Financial data analysis method and system for reconnaissance design business

Abstract

The invention relates to the technical field of financial analysis, and a financial data analysis method comprises the steps: obtaining historical structured data, and extracting important parameters affecting financial fluctuation as a monitoring array; constructing a financial data early warning model based on the cross neural network; analyzing current and ideal structured data by using the model to generate early warning information; and an analysis conclusion is obtained by replacing important parameters in the early warning information and comparing the important parameters with ideal data. Through real-time monitoring and prediction, potential financial risks are found in time, financial management is optimized, the method is especially suitable for the investigation and design industry with long project period and complex fund flow, and the method helps enterprises to reduce financial risks and improve competitiveness.

Description

Technical Field

[0001] The present invention relates to the technical field of financial analysis, and in particular to a financial data analysis method and system for survey and design business. Background Art

[0002] Financial monitoring mainly provides prompts and alarms for data anomalies in existing financial data. It helps financial personnel to find financial loopholes in a timely manner, and facilitates processing and remediation. Financial forecasting mainly effectively controls future uncertainties based on existing financial data, which is conducive to the implementation of financial plans.

[0003] The existing financial system of the survey and design industry lacks intelligence. It simply displays data and cannot fully meet the needs of financial personnel. It cannot prompt loopholes in financial data, nor can it effectively analyze financial data, which increases the financial risks of enterprises. Summary of the invention

[0004] The purpose of the present invention is to provide a financial data analysis method and system for survey and design business, so as to solve the problem that the existing financial system of the survey and design industry is not intelligent.

[0005] To achieve the above object, the present invention provides the following technical solution: a financial data analysis method for survey and design business, comprising the following steps:

[0006] Step 1: Obtain structured data of historical survey and design business, extract important parameters that affect financial fluctuations from the structured data, and use the important parameters as monitoring arrays of the structured data;

[0007] Step 2: Based on the cross neural network, use the monitoring array to build a financial data early warning model;

[0008] Step 3: Use the financial data early warning model to analyze the existing structured data and ideal structured data of the current survey and design business, and obtain the early warning information of the existing monitoring array in the current survey and design business;

[0009] Step 4: Use ideal data to replace important parameters in the warning information, and compare the warning information of the financial data warning model to obtain analytical conclusions.

[0010] Furthermore, the extraction of important parameters affecting financial fluctuations from structured data specifically includes the following steps:

[0011] Step 11: weighting the structured data using a dynamic attenuation weighted frequency adjustment method, and normalizing the weight values ​​of the structured data;

[0012] Step 12: Sort the normalized weight values ​​of the structured data, and take the top five structured data with the highest weight as important parameters;

[0013] Among them, the dynamic attenuation weighted frequency adjustment method is calculated by the formula Indicates that, W i is the comprehensive weight value of the i-th factor;

[0014] X i,j is the number of occurrences of the i-th factor in the abnormal data at the j-th stage;

[0015] Y j is the total number of abnormal data at the jth order;

[0016] β j is the importance coefficient of the jth order, reflecting the degree of influence of this stage on the enterprise's property;

[0017] α is the time decay factor, which controls the decay rate of data over time;

[0018] t i is the time interval between the data of the i-th factor and the current time, in quarters;

[0019] The monitoring arrays are direct cost fluctuation data, contract payment delay data, project schedule deviation data, revenue drop data and customer churn rate data.

[0020] Furthermore, the construction of a financial data early warning model based on a cross neural network and using a monitoring array specifically includes the following steps:

[0021] Step 21: Clean the historical structured data and perform feature standardization on the cleaned monitoring array to obtain monitoring information;

[0022] Step 22: Based on the input layer, cross layer, depth layer and output layer architecture of the cross neural network, the monitoring information is used as input to obtain an initial early warning model;

[0023] Step 23: Perform a time series interactive fusion evaluation on the initial early warning model. If the time series interactive fusion evaluation is qualified, the initial early warning model is a financial data early warning model, otherwise the initial early warning model is optimized.

