Financial data anomaly analysis system and method based on artificial intelligence

Abstract

The invention discloses a financial data anomaly analysis system and method based on artificial intelligence, and relates to the technical field of financial data. The method comprises the following steps: collecting and cleaning various financial data in real time through docking with an enterprise financial system; processing data extraction features; analyzing the feature data by using the trained deep neural network model, marking abnormal data, and setting a reasonable index threshold to further confirm an abnormal condition; deeply analyzing the financial items marked as abnormal by using an optimization algorithm, and obtaining financial optimization target data to generate specific adjustment suggestions; and financial data are collected again for monitoring again and abnormal type analysis, and a project set is constructed to calculate an association index to find potential problems in the business process, and an early warning report is generated. According to the method, automation and intelligentization of financial data analysis are realized, the detection precision and efficiency are improved, optimization suggestions with strong pertinence are provided, and enterprises are helped to deal with financial risks in time.

Description

Technical Field

[0001] The present invention belongs to the technical field of financial data, and particularly relates to a financial data anomaly analysis system and method based on artificial intelligence. Background Art

[0002] In enterprise operation, financial management is an important means to ensure the effective allocation of resources, cost control and profit maximization. Accurate financial analysis can help enterprises identify potential risks and make strategic decisions. With the expansion of enterprise scale and the diversification of business, the amount of financial data generated has increased exponentially, including data in multiple dimensions such as revenue, cost, profit, and cash flow. These data are not only huge in quantity but also complex in structure, increasing the difficulty of analysis. Traditional financial analysis methods usually rely on manual experience judgment or simple statistical models, which are difficult to process massive data and extract valuable information from it. In addition, these methods have poor adaptability to non-linear and dynamically changing data and are prone to missing important information.

[0003] In recent years, significant progress has been made in deep learning technology, especially showing powerful capabilities in fields such as image recognition and natural language processing. These technologies provide new ideas and tools for solving complex financial data analysis problems. With the increasing requirements of enterprises for real-time monitoring and rapid response to market changes, how to achieve automatic collection, cleaning, feature extraction, and anomaly detection of financial data has become a research hotspot. Using AI technology can greatly improve the analysis efficiency and accuracy. The deficiencies of current solutions of some existing financial anomaly detection systems include lack of comprehensiveness, inflexible setting of static thresholds, and inaccurate optimization suggestions.

[0004] Lack of comprehensiveness: Existing financial anomaly detection systems often only focus on a single dimension of financial data, such as revenue or cost, while ignoring data in other related dimensions, such as profit and cash flow. This one-sidedness leads to insufficient comprehensiveness in anomaly detection and is prone to missing important information.

[0005] Inflexible setting of static thresholds: Existing financial anomaly detection systems usually rely on fixed thresholds to determine whether data is abnormal. However, these thresholds often cannot adapt to changes in the enterprise operation environment and market conditions, resulting in false alarms or missed reports.

[0006] Inaccurate optimization suggestions: The optimization suggestions provided by existing financial anomaly detection systems are often based on fixed rules or templates, lacking pertinence and operability. These suggestions often cannot be directly applied to the actual situation of enterprises, resulting in poor implementation effects.

[0007] The present invention aims to establish a more intelligent, comprehensive, and dynamic financial data anomaly analysis framework through advanced AI technology. It can not only automatically identify various types of financial anomalies, but also deeply analyze the causes of anomalies and give specific optimization suggestions to help enterprises better manage and optimize their financial conditions and enhance their competitiveness. At the same time, by constructing project sets and calculating correlation indicators, the problem points in specific business processes can be more accurately located, so as to guide enterprises to take targeted measures for improvement. Summary of the Invention

[0008] In order to overcome the above-mentioned disadvantages and deficiencies of the prior art, the first object of the present invention is to provide a financial data anomaly analysis system based on artificial intelligence; the second object of the present invention is to provide a financial data anomaly analysis system based on artificial intelligence.

