CN113177733B

CN113177733B - Middle and small micro enterprise data modeling method and system based on convolutional neural network

Info

Publication number: CN113177733B
Application number: CN202110554758.XA
Authority: CN
Inventors: 王鑫; 王莹; 陈进东; 张健; 曹丽娜
Original assignee: Beijing Information Science and Technology University
Current assignee: Beijing Information Science and Technology University
Priority date: 2021-05-20
Filing date: 2021-05-20
Publication date: 2023-05-02
Anticipated expiration: 2041-05-20
Also published as: CN113177733A

Abstract

The invention discloses a medium and small micro enterprise data modeling method and a system based on a convolutional neural network, which are specific application of the convolutional neural network in the field of medium and small micro enterprise information processing, further expand the application range of a deep learning algorithm, have higher pertinence and practicability, use time sequence information (such as financial information and the like) and non-time sequence information (such as enterprise information and the like) for enterprise information, have higher innovation in the specific application of the convolutional neural network, and explain the specific method and flow for extracting and fusing the time sequence information and the non-time sequence information characteristics through application scenes of credit risk identification.

Description

Middle and small micro enterprise data modeling method and system based on convolutional neural network

Technical Field

The invention belongs to the technical field of enterprise data processing, and particularly relates to a medium-small micro enterprise data modeling method and system based on a convolutional neural network.

Background

The data of the enterprise comprises data of time series characteristics and data of non-time series characteristics, wherein the data of the time series characteristics are mainly financial data, the data of the non-time series characteristics are mainly non-financial data, the financial data are important indexes for measuring repayment capacity, development capacity, profitability and business capacity of one company, and the non-financial data comprise enterprise information such as registered capital, established years, educational background, business practice years and the like, and can reflect the reality of one company to a certain extent.

Under the background of big data, internet financial data is increasingly complicated, and a comprehensive and objective online data acquisition system is constructed by using basic information and dynamic financial indexes of small and medium enterprises acquired on the internet, so that the complex financial data with time sequence characteristics and enterprise information with non-time sequence characteristics can be effectively processed by using a deep learning method. The convolutional neural network is used as a deep learning method, and financial data with time sequence characteristics and complex financial data can be judged more accurately. In recent years, convolutional neural networks have been introduced into the internet, financial arts, to assess personal, customer and business credit risk. The convolutional neural network not only can automatically extract the characteristics from the data, but also has strong learning ability. By adjusting the network structure, the convolutional neural network can not only perform deep learning on data with time sequence characteristics and perform accurate predictive analysis, but also analyze non-financial data such as enterprise information.

Aiming at the problems that in the prior art, an inaccurate model exists for data processing of small and medium-sized micro enterprises, and the problems of high cost, complexity, low timeliness and the like exist, a convolutional neural network is introduced, and the convolutional neural network is combined with a processing method of time series data and non-time series data, so that a data modeling method of the small and medium-sized micro enterprises based on the convolutional neural network is provided.

Disclosure of Invention

The invention aims to provide a medium-small micro enterprise data modeling method and system based on a convolutional neural network, so that the problems of high cost, high complexity and low timeliness of enterprise credit risk evaluation are solved, the processed data are more accurate, and the method has more important guiding significance for measuring main indexes of enterprises.

In order to achieve the above purpose, the present invention adopts the following technical scheme:

a method for modeling data of a medium and small micro enterprise based on a convolutional neural network, the method comprising:

(1) Collecting related data of small and medium-sized enterprises;

(2) Preprocessing the collected data;

(3) Constructing a convolutional neural network model;

(4) Training and testing the constructed convolutional neural network model, and optimizing and determining parameters of the convolutional neural network model.

Preferably, the preprocessing in step (2) includes a missing value processing and a data normalization processing. The missing value processing is to carry out missing value filling on the downloaded data set, namely filling the current missing value by using the global average value of the index sequence to obtain a complete data set; the data normalization process uses a z-score normalization process.

The method for training and testing the constructed convolutional neural network model in the step (4) comprises the following steps: the method comprises the steps of randomly dividing collected enterprise data into a training set and a testing set, taking the training set data as input, carrying out iterative training on a convolutional neural network model, then testing the model by the testing set, and carrying out parameter adjustment according to a test result.

The convolutional neural network model comprises two parallel sub convolutional neural networks which are respectively used for receiving data of time sequence characteristics and data of non-time sequence characteristics.