[0024] Furthermore, the time series interactive fusion evaluation of the initial warning model specifically includes the following steps:

[0025] Step 231: Filter historical structured data with a project cycle greater than 4 quarters and containing a monitoring array, and execute step 21 for the filtered historical structured data;

[0026] Step 232: construct a feature interaction matrix using the monitoring information obtained in step 231;

[0027] Step 233: Perform a time series consistency test on the feature interaction matrix to obtain a time series interaction fusion evaluation result.

[0028] Furthermore, the constructing of the feature interaction matrix specifically includes the following steps:

[0029] Step 2321: Divide the monitoring information obtained in step 231 into time windows to obtain a number of overlapping fragment information;

[0030] Step 2322: Calculate the Spearman rank correlation coefficients between the five important parameters in the segment information, and use the Spearman rank correlation coefficients to construct an initial interaction matrix;

[0031] Step 2323: introducing a time decay factor to weight the initial interaction matrix, and performing weighted average update and normalization processing on the weighted initial interaction matrix to obtain a feature interaction matrix;

[0032] Among them, the window period is monthly and the overlapping length is not less than one week.

[0033] Furthermore, the time series consistency test of the feature interaction matrix to obtain the time series interaction fusion evaluation result specifically includes the following steps:

[0034] Step 2331: Calculate the difference between the predicted value of the initial warning model and the actual value at the time corresponding to the fragment information, and average the difference values ​​of all fragment information, and use the average value as the consistency score of the long-term trend;

[0035] Step 2332: Calculate the fluctuation range between the predicted value and the actual value of the initial warning model at the corresponding time of the adjacent fragment information, and average all the fluctuation ranges, and use the average range value as the short-term fluctuation consistency score;

[0036] Step 2333: Calculate the deviation between the Spearman rank correlation coefficient of the initial warning model at the time corresponding to the fragment information and the actual Spearman rank correlation coefficient, and average all the deviation values, and use the average value as the interaction effect capture ability score;

[0037] Step 2334: Calculate the change rate between the predicted value and the actual value of the initial warning model at the corresponding time of the adjacent fragment information, and average all the change rates, and use the average change rate as the interaction effect stability score;

[0038] Step 2335: Determine whether the long-term trend consistency score, short-term volatility consistency score, interaction effect capture ability score and interaction effect stability score all meet the model qualification threshold. If so, the time series interaction fusion evaluation result is qualified, otherwise it is unqualified.

[0039] Furthermore, the warning information is obtained by inputting the existing structured data of the current survey and design business into the financial data warning model. If the degree of deviation between the monitoring array predicted by the financial data warning model and the ideal structured data is greater than the safety threshold, then the warning information is the important parameters of the risk in the monitoring array and the predicted risk time; otherwise, the warning information is a low risk of financial information.

[0040] Furthermore, the analysis conclusion obtained by comparing the changes in the warning information of the financial data warning model specifically includes: if the warning information changes to low risk of financial information, the analysis conclusion is high risk of financial data, otherwise the analysis conclusion is unclear financial risk.

[0041] A financial data analysis system for survey and design business, including the following modules:

[0042] Data collection and preprocessing module, used to obtain structured data of historical and current survey and design business, and clean, standardize and normalize the data;

[0043] The financial data early warning model construction module is based on a cross neural network and uses a monitoring array to build a financial data early warning model and conduct a time series interactive fusion evaluation on it;

[0044] The early warning information generation module uses the financial data early warning model to analyze the existing structured data of the current survey and design business and generate early warning information;

[0045] The ideal structured data comparison module is used to obtain the ideal structured data and compare it with the warning information of the current monitoring array to generate analysis conclusions.