[0009] The first object of the present invention adopts the following technical solution:

[0010] A financial data anomaly analysis system based on artificial intelligence, including

[0011] A data collection and processing module: collecting financial data, calculating and analyzing to obtain the revenue growth rate, cost ratio, net cash flow change rate, and profit fluctuation coefficient;

[0012] An abnormal data judgment module: forming analysis data through numbers and characteristic data, inputting it into a trained abnormal judgment model to obtain results, and at the same time comparing the actual indicators with the preset thresholds, and marking the data outside the range as abnormal data;

[0013] An abnormal judgment model training module: setting judgment results and digital labels for the analysis data, converting them into feature vectors, and training a deep neural network model with the goal of minimizing the prediction error sum until the error converges;

[0014] A financial optimization target data acquisition module: analyzing the financial items corresponding to the abnormal data, constructing a selection label set for each item with a reasonable range preset, and calculating to obtain the financial optimization target data;

[0015] An abnormal data processing suggestion module: generating adjustment suggestions for abnormal financial items based on the financial optimization target data;

[0016] A re-monitoring and analysis module: re-collecting data, extracting features, and using them for the abnormal judgment model to judge. If there are anomalies, analyze the types of anomalies;

[0017] A project correlation analysis module: forming a set of upstream and downstream financial items, calculating correlation indicators and comparing them with the preset range, marking the abnormal set exceeding the threshold, and generating a warning report.

[0018] The second object of the present invention adopts the following technical solution:

[0019] The financial data anomaly analysis method based on artificial intelligence is used to realize the financial data anomaly analysis system based on artificial intelligence. The method flow is as follows:

[0020] Step 1: Connect with the enterprise financial system and use the data interface to collect various financial data in real time. After collection, clean the data and remove data records with duplicates, errors or missing values.

[0021] Step 2: Process the collected financial data and extract feature data, which include revenue growth rate, cost ratio, profit fluctuation coefficient, and net cash flow change rate;

[0022] Step 3: Analyze the extracted feature data to determine whether to mark some of the feature data as abnormal data. Use the financial data number and feature data as a set of analysis data, input the trained abnormality judgment model to obtain the judgment result, and compare the actual index with the preset reasonable index threshold to mark the abnormal data.

[0023] Step 4: Conduct in-depth analysis on the financial items corresponding to the abnormal data to obtain the financial optimization target data, which is obtained by setting a reasonable range, constructing a selection label set, and iteratively calculating using the antibody group algorithm;

[0024] Step 5: Based on the obtained financial optimization target data, give corresponding adjustment suggestions for abnormal data of different financial items;

[0025] Step 6: Recollect financial data and extract feature data, and use the anomaly judgment model again to determine whether there is abnormal data. If so, perform an anomaly type analysis;

[0026] Step 7: Group the financial projects with upstream and downstream relationships into project sets, calculate the correlation index for each project set, and compare the calculated correlation index with the preset reasonable range. If it exceeds the set threshold, mark the project set as an abnormal set and generate an early warning report.

[0027] Preferably, the specific method of extracting characteristic data is: calculating the year-on-year and month-on-month growth rates of revenue data as the revenue growth rate; calculating the proportion of cost data in total cost to obtain the cost proportion; analyzing the fluctuation of profit data in multiple accounting periods to obtain the profit fluctuation coefficient; calculating the ratio of the change in net cash flow to the previous period to determine the net cash flow change rate.

[0028] Preferably, the training process of the anomaly judgment model is as follows: First, set corresponding judgment results for a group of analysis data, where a is an integer greater than 1. Set different numerical tags for different judgment results. Convert the analysis data and the corresponding judgment tags into a corresponding set of feature vectors. Use a set of predicted judgment tags corresponding to each group of analysis data as the output, use the actual judgment tags corresponding to each group of analysis data as the prediction target, and use minimizing the sum of prediction errors of all analysis data as the training target to train the deep neural network model until the sum of prediction errors converges and then stop training.

