Economic index abnormity alarm system based on intelligent data analysis

Abstract

The invention discloses an economic index abnormity alarm system based on intelligent data analysis, and the system comprises a data collection module which is used for collecting economic index related data, an analysis module which is used for analyzing the fluctuation characteristics of economic index data based on a preset algorithm model, and a detection and alarm module. The abnormal alarm module is used for generating abnormal alarm information when detecting that the fluctuation of the index exceeds a preset threshold value; according to the economic index abnormity alarm system based on intelligent data analysis, the detection precision of economic data abnormity is remarkably improved, high stability and flexibility can be kept and the robustness of the system is enhanced when the system deals with a complex and changeable economic environment, so that the system can still keep high accuracy when facing economic data with more noise, and the economic index abnormity alarm system is suitable for popularization and application. And the calculation cost and the energy consumption during the operation of the system are obviously reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis and early warning, and in particular to an economic indicator abnormality alarm system based on intelligent data analysis. Background Art

[0002] In the modern economic management of international trade, with the development of big data technology, more and more decisions rely on real-time analysis and anomaly detection of massive economic data. At present, through the integration of modern big data technology, machine learning algorithms and professional knowledge, real-time or regular automatic analysis of macroeconomic data has been realized, and whether there is a significant deviation in the indicators can be determined based on the established model, so as to timely warn of potential economic risks.

[0003] Existing economic indicator anomaly alarms usually include the following steps: First, the system collects and processes massive economic data, and then applies statistical methods such as Z-score or machine learning algorithms such as isolation forest algorithm to detect anomalies in the data. In this way, the system can identify fluctuations in economic indicators that deviate from normal trends and trigger a preset early warning mechanism when anomalies are detected. This process can help policymakers quickly pay attention to potential risks or emergencies and make corresponding decision adjustments. However, there are some significant challenges and limitations in the application of existing technologies. First, accurately identifying abnormal changes in economic indicators in large amounts of complex economic data is still a daunting task. Economic data has high-dimensional characteristics and often contains noise. How to avoid false positives (false positives) and false negatives while improving the algorithm's ability to mine subtle patterns has become a key difficulty. Second, due to the complexity and dynamics of the economic system, a single detection algorithm or model is difficult to adapt to changes in all situations. Therefore, the existence of these problems limits the application effect and decision support capabilities of such systems in complex economic environments. Summary of the invention

[0004] The purpose of the present invention is to provide an economic indicator abnormality alarm system based on intelligent data analysis to solve the problem that in the prior art, accurately identifying abnormal changes in economic indicators in large amounts of complex economic data is still a difficult task, and due to the complexity and dynamics of the economic system, a single detection algorithm or model is difficult to adapt to changes in all situations.

[0005] To achieve the above object, the present invention provides the following technical solution: an economic indicator abnormality alarm system based on intelligent data analysis, the system comprising:

[0006] Data collection module, used to collect data related to economic indicators;

[0007] The analysis module is used to analyze the fluctuation characteristics of economic indicator data based on a preset algorithm model;

[0008] A detection and alarm module, used to generate abnormal alarm information when it is detected that the fluctuation of the indicator exceeds a preset threshold;

[0009] The analysis module analyzes the fluctuation characteristics of economic indicator data based on a preset algorithm model, including collecting time series information of economic data, extracting periodicity and trend components in the time series, analyzing the change law of the time series components of economic data, and calculating the standard deviation value of economic data based on the change law;

[0010] The detection and alarm module generates abnormal alarm information when it detects that the fluctuation of the indicator exceeds the preset threshold, including setting standard upper and lower thresholds for measuring abnormalities, triggering an alarm after detecting that the latest reading exceeds the range, and starting the machine learning model to analyze whether the current event is a true abnormality rather than a normal variation in a statistical sense. If it is determined to be an abnormality, an email or text message is sent to relevant personnel, a detailed report is generated and archived simultaneously, and an alarm interface is pushed to the management end for manual inspection when necessary.

[0011] Preferably, the data acquisition module collects economic indicator related data including:

[0012] Receive transaction data from financial exchanges in real time;

[0013] Extract historical financial reports;

[0014] Get information on economic indices and market news;

[0015] Get the latest money supply indicators;

[0016] Crawl macroeconomic data published on the Internet.