[0046] The present invention can timely discover potential financial risks by acquiring and analyzing the structured data of the current survey and design business in real time, and using the financial data early warning model for prediction. The financial data early warning model is based on a cross neural network, and introduces designs such as time series interactive fusion evaluation and ideal data comparison to achieve real-time monitoring and accurate early warning of the company's financial status. It can not only help companies to discover potential financial risks in a timely manner, but also optimize financial management and reduce financial risks through comparison and accurate early warning, ensuring that the company maintains a stable operation in a complex financial environment. Especially for the survey and design industry, where the project cycle is long and the capital flow is complex, this solution can provide companies with more scientific and reasonable decision-making support and enhance the overall competitiveness of the company. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 The present invention is an exemplary flow chart of a financial data analysis method for survey and design business provided in accordance with the present application. DETAILED DESCRIPTION

[0048] The following will be combined with the drawings in the embodiments of the present invention to clarify the technical solutions in the embodiments of the present invention; it is clearly described in detail. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] A financial data analysis method for survey and design business, comprising the following steps:

[0050] Step 1: Obtain structured data of historical survey and design business, extract important parameters that affect financial fluctuations from the structured data, and use the important parameters as monitoring arrays of the structured data;

[0051] Step 2: Based on the cross neural network, use the monitoring array to build a financial data early warning model;

[0052] Step 3: Use the financial data early warning model to analyze the existing structured data and ideal structured data of the current survey and design business, and obtain the early warning information of the existing monitoring array in the current survey and design business;

[0053] Step 4: Use ideal data to replace important parameters in the warning information, and compare the warning information of the financial data warning model to obtain analytical conclusions.

[0054] The present invention can timely discover potential financial risks by acquiring and analyzing the structured data of the current survey and design business in real time, and using the financial data early warning model for prediction. The financial data early warning model is based on a cross neural network, and introduces designs such as time series interactive fusion evaluation and ideal data comparison to achieve real-time monitoring and accurate early warning of the company's financial status. It can not only help companies to discover potential financial risks in a timely manner, but also optimize financial management and reduce financial risks through comparison and accurate early warning, ensuring that the company maintains a stable operation in a complex financial environment. Especially for the survey and design industry, where the project cycle is long and the capital flow is complex, this solution can provide companies with more scientific and reasonable decision-making support and enhance the overall competitiveness of the company.

[0055] Specifically, extracting important parameters that affect financial fluctuations from structured data includes the following steps:

[0056] Step 11: Use the dynamic decay weighted frequency adjustment method to weight the structured data, and normalize the weight value of the structured data; the dynamic decay weighted frequency adjustment method is a method for weighting structured data proposed by the present invention, which aims to dynamically adjust the weight of each factor according to the time characteristics of the data, abnormal conditions, and the degree of impact on the financial affairs of the enterprise. This method introduces the time decay factor and the stage importance coefficient to ensure that the recent data and the data of the key stage occupy a more important position in the analysis, so as to more accurately reflect the current financial situation.

[0057] Step 12: Sort the normalized weight values ​​of the structured data, and use the top five structured data as important parameters; the normalization process is to eliminate the dimensional differences between different financial indicators and ensure the comparability of each factor in the weight calculation. By normalizing the weight values ​​of the structured data, the weight values ​​of all factors can be mapped to the interval of [0, 1], which is convenient for subsequent sorting and screening.

[0058] The dynamic attenuation weighted frequency adjustment method is calculated by the formula It indicates that,

[0059] W i is the comprehensive weight value of the i-th factor;

[0060] X i,j is the number of occurrences of the i-th factor in the abnormal data at the j-th stage;

[0061] Y j is the total number of abnormal data at the jth order;

[0062] β j is the j-th order importance coefficient, reflecting the degree of impact of this stage on the company's finances; assuming that the project is divided into multiple key stages, such as project launch, design, construction, acceptance, etc. Each stage has a different impact on finance, so a different weight adjustment mechanism needs to be set for each stage.

[0063] Project start-up phase: Beta 1 =0.5 (large initial investment, but relatively small financial impact);

[0064] Design phase: Beta 2 =0.7 (cost control and revenue recognition in the design phase are relatively stable);

[0065] Construction phase: β 3 =1.0 (the cost and risk during the construction phase are high, and the financial impact is significant);

[0066] Acceptance stage: Beta 4 =0.8 (Revenue recognition is more concentrated during the acceptance phase, but cost fluctuations are smaller).

[0067] α is the time decay factor, which controls the decay rate of data over time. It usually takes a value between 0 and 1. The closer it is to 1, the slower the data decays, and the closer it is to 0, the faster the data decays.