[0029] Preferably, the process of using the antibody swarm algorithm for iterative calculation to obtain the financial optimization target data includes: setting a reasonable range for each financial item, randomly selecting values from the reasonable ranges of each financial item, combining them into selection data and constructing a selection label set; marking the financial items containing abnormal data as abnormal items, randomly selecting unmarked abnormal items and marking them as selected items, and randomly generating an initial antibody swarm of size q and presetting an iteration threshold based on the selection label set of the selected items; determining the fitness function, calculating the fitness of each antibody in the initial antibody swarm, selecting r antibodies with the largest fitness to construct a memory population, cloning the antibodies in the memory population according to the cloning scale, mutating the antibodies in the cloned antibody swarm using a cloud adaptive mutation operator, randomly selecting w antibodies from the cloned antibody swarm as parent antibodies for recombination to generate offspring antibodies S c , combining the mutated antibody swarm and the recombined antibody swarm to obtain a combined antibody swarm, randomly generating n new antibodies to replace the n antibodies with the smallest fitness in the combined antibody swarm. If the number of loop times t is less than the iteration threshold, repeat the relevant steps. If the number of iteration times t is equal to the iteration threshold, obtain the selection label corresponding to the antibody with the largest fitness in the combined antibody swarm, and obtain the selection parameters corresponding to the screened business links according to the selection label as the financial optimization target data.

[0030] Preferably, the fitness function is:

[0031] fi = ω1×(NHi)+ω2×(WRi);

[0032] Where, fi is the fitness of the i-th antibody, NHi is the financial risk index corresponding to the i-th antibody, WRi is the financial health deviation index corresponding to the i-th antibody, ω1 and ω2 are preset proportionality coefficients, and i ∈ [1, q].

[0033] Preferably, the calculation formula for the cloning scale is:

[0034]

[0035] Φ i = min(exp(||X i -X j ||));

[0036] where Ri is the cloning scale of the i-th antibody, Int is the ceiling function, q is the antibody population scale, Φ i is the affinity of the i-th antibody, min is the minimum value function, exp is the natural exponential function, ||X i -X j || is the Euclidean distance between the i-th antibody and the j-th antibody, i≠j, j∈[1,q].

[0037] Preferably, the formula for calculating the offspring antibody is:

[0038]

[0039] where X w ' is the w-th parent antibody, S w is the w-th scaling factor, the scaling factor is a randomly generated real number and not all of the w scaling factors are 0.

[0040] Preferably, the analysis of abnormal types includes: if the abnormal data is that the revenue growth rate is extremely low, it is judged as a revenue growth dilemma; if the cost ratio is too high, it is judged as ineffective cost control; if the profit fluctuation coefficient is too large, it is judged as a profit stability risk; if the change rate of the net cash flow is abnormal, it is judged as a cash flow risk.

[0041] In summary, due to the adoption of the above technical solutions, the beneficial effects of the present invention are:

[0042] 1. By using a deep neural network model to analyze the feature data, the present invention can more accurately identify potential abnormal situations. Compared with the traditional static threshold judgment method, this AI-based method can adapt to the changes in complex business environments and improve the accuracy of anomaly detection; it not only focuses on the changes in individual financial indicators (such as revenue growth rate, cost ratio, etc.), but also considers the relevance between different financial items, and discovers possible deep-seated business process problems by constructing a project set to calculate correlation indicators. This method makes the risk assessment more comprehensive and dynamic, and helps enterprises adjust their strategies in a timely manner to cope with market changes;

[0043] 2. By obtaining the optimal financial optimization target data, the present invention provides specific adjustment suggestions for different types of anomalies. For example, for the situation of a high cost ratio, it is recommended to optimize the procurement channels or renegotiate prices; for the situation of a low revenue growth rate, strategies such as expanding market channels or launching new products are proposed. These suggestions are based on specific data analysis results, are highly targeted, and can help enterprises effectively solve problems and improve their financial conditions;

[0044] 3. When dealing with abnormal data, the present invention adopts a variety of complex optimization algorithms such as the clonal selection algorithm and the cloud adaptive mutation operator to find the best solution. The application of these technologies not only improves the search ability and convergence speed of the algorithm, but also effectively solves the optimization problems under nonlinear and complex constraint conditions that are difficult to handle by traditional methods. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0046] Figure 1 FIG. shows a block diagram of the financial data anomaly analysis system based on artificial intelligence of the present invention;