[0017] Preferably, the analysis module analyzes the changing rules of the time series components of economic data including:

[0018] The improved Hurst index and autoregressive moving average model are used to jointly analyze the changing rules of time series components and calculate the Hurst index H of each k , the specific formula is:

[0019]

[0020] Among them, H k is the Hurst exponent, R / S is the relative fluctuation range, and n is the number of samples;

[0021] The decomposed trend component r(t) is fitted with an autoregressive moving average model. The specific formula is:

[0022] r(t)=φ1r(t-1)+φ2r(t-2)+…+θ1∈(t-1)+θ2∈(t-2)+…+∈(t),

[0023] where r(t) is the trend component at time t, t is time, □1,□2,…,□i are the autoregressive coefficients in the model, θ1,θ2,…,θ j is the moving average coefficient in the model, and □(t) is the random error term.

[0024] Preferably, the analysis module calculates the standard deviation value of economic data based on the change law, including:

[0025] Statistical historical data sets of economic indicators within a set period of time;

[0026] Determine the ratio of the volatility component in the historical data set to the overall level;

[0027] The standard deviation is calculated based on the proportion and overall level of the volatility components;

[0028] If the standard deviation of economic indicator data in a certain period exceeds the set threshold T, it indicates abnormal fluctuations.

[0029] Preferably, the calculating of the standard deviation based on the proportion of the volatility component and the overall level comprises:

[0030] The proportion of the quantitative volatility component, B, is a parameter in the standard deviation formula;

[0031] Use the overall mean M to adjust the sensitivity of the calculation formula;

[0032] To obtain the standard deviation σ, use the formula:

[0033]

[0034] Where σ is the standard deviation, x i is the i-th data point, M is the average value, B is the coefficient of the volatility component, and n is the number of data points

[0035] The obtained standard deviation result is used to compare with the predefined abnormal fluctuation limit.

[0036] Preferably, comparing the obtained standard deviation result with a predefined upper limit of abnormal fluctuation includes:

[0037] Let the standard deviation result be SD;

[0038] Compare the standard deviation SD value with the previously set upper limit of the safety fluctuation range T;

[0039] If SD>T, the judgment result is 1, which means there is a potential volatility risk;

[0040] If SD≤T, the result is 0, which means it is within the safe range;

[0041] If SD exceeds the upper limit T, it is marked as potential volatility risk.

[0042] Preferably, determining the ratio of the volatility component in the historical data set to the overall level comprises:

[0043] In the historical data set obtained, the volatility component and the overall level are first separated, and the volatility component is represented by calculating the deviation between the economic indicator value at each time point and its trend level, specifically:

[0044] V t =X t -T t ;

[0045] Among them, V t is the volatility component, X t is the value of the economic indicator at time t, T t It is the trend component;

[0046] The ratio of the volatility component to the overall level is expressed as:

[0047] R t =V t / T t ;

[0048] Among them, R t It is the ratio of the volatility component to the overall level.

[0049] Preferably, the detection and alarm module starts the machine learning model to analyze whether the current event is a true anomaly rather than a normal variation in a statistical sense, including:

[0050] Use known abnormal events and normal events in historical data to supervise the model. The loss function during training can use weighted cross entropy to cope with the situation where there are fewer abnormal event samples. Specifically:

[0051]

[0052] Where L is the loss function, m is the total number of training samples, wi is the weight of the i-th sample, yi is the actual label of the i-th sample, is the model prediction value of the ith sample.

[0053] Preferably, the model is predicted, specifically:

[0054] For the current input feature matrix F t Make a prediction and output a probability value P anomaly, indicating the possibility that the current event is abnormal, by setting the dynamic threshold T dynamic To determine whether to trigger the final alarm,

[0055] If P anomaly >T dynamic , it is considered abnormal.