[0068] t i is the time interval between the data of the i-th factor and the current time, in quarters;

[0069] Abnormal data refers to financial data that is beyond the normal range and may indicate potential risks or opportunities. For example, a sudden and significant increase in direct costs or delayed contract payments are all abnormal data.

[0070] The monitoring array is direct cost fluctuation data, contract payment delay data, project schedule deviation data, revenue drop data and customer churn rate data. The top five structured data with the highest weights are selected as important parameters. These five parameters will play a key role in the subsequent financial data analysis.

[0071] Direct cost fluctuation data is the most direct expenditure item in the operation of an enterprise. Its fluctuation directly affects the profit level of the enterprise. If the direct cost fluctuates abnormally, it may indicate that the project budget is overspent or the cost control is improper. Contract payment delay data: Contract payment delay may lead to tight cash flow of the enterprise and affect the normal operation of the enterprise. Especially for the survey and design industry, the project cycle is long, and the timeliness of contract payment is particularly important. Project progress deviation data: Project progress deviation reflects the execution of the project. Progress delay may lead to problems such as extended construction period and increased cost. Therefore, project progress deviation is an important indicator for assessing the health of the project. Revenue drop data: Revenue drop may be caused by changes in market demand, customer loss, etc., which directly affects the profitability of the enterprise. Timely discovery of signs of revenue drop can help the enterprise take measures to cope with market changes. Customer churn rate data: Customer churn rate reflects the market competitiveness and customer satisfaction of the enterprise. If the customer churn rate is too high, it may indicate that the enterprise is at a disadvantage in market competition and needs to strengthen customer relationship management.

[0072] Based on the cross neural network, the construction of the financial data early warning model using the monitoring array includes the following steps:

[0073] Step 21: Clean the historical structured data and perform feature standardization on the cleaned monitoring array to obtain monitoring information; data preprocessing can ensure the high quality of data input into the model and avoid analysis deviations caused by data quality issues.

[0074] Step 22: Based on the input layer, cross layer, depth layer and output layer architecture of the cross neural network, the monitoring information is used as input to obtain the initial early warning model; the cross neural network can capture the complex interactive relationship between multi-dimensional features by introducing the cross layer, and improve the prediction accuracy of the model. Especially in the survey and design industry, the project cycle is long, the capital flow is complex, and there are complex interactions between multiple financial indicators. The cross neural network can more accurately reflect these interactive effects. Specifically, the input layer receives the standardized monitoring information as the input of the model.

[0075] The dimension of the input layer is equal to the number of features in the monitoring array (i.e., 5 important parameters: direct cost fluctuations, contract payment delays, project schedule deviations, revenue drops, and customer churn rates);

[0076] The cross layer is the core part of the cross neural network, responsible for capturing the interaction between different features. The cross layer is calculated by formula x (k+1) =x k *(W k x (k) +bx)+x (k) Calculate. Where x (k) is the input of the first layer, W k is the weight matrix of the cross layer, b k The bias term, * is the element-by-element multiplication. The cross layer gradually enhances the interaction effect between important parameters in the monitoring array by stacking layer by layer, and finally outputs a more complex feature combination.

[0077] The deep layer consists of multiple fully connected layers and is responsible for processing the complex feature combinations output by the cross-layer. The deep layer introduces nonlinear transformations through nonlinear activation functions to improve the expressiveness of the model. The number of deep layers and the number of neurons in each layer can be adjusted according to actual needs.

[0078] The output layer is responsible for generating the final warning information. For the financial risk warning problem, the output layer can use the Sigmoid activation function to output a probability value between 0 and 1, indicating the possibility of financial risk.

[0079] Step 23: Perform a time series interaction fusion evaluation on the initial warning model. If the time series interaction fusion evaluation is qualified, the initial warning model is a financial data warning model, otherwise the initial warning model is optimized. The time series interaction fusion evaluation is divided into three steps: screening suitable historical data, building a feature interaction matrix, and performing a time series consistency test. Each step is crucial and together determines the reliability and effectiveness of the evaluation results. The time series interaction fusion evaluation can comprehensively test the model's prediction ability and its ability to capture complex interactions between multi-dimensional features by comprehensively evaluating the model's long-term trend consistency, short-term fluctuation consistency, interaction effect capture ability, and interaction effect stability. This evaluation method not only considers the static prediction accuracy of the model, but also its performance in dynamic time series data to ensure that the model can adapt to the company's changing financial environment.