[0047] Figure 2 FIG. shows a flowchart of the financial data anomaly analysis method based on artificial intelligence of the present invention;

[0048] Figure 3 FIG. shows a flowchart of the re-monitoring and anomaly type analysis of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0050] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more exemplary embodiments. In the following description, many specific details are provided to give a full understanding of the exemplary embodiments of the present disclosure. However, those skilled in the art will realize that the technical solutions of the present disclosure can be practiced without one or more of the specific details, or other methods, components, steps, etc. can be adopted. In other cases, well-known structures, methods, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0051] Embodiment 1:

[0052] Referring to Figure 1 as shown, the financial data anomaly analysis system based on artificial intelligence in this embodiment includes

[0053] Data Acquisition and Processing Module: Connects to the enterprise financial system, uses data interfaces to collect various financial data in real time, such as revenue data, cost data, profit data, cash flow data, etc., and cleans the data after collection; processes the collected financial data to extract feature data. Specifically, calculates the year-on-year and month-on-month growth rates of revenue data as the revenue growth rate; calculates the proportion of cost data in the total cost to obtain the cost ratio; analyzes the fluctuations of profit data over multiple accounting periods to obtain the profit fluctuation coefficient; calculates the ratio of the change range of the net cash flow to the previous period to determine the net cash flow change rate.

[0054] Abnormal Data Judgment Module: Analyzes the extracted feature data to determine whether to mark some data in the feature data as abnormal data. By setting numbers for different financial data, takes the financial data number and feature data as a set of analysis data, inputs it into the trained abnormal judgment model, outputs the judgment label and obtains the judgment result. At the same time, compares the actual indicators of each financial item with the preset reasonable indicator thresholds, and marks them as abnormal data if they exceed the range.

[0055] Abnormal Judgment Model Training Module: Presets corresponding judgment results for a certain number of analysis data in advance, and sets digital labels for different judgment results. Converts the analysis data and judgment labels into feature vectors, and trains the deep neural network model with the goal of minimizing the sum of prediction errors until the error converges.

[0056] Financial Optimization Target Data Acquisition Module: Conducts in-depth analysis on the financial items corresponding to the abnormal data to obtain financial optimization target data. Sets a reasonable range for each financial item, constructs a selection label set, and uses the antibody group algorithm (including operations such as generating the initial antibody group, calculating fitness, cloning, mutation, recombination, etc.) for iterative calculation. Finally, obtains the selection label corresponding to the antibody with the maximum fitness in the combined antibody group to get the financial optimization target data.

[0057] Abnormal Data Processing Suggestion Module: Based on the obtained financial optimization target data, gives corresponding adjustment suggestions for the abnormal data of different financial items. For example, provides specific optimization suggestions for situations such as an abnormally high cost ratio or an abnormally low revenue growth rate.

[0058] Re-monitoring and Analysis Module: Re-collects financial data and extracts feature data, and uses the abnormal judgment model again to determine whether there is abnormal data. If there is abnormal data, conducts abnormal type analysis, such as judging as revenue growth dilemma, ineffective cost control, profit stability risk, cash flow risk, etc.

[0059] Project Association Analysis Module: It forms project sets with financial projects in an upstream-downstream relationship, calculates association indicators for each project set, and compares the calculated association indicators with a preset reasonable range. If the threshold is exceeded, the project set is marked as an abnormal set, and a warning report is generated indicating possible business process problems.

[0060] The beneficial effects of this embodiment are as follows: efficiently collecting and processing financial data, accurately judging abnormal data, and improving the judgment accuracy through a trained model; deeply analyzing abnormal data and providing financial optimization suggestions; continuously monitoring and analyzing to detect risks in a timely manner; project association analysis and warning to assist enterprises in optimizing business processes and improving financial management levels.