Deconvolution neural networks for time series characteristic data: the first layer of convolution layer uses 1X 3 and 3X 1 convolution kernels to extract the related characteristics among different indexes and the time sequence characteristics of the same index, then the characteristics are further extracted by a 2X 2 maximum pooling layer, then a convolution layer only comprising 3X 3 single convolution kernels is connected, and finally downsampling is carried out by a 2X 2 maximum pooling layer;

deconvolution neural networks for non-time series characteristic data: only one layer of 1 x 3 convolution kernels is used, followed by a layer of 2 x 2 pooling layers for feature extraction;

and finally, the output matrixes of the two sub-convolution neural networks are tiled through a flat, multidimensional input is unidimensionally combined and then passes through a full-connection layer, and the output layer after the full-connection layer selects a softmax function as an output classifier.

The convolutional neural network model is optimized by specifically adopting: the target loss function is 'binary-cross sentropy', the optimizer is Adam, the standard for measuring the quality of the model is precision, then in the process of training the model, the value of the evaluation-split is 0.2, the sequence of input samples is randomly disturbed before each epoch, the data size of each batch of training is 10, and the training is carried out for 50 times.

The invention also discloses a medium and small micro enterprise data processing system based on the convolutional neural network, which comprises:

at least one memory cell;

at least one processing unit;

the storage unit stores at least one instruction; the instructions are loaded by at least one processing unit and perform the steps of:

collecting related data of small and medium-sized enterprises;

preprocessing the collected data;

constructing a convolutional neural network model;

training and testing the constructed convolutional neural network model, and optimizing and determining parameters of the convolutional neural network model.

The processing unit includes:

a data acquisition subunit: the method is used for collecting related data of small and medium-sized enterprises;

a data preprocessing subunit: preprocessing the collected data;

model building subunit: the method is used for constructing a convolutional neural network model;

training subunit: the method is used for training and testing the constructed convolutional neural network model.

The invention has the advantages that:

(1) The method is a specific application of the convolutional neural network in the field of information processing of small and medium-sized micro enterprises, further expands the application range of a deep learning algorithm, has higher pertinence and practicability, uses time sequence information (such as financial information and the like) and non-time sequence information (such as enterprise information and the like) for the information of the enterprises, has higher innovation in the specific application of the convolutional neural network, and can more accurately embody the basic condition of the small and medium-sized micro enterprises through quantitative and qualitative multidimensional data analysis.

(2) In the data processing process, different processing methods are respectively adopted for the data with the time sequence characteristics and the data without the time sequence characteristics, and in order to improve the classification accuracy of the data, two convolution layers and two pooling layers are adopted for the sub-convolution network connected with the data with the time sequence characteristics in consideration of more input characteristics of the data with the time sequence characteristics. Considering that the input features of non-time series characteristic data are fewer, in order to avoid the problems of gradient elimination and explosion, overfitting phenomenon is generated, so that only one convolution layer and one pooling layer are adopted in the sub-convolution network connected with the non-time series characteristic data. In order to verify the effectiveness of the method, the data of the middle and small micro enterprises are used for evaluation, and the result shows that the accuracy of the training set is 0.95, the accuracy of the verification set is 0.99, the classification accuracy is higher, and the fitting phenomenon is not generated. Therefore, different processing methods are adopted for the data with the time sequence characteristics and the data without the time sequence characteristics, so that the processing accuracy of enterprise data is improved, and the basic situation of a middle and small micro enterprise can be more accurately embodied.

Drawings

FIG. 1 is a graph of training set accuracy using different numbers of convolutions and pooling layers, respectively;

FIG. 2 is a graph of verification set accuracy using different numbers of convolutions and pooling layers, respectively;

FIG. 3 is a graph of training set accuracy using both 2-layer convolution and 2-layer pooling;

FIG. 4 is a graph of verification set accuracy using both 2-layer convolution and 2-layer pooling;

FIG. 5 is a graph of training set accuracy using both layer 1 convolution and layer 1 pooling;

FIG. 6 is a graph of verification set accuracy using both layer 1 convolution and layer 1 pooling;

fig. 7 is a diagram of a data processing process for a medium or small micro enterprise using a convolutional neural network.

Detailed Description

The invention will be further described with reference to specific embodiments, and advantages and features of the invention will become apparent from the description.

Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art. In addition, any methods and materials similar or equivalent to those described herein can be used in the present invention. The preferred methods and materials described herein are presented for illustrative purposes only.