[0080] The time series interactive fusion evaluation of the initial warning model specifically includes the following steps:

[0081] Step 231: Filter historical structured data with a project cycle greater than 4 quarters and containing monitoring arrays, and perform step 21 on the filtered historical structured data; the purpose of selecting a longer project cycle is to ensure that the evaluation results can reflect the long-term financial trend of the enterprise and avoid misjudgment caused by short-term fluctuations. For the survey and design industry, the project cycle is usually longer, so selecting a longer project cycle can better capture changes in the financial status of the enterprise. The parameters in the monitoring array are the basis for building a financial data early warning model. The lack of any parameter will affect the prediction accuracy and reliability of the model.

[0082] Step 232: Use the monitoring information obtained in step 231 to construct a feature interaction matrix; the feature interaction matrix is ​​used to capture the complex interaction relationship between multiple financial indicators. Traditional neural network models usually only focus on the linear relationship of a single feature and have difficulty capturing the nonlinear interaction effect between multidimensional features. By constructing a feature interaction matrix, the cross neural network can better capture the interaction between different financial indicators and improve the prediction accuracy of the model.

[0083] Step 233: Perform a time series consistency test on the feature interaction matrix to obtain a time series interaction fusion evaluation result.

[0084] Furthermore, constructing the feature interaction matrix specifically includes the following steps:

[0085] Step 2321: Divide the monitoring information obtained in step 231 into time windows to obtain a number of overlapping fragments of information, with a window period of one month and an overlapping length of no less than one week. The purpose of time window division is to ensure the dynamic and representativeness of the evaluation results. Overlapping time fragments capture financial fluctuations in different time periods.

[0086] Step 2322: Calculate the Spearman rank correlation coefficient between the five important parameters in the fragment information, and use the Spearman rank correlation coefficient to construct the initial interaction matrix; the Spearman rank correlation coefficient is a non-parametric statistical method that can effectively capture nonlinear interaction effects. By calculating the Spearman rank correlation coefficient between direct cost fluctuations, contract payment delays, project schedule deviations, revenue drops, and customer churn rates, the interaction between different financial indicators can be quantified. The Spearman rank correlation coefficient is expressed by the formula Calculate, where d i is the rank difference of the i-th observation, and n is the number of samples. The initial interaction matrix reflects the interaction between different financial indicators, and each element in the matrix represents the strength of the correlation between two features. For example, element A in the initial interaction matrix ij Represents the Spearman rank correlation coefficient between the i-th feature and the j-th feature in the monitoring array. The initial interaction matrix is ​​a 5×5 symmetric matrix, where the elements on the diagonal are 1 (indicating the correlation between the same feature and itself), and the elements on the off-diagonal represent the correlation between different features.

[0087] Step 2323: Introduce the time decay factor to weight the initial interaction matrix, and perform weighted average update and normalization on the weighted initial interaction matrix to obtain the feature interaction matrix; in order to ensure that recent data occupies a more important position in the evaluation, introduce the time decay factor α to weight the initial interaction matrix. Specifically, the weight of the initial interaction matrix is ​​calculated by formula A ij ′=A ij ×α t Indicates that t is the time interval (in quarters), α is the time decay factor, and its value range is (0, 1]. The closer it is to 1, the slower the data decays, and the closer it is to 0, the faster the data decays. The value of the time decay factor can be adjusted according to the industry characteristics and business cycle of the enterprise. The weighted initial interaction matrix can better reflect the current financial situation and avoid analysis bias caused by too much historical data. The normalized feature interaction matrix not only retains the original interaction relationship, but also eliminates the dimensional difference, which is convenient for subsequent time series consistency testing.