[0061] Embodiment 2:

[0062] Refer to Figure 2 As shown, the method for analyzing abnormal financial data based on artificial intelligence in this embodiment is as follows:

[0063] Step 1: Connect to the enterprise financial system and use the data interface to collect various types of data in the system in real time, including but not limited to various financial data such as revenue data, cost data, profit data, and cash flow data. After collection, clean the data to remove data records with a large number of duplicates, errors, or missing values.

[0064] Step 2: Process the collected financial data to extract feature data. The feature data includes revenue growth rate, cost ratio, profit fluctuation coefficient, cash flow net change rate, etc.

[0065] Calculate the year-on-year and month-on-month growth rates of revenue data as the revenue growth rate.

[0066] Calculate the proportion of cost data in the total cost to obtain the cost ratio.

[0067] Analyze the fluctuation of profit data over multiple accounting periods to obtain the profit fluctuation coefficient.

[0068] Calculate the ratio of the change amplitude of the net cash flow to the previous period to determine the net cash flow change rate.

[0069] Step 3: Analyze the extracted feature data to determine whether to mark some of the data in the feature data as abnormal data.

[0070] Set different numbers for different financial data, use the financial data number and the feature data as a set of analysis data, input each set of analysis data into the trained abnormal judgment model respectively, output the corresponding judgment label, and obtain the corresponding judgment result according to the judgment label. The judgment results include revenue anomaly, cost anomaly, profit anomaly, cash flow anomaly, all normal, and all abnormal.

[0071] The training process of the anomaly judgment model includes: pre-setting corresponding judgment results for a group of analysis data, where a is an integer greater than 1, and setting different numerical labels for different judgment results; marking the numerical labels of the judgment results as judgment labels, and converting the analysis data and the corresponding judgment labels into a corresponding set of feature vectors; using each set of feature vectors as the input of the anomaly judgment model, the anomaly judgment model outputs a set of predicted judgment labels corresponding to each set of analysis data, and uses the actual judgment label corresponding to each set of analysis data as the prediction target, and the actual judgment label is the numerical label of the judgment result collected in advance corresponding to the analysis data; taking the minimization of the sum of the prediction errors of all analysis data as the training target; training the anomaly judgment model until the sum of the prediction errors reaches convergence and then stopping the training; the anomaly judgment model is a deep neural network model.

[0072] If the judgment result is abnormal revenue, then mark the revenue growth rate as abnormal data;

[0073] If the judgment result is abnormal cost, then mark the cost ratio as abnormal data;

[0074] If the judgment result is abnormal profit, then mark the profit volatility coefficient as abnormal data;

[0075] If the judgment result is abnormal cash flow, then mark the change rate of net cash flow as abnormal data;

[0076] If the judgment result is all abnormal, then mark the above-mentioned multiple feature data as abnormal;

[0077] If the judgment result is all normal, then do not mark the feature data.

[0078] At the same time, preset corresponding reasonable index thresholds for each financial item, including reasonable revenue growth rate thresholds, cost ratio thresholds, etc., and compare the actual indexes of each financial item with the corresponding thresholds respectively; if the actual index exceeds the reasonable range, then mark the index as abnormal data.

[0079] Step Four: Conduct in-depth analysis on the financial items corresponding to the abnormal data to obtain the corresponding financial optimization target data.

[0080] Set a reasonable range for each financial item, such as a reasonable range for the revenue growth rate, a reasonable range for the cost ratio, etc. Randomly select values from within the reasonable range of each financial item to form selection data, and set unique numerical labels for each set of selection data to construct a selection label set.

[0081] Mark the financial items containing abnormal data as abnormal items, and cancel the abnormal marks for items with only abnormal profit fluctuation coefficients but stable overall profitability. Randomly select an abnormal item that has not been marked as selected and mark it as a selected item. According to the set of selection labels of the selected item, randomly generate an initial antibody population X(0) = {x1, x2, x3,..., xn} with a size of q; at the same time, preset the iteration threshold.