Embodiment 1A data modeling method for small and medium sized micro enterprises based on convolutional neural network

1. Collecting enterprise related data:

first, small and medium-sized enterprises and startup manufacturing industries are selected from a national security database, wherein the small and medium-sized enterprises and startup manufacturing industries are 1158 in total from ST and non-ST companies, and the financial factors are divided into 4 dimensions of profit capability, debt repayment capability, operation capability and growth capability, and the non-financial factors are divided into 2 dimensions of enterprise quality and enterprise owner quality by referring to a small and medium-sized enterprise credit evaluation study based on an improved dynamic combination evaluation method published in management academic report in Zhang Faming 2019. The financial index selects data of 12 months 31 days 2017, 12 months 31 days 2018 and 12 months 31 days 2019, and the non-financial index selects data of 12 months 31 days 2019. The specific index selections are shown in table 1:

basic information of small and micro enterprises in Table 1

Secondly, quantifying non-financial data under the basic condition of an enterprise, wherein a specific quantification method can refer to a small and micro enterprise credit evaluation research based on an improved dynamic combination evaluation method published in management report in 2019, directly quantifying non-financial index data such as enterprise registered capital, established years, educational background, enterprise owner practitioner years and the like into values between [0 and 1] according to thresholds, for example, dividing the registered capital into three sections according to two thresholds of 1 hundred million yuan and 10 hundred million yuan, and quantifying the three sections into 0.6, 0.8 and 1 respectively; the period of establishment is greater than 9 years, the quantification is 1, the quantification is 0.8 in (7, 9), the quantification is 0.6 in (5, 7), the quantification is 0.4 in (3, 5), the quantification is 0.2, less than or equal to 1 year, the quantification is 0, the education background is research life or more, the quantification is 1, the family quantification is 0.8, the large quantification is 0.7, the middle school and middle school quantification is 0.5, the primary quantification is 0.3, the primary quantification is 0.1, the no-learning quantification is 0, the business owner has a service life of greater than 9 years, the quantification is 1 in (7, 9), the quantification is 0.8 in (5, 7), the quantification is 0.6 in (3, 5), the quantification is 0.4 in (1, 3), the quantification is 0.2, less than or equal to 1 year, and the quantification is 0.

2. Determining a medium and small micro enterprise credit risk judgment standard:

whether or not the small and medium-sized enterprises in the market are judged by ST as a criterion whether or not there is credit risk, the Shanghai deep stock exchange is declared on the 4 th month 22 th 1998, and stock exchanges of the market companies with abnormal financial conditions or other conditions are specially processed according to stock marketing rules implemented in 1998, namely (ST), the enterprises are judged as ST refers to the conditions that the market companies have at least one of the following six conditions:

1) Continuous deficit in the last two years.

2) The last accounting annual audit result shows that its stakeholder equity is below the registered capital, i.e., each equity is below the stock denomination.

3) The registered accountant presents an audit report which cannot express comments or negative comments to the financial report of the last accounting year.

4) The most recent accounting year audited stakeholder equity deducts the portion of the registered accountant, the relevant department, that was not confirmed, below the registered capital.

5) The last annual revenue is adjusted in the last audited financial report, resulting in two consecutive annual losses of accounting.

6) The financial condition is identified as abnormal by the exchange or the Chinese license.

After becoming ST, the marketing company means that the company has a certain risk of paying off capability, the paying off capability of the company may be weakened, the company has a higher credit risk, the paying off capability is reduced, the marketing company is used as a definition standard of the credit risk, ST is an enterprise with high credit risk, and non-sT is an enterprise with low credit risk;

3. preprocessing the data of the time series characteristics and the data of the non-time series characteristics:

the obtained original data may be lost, and meanwhile, different indexes have the problems of non-uniform dimensions and the like, which are mainly represented by uneven values of index values, wherein some index values appear in the form of specific gravity, and some indexes appear in the form of ratio. Firstly, for the problem of data missing, because the index data which needs to be acquired by each enterprise is numerous, if simple deletion is used, the sample data volume is greatly reduced, which is not a preferable mode; for the problem of non-uniform dimension, if the original index value is directly analyzed, the influence of some indexes with overlarge values on the model is amplified, so that the effectiveness of the model is influenced. Thus, the data requires some pre-processing before training the model using the sample data. The preprocessing of the data comprises two steps of missing value processing and data standardization processing.

Missing value processing: because the obtained original data set has the conditions of deficiency and the like, the downloaded data set is subjected to deficiency value filling, namely the current deficiency value is filled by the global average value of the index sequence, and a complete data set is obtained.