[0088] The time series consistency test ensures the stability and accuracy of the model in different time periods by performing a multi-dimensional evaluation of the feature interaction matrix. The time series consistency test of the feature interaction matrix to obtain the time series interaction fusion evaluation results specifically includes the following steps:

[0089] Step 2331: The consistency score of the long-term trend measures the prediction ability of the model on the long-term trend. The difference between the initial warning model prediction value and the actual value at the corresponding time of the fragment information is calculated, and the difference value of all fragment information is averaged to obtain the consistency score of the long-term trend. The higher the score, the stronger the prediction ability of the model on the long-term trend.

[0090] Step 2332: The short-term fluctuation consistency score measures the model's ability to predict short-term fluctuations. The consistency score of short-term fluctuations is obtained by calculating the fluctuation range between the initial warning model's predicted value and the actual value at the corresponding time of adjacent fragment information and averaging all fluctuation ranges. The higher the score, the stronger the model's ability to predict short-term fluctuations.

[0091] Step 2333: The interaction effect capture ability score measures the model's ability to capture the interaction effect between multi-dimensional features. The interaction effect capture ability score is obtained by calculating the deviation between the Spearman rank correlation coefficient of the initial warning model at the time corresponding to the fragment information and the actual Spearman rank correlation coefficient, and averaging all deviations. The higher the score, the stronger the model's ability to capture the interaction effect.

[0092] Step 2334: The interaction effect stability score measures the stability of the model for the interaction effect in different time periods. The interaction effect stability score is obtained by calculating the change rate between the predicted value and the actual value of the initial warning model at the corresponding time of the adjacent fragment information and averaging all the change rates. The higher the score, the better the stability of the model in different time periods.

[0093] Step 2335: If the long-term trend consistency score, short-term volatility consistency score, interaction effect capture ability score and interaction effect stability score all meet the preset qualified threshold, the time series interaction fusion evaluation result is considered qualified, and the initial warning model can be used as the final financial data warning model. If any evaluation indicator does not meet the qualified threshold, the time series interaction fusion evaluation result is considered unqualified, and the initial warning model needs to be optimized. The main means of optimization include adjusting the number of cross layers and the number of neurons in each layer, adjusting the number of layers in the depth layer and the number of neurons in each layer, trying different activation functions, adjusting the loss function, introducing regularization terms, etc.

[0094] The time series interaction fusion evaluation method proposed in this solution ensures the robustness and generalization ability of the model by comprehensively evaluating the long-term trend consistency, short-term fluctuation consistency, interaction effect capture ability and interaction effect stability. This evaluation method not only considers the prediction accuracy of the model, but also considers the model's adaptability to time series data and its ability to capture interaction effects, thus improving the overall performance of the model.

[0095] The early warning information is obtained by inputting the existing structured data of the current survey and design business into the financial data early warning model. If the deviation between the monitoring array predicted by the financial data early warning model and the ideal structured data is greater than the safety threshold, the early warning information is the important parameters with risks in the monitoring array and the predicted risk time. Otherwise, the early warning information is the low risk of financial information. According to the actual situation and risk tolerance of the enterprise, a safety threshold is set in advance. This threshold is used to measure whether the deviation between the monitoring array and the ideal structured data is within an acceptable range. The setting of the safety threshold can be adjusted based on the experience of historical data, industry standards or expert opinions. Based on the interaction effect of historical data and current data, the financial data early warning model can not only identify the important parameters with risks, but also predict the time point when the risk may occur. Based on the output of the model, the system will provide risk warnings for the next quarter to help enterprises prepare for response in advance.

[0096] By comparing the changes in the early warning information of the financial data early warning model, the analysis conclusions include: if the early warning information changes to low risk of financial information, the analysis conclusion is high risk of financial data, otherwise the analysis conclusion is unclear financial risk. If the early warning information of the financial data early warning model does not change, or changes from one risk state to another (for example, from "direct cost fluctuations" to "delayed contract payments"), it indicates that the company's financial situation is still uncertain and no clear conclusion can be drawn, requiring manual intervention and investigation.