[0082] Determine the fitness function:

[0083] fi = ω1×(NHi) + ω2×(WRi);

[0084] Where, fi is the fitness of the i-th antibody, NHi is the financial risk index corresponding to the i-th antibody (such as debt risk index, etc.), WRi is the financial health deviation index corresponding to the i-th antibody, ω1 and ω2 are preset proportionality coefficients, and i ∈ [1, q]. By converting the selection labels corresponding to the antibodies into selection data, combining with financial item numbers, etc. as test data, and inputting them into the trained financial risk prediction model and financial health prediction model, the corresponding financial risk index and financial health deviation index are obtained.

[0085] Calculate the fitness of each antibody in the initial antibody population X(0), select r antibodies with the largest fitness, and construct the memory population Xm(t). Clone the antibodies in the memory population Xm(t) according to the cloning scale, and generate the cloned antibody population Xc(t). The calculation formula for the cloning scale is:

[0086]

[0087] Φ i =min(exp(||X i -X j ||));

[0088] Where, Ri is the cloning scale of the i-th antibody, Int is the ceiling function, q is the size of the antibody population, Φ i is the affinity of the i-th antibody, min is the minimum value function, exp is the natural exponential function, ||X i -X j || is the Euclidean distance between the i-th antibody and the j-th antibody, i ≠ j, j ∈ [1, q].

[0089] Mutate the antibodies in the cloned antibody population X c (t) using the cloud adaptive mutation operator to generate the mutated antibody population X d (t). Randomly select w antibodies from the cloned antibody population X c (t) as parent antibodies for recombination to generate the offspring antibody S c , until the cloned antibody population Xc Stop when the number of unselected antibodies in (t) is less than w, and generate a recombinant antibody population X r (t), the offspring antibody S c The calculation formula is:

[0090]

[0091] where X w ' is the w-th parental antibody, and S w is the w-th scaling factor, which is a randomly generated real number and not all w scaling factors are 0.

[0092] Merge the mutant antibody population X d (t) and the recombinant antibody population X r (t) to obtain the merged antibody population X L (t), randomly generate n new antibodies, and replace the n antibodies with the lowest fitness in the merged antibody population X L (t).

[0093] If the number of iterations t is less than the iteration threshold, let t = t + 1, and select the r antibodies with the highest fitness in the merged antibody population X L (t) to form a new memory population X m (t + 1), and repeat the above relevant steps;

[0094] If the number of iterations t is equal to the iteration threshold, obtain the selection label corresponding to the antibody with the highest fitness in the merged antibody population X L (t), obtain the selection parameters corresponding to the selected business process according to the selection label, and use them as the financial optimization target data. The selected business process is the previously marked selected business process;

[0095] Repeat the above steps until all abnormal business processes are marked as selected business processes, and the loop ends.

[0096] Step Five: Abnormal data processing and adjustment suggestions.

[0097] Based on the obtained financial optimization target data, give corresponding adjustment suggestions for the abnormal data of different financial items. If the cost ratio is abnormally high, it can be recommended that the enterprise optimize the procurement channels, renegotiate prices with suppliers, or evaluate and cut unnecessary expenses; if the revenue growth rate is abnormally low, it can be recommended to expand market channels, launch new products, or optimize marketing strategies, etc.

[0098] Step Six: Re-monitoring and abnormal type analysis.

[0099] Refer to Figure 3 As shown, the process of re-monitoring and abnormal type analysis is as follows:

[0100] S61. Re - collect financial data and extract feature data, and then use the anomaly judgment model again to determine whether there is abnormal data.

[0101] S62. If there is still abnormal data, conduct an analysis of the anomaly type:

[0102] If the abnormal data is that the revenue growth rate is extremely low, it is judged as a revenue growth dilemma;

[0103] If the cost ratio is too high, it is judged that cost control is ineffective;

[0104] If the profit fluctuation coefficient is too large, it is judged as a profit stability risk;

[0105] If the change rate of the net cash flow is abnormal, it is judged as a cash flow risk.

[0106] S63. Project set construction: Combine financial items with upstream and downstream relationships into project sets. For example, procurement cost and raw material inventory, sales revenue and accounts receivable, etc.

[0107] S64. Associated index calculation: For each project set, calculate the associated index, such as the ratio change between procurement cost and raw material inventory.