Data standardization processing: because the magnitudes of the data sets are different, such as surplus cash guarantee multiple, sales profit margin, capital income rate and the like, and huge differences exist among the magnitudes of the data, in order to eliminate the influence of the different magnitudes among the data, the data of the different magnitudes are uniformly converted into the same magnitude, so that the data are subjected to z-score standardization processing, firstly, the original data are assumed to be subjected to normal distribution, the data are processed through the mean value and the standard deviation of the data sequence in the standardization process, the processed original data conform to the normal standard distribution, and the median value (mu) of the set of observed values is subtracted from the observed values and then divided by the standard deviation (sigma), thereby being beneficial to improving the training speed and the prediction accuracy of the model. The formula is as follows:

x′＝(x-μ)/σ

4. constructing a convolutional neural network model:

and constructing a convolutional neural network model, and forming a stable convolutional neural network model through the design of a structure in the convolutional neural network, a convolutional layer, a pooling layer, an output layer Softmax and a full-connection layer.

5. Training and testing the constructed convolutional neural network model, and learning and optimizing parameters of the convolutional neural network model:

the method comprises the steps of randomly dividing data in an enterprise data information base into a training set and a testing set, taking the training set data as input, carrying out iterative training on a convolutional neural network model, then testing the model by the testing set, and carrying out parameter adjustment according to a test result, so that the model is stable finally.

The model comprises two parallel sub-convolutional neural networks, a first sub-network and a second sub-network, which are respectively used for inputting financial data with time series characteristics and non-financial data without time series characteristics.

In the data processing process, different processing methods are respectively adopted for the data with the time sequence characteristics and the data without the time sequence characteristics, and in order to improve the classification accuracy of the data, two convolution layers and two pooling layers are adopted for a sub-convolution network connected with the data with the time sequence characteristics in consideration of more input characteristics of the data with the time sequence characteristics. Considering that the input features of the non-time series feature data are fewer, in order to avoid the problems of gradient elimination and explosion caused by excessive layers of the convolution layer and the pooling layer, the overfitting phenomenon is generated, so that the sub-convolution network connected with the non-time series feature data only adopts one convolution layer and one pooling layer, as can be seen from fig. 1 and 2, the accuracy of the training set is 0.95, the accuracy of the verification set is 0.99, the classification accuracy is higher, and the overfitting phenomenon is not generated.

If the number of layers of the convolution layer and the pooling layer is too large, problems of gradient extinction and explosion may be faced, and even an overfitting phenomenon may occur. In an experiment of the invention, a first sub-network is used for processing the two data, and under the condition of ensuring that other experimental parameter settings are unchanged, two layers of convolution and two layers of pooling layers are used for data of time sequence characteristics and non-time sequence characteristics, and the result is shown in fig. 3 and 4, the accuracy of a training set is 0.85, the accuracy of a verification set is 0.87, and the classification accuracy is reduced.

If the number of layers of the convolution layer and the pooling layer is too small, the recognition capability of the model on the data characteristics is weak, so that the classification accuracy is reduced. In another experiment of the invention, the second sub-network is used for processing the two data, and under the condition of ensuring that other experimental parameter settings are unchanged, the data of the time sequence characteristics and the non-time sequence characteristics are respectively convolved and pooled, and the result is that the accuracy of the training set is 0.79, the accuracy of the verification set is 0.76 and the classification accuracy is reduced as shown in fig. 5 and 6.

Therefore, specifically, the following processing method is adopted:

for a sub-convolutional neural network receiving time-series characteristic data (in this embodiment financial data): the first layer of convolution layer uses 1X 3 and 3X 1 convolution kernels to extract the related features among different indexes and the time sequence features of the same index, the convolution kernel depth is 128, padding is the same, BN is subjected to batch normalization operation, and the activation function is relu; then filling padding into the same by a 2×2 maximum pooling layer, wherein Dropout is 0.1; then a convolution layer only comprising 3 multiplied by 3 single convolution kernel is connected to further extract the characteristics, the characteristic is filled with the same, the convolution kernel depth is 128, BN is normalized, and the activation function is relu; and finally, downsampling is carried out through a 2×2 max pooling layer, padding is filled to be the same, and Dropout is 0.1.

For a sub-convolutional neural network that receives non-time-series characteristic data (this embodiment refers to non-financial data): because the non-financial data does not have time sequence characteristics, the input information is less, the part of the sub-network only uses a layer of convolution kernel of 1 multiplied by 3, the convolution kernel is filled with the same, the convolution kernel depth is 128, the BN is normalized, and the activation function is relu; a 2 x 2 pooling layer is then used for feature extraction, dropout operation at 0.1.