[0097] A financial data analysis system for survey and design business, including the following modules:

[0098] The data collection and preprocessing module is used to obtain the structured data of historical and current survey and design business, and clean, standardize and normalize the data; specifically, the data collection and preprocessing module obtains historical and current structured data from sources such as the company's internal database and ERP system, removes duplicate, erroneous or incomplete records, ensures data quality, and extracts important parameters that affect financial fluctuations (direct cost fluctuations, contract payment delays, project schedule deviations, revenue drops and customer churn rates) based on the dynamic attenuation weighted frequency adjustment method, and uses them as monitoring arrays, and normalizes the extracted monitoring arrays to ensure that the dimensions of different parameters are consistent.

[0099] The financial data early warning model construction module, based on the cross neural network, uses the monitoring array to build the financial data early warning model, and performs a time series interactive fusion evaluation on it; specifically, the financial data early warning model construction module cleans the historical structured data and performs feature standardization processing on the monitoring array; based on the input layer, cross layer, depth layer and output layer architecture of the cross neural network, the monitoring information is used as input to obtain the initial early warning model; the initial early warning model is evaluated for time series interactive fusion, and the long-term trend consistency score, short-term volatility consistency score, interaction effect capture ability score and interaction effect stability score are calculated; if the evaluation result is qualified, the initial early warning model is the final financial data early warning model; otherwise, the model is optimized.

[0100] The early warning information generation module uses the financial data early warning model to analyze the existing structured data of the current survey and design business to generate early warning information; specifically, the early warning information generation module inputs the current monitoring array into the financial data early warning model to obtain the prediction results within each time window; based on the prediction results of the model, early warning information is generated, including the warning level and the reason for the warning; if the degree of deviation of important parameters in the early warning information from the ideal structured data is greater than the safety threshold, the early warning information is "important parameters with risks and predicted risk time"; otherwise, the early warning information is "low risk of financial information".

[0101] The ideal structured data comparison module is used to obtain ideal structured data and compare it with the warning information of the current monitoring array to generate analysis conclusions. Specifically, the ideal structured data comparison module obtains ideal structured data and extracts structured data under ideal conditions from the company's historical data as a benchmark, compares the warning information of the current monitoring array with the ideal structured data, and generates analysis conclusions based on the comparison results; if the warning information of the current monitoring array becomes "low risk of financial information", the analysis conclusion is "high risk of financial data"; otherwise, the analysis conclusion is "unclear financial risk".

[0102] The above description is only a preferred specific implementation manner of the present application, and is not intended to limit the protection scope of the present application. Therefore, any equivalent changes made based on the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A financial data analysis method for survey and design business, characterized in that: The following steps are involved: Step 1: Obtain structured data of historical survey and design business, extract important parameters that affect financial fluctuations from the structured data, and use the important parameters as monitoring arrays of the structured data; Step 2: Based on the cross neural network, use the monitoring array to build a financial data early warning model; Step 3: Use the financial data early warning model to analyze the existing structured data and ideal structured data of the current survey and design business, and obtain the early warning information of the existing monitoring array in the current survey and design business; Step 4: Use ideal data to replace important parameters in the warning information, and compare the warning information of the financial data warning model to obtain analytical conclusions.

2. A financial data analysis method for survey and design business according to claim 1, characterized in that: The extracting of important parameters affecting financial fluctuations from structured data specifically includes the following steps: Step 11: weighting the structured data using a dynamic attenuation weighted frequency adjustment method, and normalizing the weight values ​​of the structured data; Step 12: Sort the normalized weight values ​​of the structured data, and take the top five structured data with the highest weight as important parameters; Among them, the dynamic attenuation weighted frequency adjustment method is calculated by the formula It indicates that, W i is the comprehensive weight value of the i-th factor; X i,j is the number of occurrences of the i-th factor in the abnormal data at the j-th stage; Y j is the total number of abnormal data at the jth order; β j is the importance coefficient of the jth order, reflecting the degree of influence of this stage on the enterprise's property; α is the time decay factor, which controls the decay rate of data over time; t i is the time interval between the data of the i-th factor and the current time, in quarters; The monitoring arrays are direct cost fluctuation data, contract payment delay data, project schedule deviation data, revenue drop data and customer churn rate data.