[0108] S65. Threshold comparison and warning: Compare the calculated associated index with the preset reasonable range. If it exceeds the set threshold, mark the project set as an abnormal set and generate a warning report indicating possible business process problems (such as cost - inventory imbalance caused by unreasonable procurement processes).

[0109] The beneficial effects of this embodiment are as follows: Realize real - time collection of financial data and accurate anomaly judgment, improve the judgment accuracy through the deep neural network model; Provide financial optimization suggestions to assist enterprises in optimizing financial management; Continuously monitor and analyze to timely discover and process abnormal data; Construct project sets, warn about business process problems, and improve the operation efficiency of enterprises.

[0110] As described above, it is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and all should be covered within the protection scope of the present invention.

[0111] The preferred embodiments of the present invention disclosed above are only used to help illustrate the present invention. The preferred embodiments do not describe all the details in detail, nor do they limit the invention to only the specific embodiments. Obviously, many modifications and variations can be made according to the content of this specification. These embodiments are selected and specifically described in this specification in order to better explain the principles and practical applications of the present invention, so that those skilled in the art can well understand and utilize the present invention. The present invention is only limited by the claims and their full scope and equivalents.

Claims

1. An abnormal financial data analysis system based on artificial intelligence, characterized in that, The system comprises Collection and processing module: collect financial data to calculate revenue growth rate, cost ratio, net cash flow change rate and profit fluctuation coefficient; Abnormal data judgment module: The module inputs the trained abnormal judgment model to obtain the result through the analysis of the number and feature data composition, and compares the actual index with the preset threshold. If the index exceeds the range, it will be marked as abnormal data. Abnormal judgment model training module: set judgment results and digital labels for analysis data, convert them into feature vectors to minimize prediction errors and train deep neural network models for the target until the error converges; Financial optimization target data acquisition module: Analyze the financial items corresponding to abnormal data, set a reasonable range for each item, build a selection label set, and calculate and obtain financial optimization target data; Abnormal data processing suggestion module: Generate adjustment suggestions for abnormal financial items based on financial optimization target data; Second monitoring and analysis module: re-collect data, extract features, use them for abnormal judgment model judgment, and analyze the abnormal type if there is an abnormality; Project correlation analysis module: It groups the upstream and downstream financial projects into sets, calculates correlation indicators and compares them with the preset range, marks abnormal sets that exceed the threshold, and generates early warning reports.

2. The method for abnormal analysis of financial data based on artificial intelligence is used to implement the system for abnormal analysis of financial data based on artificial intelligence as described in claim 1, and is characterized in that, The method flow is as follows: Step 1: Connect with the enterprise financial system and use the data interface to collect various financial data in real time. After collection, clean the data and remove data records with duplicates, errors or missing values. Step 2: Process the collected financial data to extract feature data, including revenue growth rate, cost ratio, profit fluctuation coefficient, and net cash flow change rate; Step 3: Analyze the extracted feature data to determine whether to mark some of the feature data as abnormal data. Use the financial data number and feature data as a set of analysis data, input the trained abnormality judgment model to obtain the judgment result, and compare the actual index with the preset reasonable index threshold to mark the abnormal data. Step 4: Conduct in-depth analysis on the financial items corresponding to the abnormal data to obtain the financial optimization target data, which is obtained by setting a reasonable range, constructing a selection label set, and iteratively calculating using the antibody group algorithm; Step 5: Based on the obtained financial optimization target data, give corresponding adjustment suggestions for abnormal data of different financial items; Step 6: Recollect financial data and extract feature data, and use the anomaly judgment model again to determine whether there is abnormal data. If so, perform an anomaly type analysis; Step 7: Group the financial projects with upstream and downstream relationships into project sets, calculate the correlation index for each project set, and compare the calculated correlation index with the preset reasonable range. If it exceeds the set threshold, mark the project set as an abnormal set and generate an early warning report.