The output matrixes of the two sub-convolution neural networks are tiled through a flat, multidimensional input is unidimensionally realized, the multidimensional input is combined and then passes through a full-connection layer, and finally, a softmax function is selected as an output classifier by an output layer.

Optimizing a model, wherein a target loss function is 'binary-cross sentropy', an optimizer is Adam (learning rate is 0.004), and a standard for measuring the quality of the model is accuracy; and then in the training model process, the value of the validization-split is 0.2, the sequence of input samples is randomly disturbed before each epoch, the data size of each batch of training is 10, and the training is performed for 50 times.

The training process is to process each sample in the training set, calculate losses of the output result and the label classification result for training, and the testing set is used for obtaining the output result and the label classification result for accuracy calculation, so that the training is not participated; for each sample data (i.e., data for each enterprise), training was performed in the following way so that the data tended to be stable:

1) Input enterprise financial data input_1, data structure (None, 3, 43,1); the enterprise non-financial data input_2 is input, and the data structure is (None, 1,4, 1).

2) For the first subnetwork, step 1) enterprise financial data input_1 is subjected to a first layer convolution operation: the first layer of convolution layer comprises two convolutions, using 1 x 3 and 3 x 1 convolution kernels, respectively, wherein the 1 x 3 convolution kernels are aimed at extracting correlation features between different indices, and the 3 x 1 convolution kernels are aimed at extracting timing features of the same index. According to the convolution kernel decomposition, the n multiplied by n convolution kernel is decomposed into two one-dimensional convolutions (1 multiplied by n, n multiplied by 1), so that the calculation can be accelerated to reduce the parameter scale, and the network depth and nonlinearity are increased because the number of the convolution kernels is doubled; the filling is the same, so that the size of the input vector is unchanged after convolution; the convolution kernel depth is 128, and padding is filled with the same; performing BN batch normalization operation; the activation function is ReLU, and compared with the Sigmoid function, the ReLU eliminates the gradient saturation effect in the part that x is more than or equal to 0, and the calculation of the ReLU is simpler. However, reLU itself also has drawbacks: if the input becomes negative, its gradient equals 0, and the network training will not be successfully completed. Even so, reLU is still one of the most commonly used activation functions in the current deep learning field. The activation functions used by this model are all ReLU.

Both convolution output shapes are (None, 3, 43, 128).

3) For the first subnetwork, connect operation: the results (None, 3, 43, 128) of the two convolutions of the first layer convolution operation of step 2) are output, spliced in depth, and the output results are (None, 3, 43, 256).

4) For the first subnetwork, max pooling: and 3) splicing the output results (None, 3, 43, 256) of the step 3), filling the padding into the same by a 2×2 maximum pooling layer, setting Dropout to 0.1, and outputting the result (None, 2, 22, 256).

5) For the first subnetwork, a second layer convolution operation is performed: outputting the result (None, 2, 22, 256) from the step 4) by using a convolution kernel of 3×3; the filling is same, so that the size of the input vector is unchanged after convolution; the convolution kernel depth is 128; performing BN normalization; the activation function is relu and the output is (None, 2, 22, 128).

6) For the first subnetwork, the second max pooling: downsampling the output result (None, 2, 22, 128) of step 5) through a 2×2 max pooling layer; padding is filled with the same, dropout is operated to 0.1, and the output result is (None, 1, 11, 128).

7) For the first subnetwork, the tile: the flat layer is used to "Flatten" the input, i.e., to unidimensionally unify the input, often used in the transition from the convolutional layer to the fully-connected layer, and to tile the output of step 6) (None, 1, 11, 128) to output (None, 1408).

8) For the second subnetwork, step 7) enterprise non-financial data input_2 convolution operation: a layer of 1 x 3 convolution kernels is used; the filling is same, so that the size of the input vector is unchanged after convolution; the convolution kernel depth is 128; performing BN normalization; the activation function is relu and the output is (None, 1,4, 128).

9) For the second subnetwork, max pooling: using a 2 x 2 pooling layer for the output result (None, 1,4, 128) of step 8); dropout operation was 0.1, and the output was (None, 1,2, 128).

10 For the second subnetwork, tiling layer: the flat layer is used to "Flatten" the input, i.e., to unidimensionally unify the input, often used in the transition from the convolutional layer to the fully-connected layer, and to tile the output (None, 1,2, 128) of step 9) to output (None, 256).