3. A financial data analysis method for survey and design business according to claim 2, characterized in that: The method of constructing a financial data early warning model based on a cross neural network using a monitoring array specifically includes the following steps: Step 21: Clean the historical structured data and perform feature standardization on the cleaned monitoring array to obtain monitoring information; Step 22: Based on the input layer, cross layer, depth layer and output layer architecture of the cross neural network, the monitoring information is used as input to obtain an initial early warning model; Step 23: Perform a time series interactive fusion evaluation on the initial early warning model. If the time series interactive fusion evaluation is qualified, the initial early warning model is a financial data early warning model, otherwise the initial early warning model is optimized.

4. A financial data analysis method for survey and design business according to claim 3, characterized in that: The time series interactive fusion evaluation of the initial warning model specifically includes the following steps: Step 231: Filter historical structured data with a project cycle greater than 4 quarters and containing a monitoring array, and execute step 21 for the filtered historical structured data; Step 232: construct a feature interaction matrix using the monitoring information obtained in step 231; Step 233: Perform a time series consistency test on the feature interaction matrix to obtain a time series interaction fusion evaluation result.

5. A financial data analysis method for survey and design business according to claim 4, characterized in that: The construction of the feature interaction matrix specifically includes the following steps: Step 2321: Divide the monitoring information obtained in step 231 into time windows to obtain a number of overlapping fragment information; Step 2322: Calculate the Spearman rank correlation coefficients between the five important parameters in the segment information, and use the Spearman rank correlation coefficients to construct an initial interaction matrix; Step 2323: introducing a time decay factor to weight the initial interaction matrix, and performing weighted average update and normalization processing on the weighted initial interaction matrix to obtain a feature interaction matrix; Among them, the window period is monthly and the overlapping length is not less than one week.

6. A financial data analysis method for survey and design business according to claim 5, characterized in that: The time series consistency test of the feature interaction matrix to obtain the time series interaction fusion evaluation result specifically includes the following steps: Step 2331: Calculate the difference between the predicted value of the initial warning model and the actual value at the time corresponding to the fragment information, and average the difference values ​​of all fragment information, and use the average value as the consistency score of the long-term trend; Step 2332: Calculate the fluctuation range between the predicted value and the actual value of the initial warning model at the corresponding time of the adjacent fragment information, and average all the fluctuation ranges, and use the average range value as the short-term fluctuation consistency score; Step 2333: Calculate the deviation between the Spearman rank correlation coefficient of the initial warning model at the time corresponding to the fragment information and the actual Spearman rank correlation coefficient, and average all the deviation values, and use the average value as the interaction effect capture ability score; Step 2334: Calculate the change rate between the predicted value and the actual value of the initial warning model at the corresponding time of the adjacent fragment information, and average all the change rates, and use the average change rate as the interaction effect stability score; Step 2335: Determine whether the long-term trend consistency score, short-term volatility consistency score, interaction effect capture ability score and interaction effect stability score all meet the model qualification threshold. If so, the time series interaction fusion evaluation result is qualified, otherwise it is unqualified.

7. A financial data analysis method for survey and design business according to claim 6, characterized in that: The warning information is obtained by inputting the existing structured data of the current survey and design business into the financial data warning model. If the deviation between the monitoring array predicted by the financial data warning model and the ideal structured data is greater than the safety threshold, the warning information is the important parameters of the risk in the monitoring array and the predicted risk time. Otherwise, the warning information is a low risk of financial information.

8. A financial data analysis method for survey and design business according to claim 7, characterized in that: The analysis conclusion obtained by comparing the changes in the warning information of the financial data warning model specifically includes: if the warning information changes to low risk of financial information, the analysis conclusion is high risk of financial data, otherwise the analysis conclusion is unclear financial risk.

9. A financial data analysis system for survey and design business, characterized in that: Includes the following modules: Data collection and preprocessing module, used to obtain structured data of historical and current survey and design business, and clean, standardize and normalize the data; The financial data early warning model construction module is based on a cross neural network and uses a monitoring array to build a financial data early warning model and conduct a time series interactive fusion evaluation on it; The early warning information generation module uses the financial data early warning model to analyze the existing structured data of the current survey and design business and generate early warning information; The ideal structured data comparison module is used to obtain the ideal structured data and compare it with the warning information of the current monitoring array to generate analysis conclusions.

Images