition as the new center of that class; ③ Calculate the sum of squared errors of the clustering results; (3) Iteratively update the clustering, including the following steps: ① Calculate the sum of squared errors of the new clusters; ② Compare the sum of squared errors of the current partition with the sum of squared errors of the previous partition, if the absolute value of the difference between the two does not exceed the minimum value of the sum of squared errors of the previous partition, the sum of squared errors of the current partition is calculated as follows: Stop the iteration and output the final clustering result; otherwise, repeat the above clustering steps until the clustering result converges.

4. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: The step S3, which involves performing association rule analysis on the clustered data based on the Apriori algorithm, specifically includes the following steps: (1) Scanning the dataset Generate candidate selection set from dataset ( Starting from 1), express Itemsets, each itemset contains (2) Calculate the elements; The support of each itemset is used to exclude itemsets below a certain threshold, and the resulting itemsets are considered frequent itemsets. ( (Starting from 1); (3) Frequent itemsets Combine the elements in to obtain candidates Itemset (4) Repeat steps (2) and (3) to continuously search for frequent itemsets in the dataset during the iteration process until the iteration stops. (5) After generating frequent itemsets, split each frequent itemset into two non-empty subsets and use these two subsets to form association rules. (6) Calculate the confidence of each association rule and retain the association rules that meet the minimum confidence. (7) Calculate the lift of each association rule and retain the association rules with a lift greater than 1.

5. The method for optimizing the classification of financial management in power enterprises according to claim 4, characterized in that: The formula for calculating support in step (2) is expressed as follows: , representing itemsets In the dataset The frequency of occurrence in; the confidence level in step (6) uses itemsets. The number of times the combination occurs divided by the itemset The number of occurrences is used to calculate the frequency of occurrence, and the formula is as follows: , indicating in association rules In Under the premise of occurrence The probability of occurrence; the association rules in step (7) The formula for calculating the lift is expressed as: , indicating in Under the premise of occurrence The probability of occurrence and The result is a comparison of the probabilities of each occurrence occurring individually.

6. The method for optimizing the classification of financial management in power enterprises according to claim 4, characterized in that: The iteration stopping condition in step (4) includes two conditions: ① the current frequent iterations. Itemset Unable to generate new candidates through further combination Itemset ① The iteration stops when all possible frequent itemset combinations have been found; ② In a certain iteration, all generated candidate itemsets... Itemset Since the support of these candidate itemsets is below the preset minimum support threshold, they will not be considered frequent itemsets and cannot generate new frequent itemsets. Itemset, stop iteration.

7. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: In step S4, the Pearson correlation coefficient is used to calculate the correlation, specifically as follows: In the formula: This represents the Pearson correlation coefficient. Indicates the first Feature values ​​of each sample , Representation of features The mean, Indicates the first The target value for each sample Representation of features The mean.

8. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: The convolutional neural network in step S4 includes an input layer, a convolutional layer, a pooling layer, a fully connected layer, and an output layer; the formula for calculating the convolution of the convolutional layer is expressed as: In the formula: and They represent the first Layer, First The feature output of the layer, Represents a weight vector. This represents the convolution operation. Indicates the first Layer offset.

9. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: The structure of the Long Short-Term Memory (LSTM) neural network in step S5 includes a forget gate, an input gate, an output gate, and internal memory units; wherein, the forget gate is responsible for deleting useless data information, and its calculation process is represented as follows: The input gate updates the unit state based on the input data and historical information; its calculation process is represented as follows: Internal memory units select and add candidate memory state information. A portion of the information received new memory state information. The calculation process is expressed as follows: , The output gate determines its output based on the input and memory unit; its calculation process is expressed as follows: , ;in: and This represents the activation function. , , , and , , , These represent the weight matrices for the forget gate, input gate, cell state, and output gate, respectively. This represents the state matrix of the previous time step. express The input vector at time t, , , , These represent the offsets of the forget gate, input gate, cell state, and output gate, respectively. This indicates the hidden state within a unit.

10. The method for optimizing the classification of financial management in power enterprises according to claim 1, characterized in that: In step S5, the attention mechanism is used to provide the decoder with information about the hidden state of each encoder, and when the encoder encodes the elements of the input sequence, it will obtain multiple semantic codes composed of different sequence elements and different weight parameters. Its calculation formula is expressed as: , , Indicates the sequence number One element, Indicates the sequence length. For encoding functions, Represents element right The importance of Let be a probability calculated using softmax. This indicates the degree of matching between the element to be encoded and other elements.

Images