11 A splice operation is performed between the first sub-network and the second sub-network, and the output result (None, 1408) of step 7) is spliced with the output result (None, 256) of step 10), and the output result is (None, 1664).

12 Full connect operation): and (3) performing full connection operation on the output result (None, 1664) of the step 11), wherein units=128, the activation function is relu, and the output result is (None, 128).

18 Output layer): the softmax function is selected as the output classifier.

19 Compiling a model, optimizing the model: the objective loss function is "binary-cross sentropy" (two-class loss function), the optimizer is Adam (learning rate is 0.004), and the standard for measuring the quality of the model is accuracy.

20 Training model: the value of validization-split (0.2) is that one fifth of the data in the training set is designated as a verification set in each training, and the verification set is used for testing relevant indexes of the model, such as accuracy, loss function and the like after each round of training is finished. And the shuffle value is set to true, which indicates that the sequence of input samples is randomly disordered before each epoch in the training process, so that the randomness of the verification set is ensured, and the robustness of the model is improved. The size of batch-size (10) of each batch of data represents that the size of each batch of training data is 10 in the training process. The number of exercises epochs (value 50) represents 50 exercises.

Example 2

A convolutional neural network-based medium and small micro-enterprise data modeling system, the system comprising:

at least one memory cell;

at least one processing unit;

collecting related data of small and medium-sized enterprises;

preprocessing the collected data;

constructing a convolutional neural network model;

training and testing the constructed convolutional neural network model, and learning and optimizing parameters of the convolutional neural network model.

The processing unit includes:

a data preprocessing subunit: preprocessing the collected data;

The specific data processing procedure of the above system adopts the method of example 1.

The above-mentioned embodiments are merely preferred embodiments of the present invention, which are not intended to limit the scope of the present invention, and other embodiments can be easily made by those skilled in the art through substitution or modification according to the technical disclosure in the present specification, so that all changes and modifications made in the principle of the present invention shall be included in the scope of the present invention.

Claims

1. The medium-small micro enterprise data modeling method based on the convolutional neural network is characterized by comprising the following steps of:

collecting related data of small and medium-sized enterprises;

preprocessing the collected data;

constructing a convolutional neural network model;

training and testing the constructed convolutional neural network model, and optimizing and determining parameters of the convolutional neural network model;

the medium-small micro enterprise related data comprises financial data and non-financial data, the financial data is time series characteristic data, the non-financial data is non-time series characteristic data, and the convolutional neural network model comprises two parallel sub convolutional neural networks which are respectively used for receiving the time series characteristic data and the non-time series characteristic data;

for a sub-convolutional neural network receiving time-series characteristic data: the first layer of convolution layer uses 1X 3 and 3X 1 convolution kernels to extract the related characteristics among different indexes and the time sequence characteristics of the same index, then the characteristics are further extracted by a 2X 2 maximum pooling layer, then a convolution layer only comprising 3X 3 single convolution kernels is connected, and finally downsampling is carried out by a 2X 2 maximum pooling layer;

2. The method of claim 1, wherein the preprocessing in step (2) comprises a missing value processing and a data normalization processing.

3. The method according to claim 2, wherein the missing value processing is to perform missing value filling on the downloaded data set, namely filling the current missing value with a global average value of the index sequence to obtain a complete data set; the data normalization process uses a z-score normalization process.

4. The method of claim 1, wherein the training and testing of the constructed convolutional neural network model in step (4) is as follows: the method comprises the steps of randomly dividing collected enterprise data into a training set and a testing set, taking the training set data as input, carrying out iterative training on a convolutional neural network model, then testing the model by the testing set, and carrying out parameter adjustment according to a test result.

5. The method according to claim 1, wherein optimizing the convolutional neural network model specifically uses: the target loss function is 'binary-cross sentropy', the optimizer is Adam, the standard for measuring the quality of the model is precision, then in the process of training the model, the value of the evaluation-split is 0.2, the sequence of input samples is randomly disturbed before each epoch, the data size of each batch of training is 10, and the training is carried out for 50 times.

6. A medium and small micro enterprise data processing system based on convolutional neural network constructed according to any one of claims 1-5, wherein the system comprises:

at least one memory cell;

at least one processing unit;

collecting related data of small and medium-sized enterprises;

preprocessing the collected data;

constructing a convolutional neural network model;

7. The system of claim 6, wherein the processing unit comprises:

a data preprocessing subunit: preprocessing the collected data;