3. The method for abnormal analysis of financial data based on artificial intelligence according to claim 2, characterized in that, The specific method of extracting characteristic data is: calculating the year-on-year and month-on-month growth rates of revenue data as the revenue growth rate; calculating the proportion of cost data in total cost to obtain the cost proportion; analyzing the fluctuation of profit data in multiple accounting periods to obtain the profit fluctuation coefficient; calculating the ratio of the change in net cash flow to the previous period to determine the net cash flow change rate.

4. The method for abnormal analysis of financial data based on artificial intelligence according to claim 2, wherein The training process of the abnormal judgment model is as follows: setting corresponding judgment results for a group of analysis data in advance, where a is an integer greater than 1, setting different digital labels for different judgment results, converting the analysis data and the corresponding judgment labels into a corresponding set of feature vectors, taking a set of predicted judgment labels corresponding to each group of analysis data as output, taking the actual judgment label corresponding to each group of analysis data as the prediction target, taking minimizing the sum of prediction errors of all analysis data as the training target, training the deep neural network model until the sum of prediction errors reaches convergence and stopping the training.

5. The method for abnormal analysis of financial data based on artificial intelligence according to claim 2, characterized in that, The process of using the antibody swarm algorithm for iterative calculation to obtain financial optimization target data includes: setting a reasonable range for each financial item, randomly selecting values from within the reasonable ranges of each financial item, combining them into selection data and constructing a selection label set; marking the financial items containing abnormal data as abnormal items, randomly selecting abnormal items that have not been marked as selected and marking them as selected items, randomly generating an initial antibody swarm of size q and presetting an iteration threshold based on the selection label set of the selected items; determining a fitness function, calculating the fitness of each antibody in the initial antibody swarm, selecting r antibodies with the largest fitness to construct a memory population, cloning the antibodies in the memory population according to the cloning scale, mutating the antibodies in the cloned antibody swarm using a cloud self-adaptive mutation operator, randomly selecting w antibodies from the cloned antibody swarm as parent antibodies for recombination to generate offspring antibodies S c , combining the mutated antibody swarm and the recombined antibody swarm to obtain a combined antibody swarm, randomly generating n new antibodies to replace the n antibodies with the smallest fitness in the combined antibody swarm, repeating the relevant steps if the number of cycles t is less than the iteration threshold, and if the number of iterations t is equal to the iteration threshold, obtaining the selection label corresponding to the antibody with the largest fitness in the combined antibody swarm, and obtaining the selection parameters corresponding to the screened business processes as the financial optimization target data.

6. The method for abnormal analysis of financial data based on artificial intelligence according to claim 5, characterized in that, The fitness function is: fi=ω1×(NHi)+ω2×(WRi); Among them, fi is the fitness of the i-th antibody, NHi is the financial risk indicator corresponding to the i-th antibody, WRi is the financial health deviation indicator corresponding to the i-th antibody, ω1 and ω2 are preset proportional coefficients, i∈[1,q].

7. The method for abnormal analysis of financial data based on artificial intelligence according to claim 5, characterized in that The clone scale calculation formula is: Φ i = min(exp(||X i - X j ||)); where Ri is the cloning scale of the i-th antibody, Int is the ceiling function, q is the scale of the antibody population, Φ i is the affinity of the i-th antibody, min is the minimum value function, exp is the natural exponential function, ||X i -X j || is the Euclidean distance between the i-th antibody and the j-th antibody, i≠j, j∈[1,q].

8. The method for abnormal analysis of financial data based on artificial intelligence according to claim 5, wherein The calculation formula for the progeny antibody is: Among them, X w ' is the w-th parental antibody, and S w is the w-th scaling factor. The scaling factor is a randomly generated real number and not all of the w scaling factors are 0.

9. The method for abnormal analysis of financial data based on artificial intelligence according to claim 2, wherein The abnormal type analysis includes: if the abnormal data is an abnormally low revenue growth rate, it is judged as a revenue growth dilemma; if the cost ratio is too high, it is judged as poor cost control; if the profit fluctuation coefficient is too large, it is judged as a profit stability risk; if the net cash flow change rate is abnormal, it is judged as a cash flow risk.

Images