[0056] Preferably, the threshold T dynamic Make dynamic adjustments, specifically:

[0057] Introduce a dynamic threshold adjustment mechanism based on Bayesian optimization to continuously optimize the threshold T through online learning dynamic , to minimize the risk of false positives and false negatives, Bayesian optimization updates the prior distribution of the model through historical detection results and actual feedback, and adjusts the threshold at each detection to balance the accuracy and timeliness of the alarm. The formula is:

[0058]

[0059] Among them, Tdynamic is the dynamic threshold, p(T|data) is the prior distribution updated according to historical data, and Loss(T) is the loss function after the alarm is triggered.

[0060] It can be seen from the above technical solution that the present invention has the following beneficial effects:

[0061] The economic indicator abnormality alarm system based on intelligent data analysis collects economic indicator related data through the data acquisition module, the analysis module analyzes the fluctuation characteristics of the economic indicator data based on a preset algorithm model, and the detection and alarm module generates abnormal alarm information when it detects that the fluctuation of the indicator exceeds the preset threshold. It significantly improves the detection accuracy of economic data anomalies, and can maintain high stability and flexibility when dealing with complex and changeable economic environments. The robustness of the system is enhanced, so that it can still maintain high accuracy when facing economic data with more noise, and significantly reduces the computing cost and energy consumption during system operation. It solves the problem that in the existing technology, accurately identifying abnormal changes in economic indicators in large amounts of complex economic data is still a difficult task, and due to the complexity and dynamics of the economic system, a single detection algorithm or model is difficult to adapt to changes in all situations. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 It is a schematic diagram of the connection of the system modules of the present invention. DETAILED DESCRIPTION

[0063] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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.

[0064] like Figure 1 As shown, an economic indicator abnormality alarm system based on intelligent data analysis, the system includes: a data acquisition module for collecting economic indicator related data, an analysis module for analyzing the fluctuation characteristics of economic indicator data based on a preset algorithm model, and a detection and alarm module for generating abnormal alarm information when the fluctuation of the indicator is detected to exceed a preset threshold. First, we are required to collect economic indicator related data. The source of these data can be official information such as data publicly released by the central bank, statistical departments, etc., or obtained from news sources and various announcements on the Internet through crawlers, or even combined with information provided by third-party commercial data service providers to form a comprehensive economic data set; analyze the fluctuation characteristics of the economic indicator data based on a preset algorithm model; at this stage, a series of data analysis methods will be used to evaluate the normal range and fluctuation range of economic variables. In this part, we can use statistical methods to analyze the past Historical data is analyzed to establish a standard or so-called normal range as a baseline. In addition, more complex machine learning methods may be used to train classifiers to identify patterns and predict significant changes that may occur. In this process, the model training process needs to be continuously optimized and adjusted until a high correct prediction accuracy is obtained. The data used for monitoring is obtained in real time or regularly to ensure that new changes can be responded to quickly. When the algorithm finds that the fluctuation of a certain indicator exceeds the set warning value after comparing it with the standard interval or training model, a warning signal will be generated. For example, if we are paying attention to the growth rate data of GDP, assuming that a stable growth rate model in the range of 2% to 4% is established based on the data of the past years, when the actual GDP growth rate in a certain quarter reaches -1% or 7%, the system's automatic alarm program is triggered and the decision-makers of the relevant departments are reported to report this large deviation from the normal range. The above process effectively alleviates the problems of low efficiency and difficulty in ensuring accuracy of manual identification in the face of massive information by introducing automation technology and advanced algorithms. It enables us to quickly locate changes in importance and take necessary preventive or remedial measures before the problem evolves into a greater harm, which is of great significance to ensuring macroeconomic security and financial stability. This invention provides a technical solution to solve the technical problem that traditional means are difficult to respond to economic fluctuations in a complex environment in a timely and effective manner. After implementation, it can achieve real-time monitoring and early warning of key financial or trade indicators caused by emergencies, which are applied in international trade.

[0065] The analysis module analyzes the fluctuation characteristics of economic indicator data based on a preset algorithm model, including collecting time series information of economic data, extracting periodicity and trend components in the time series, analyzing the changing patterns of the time series components of economic data, and calculating the standard deviation value of economic data based on the changing patterns.

[0066] First, the analysis module receives the time series data of economic indicators provided by the data acquisition module. These data usually record the values ​​of economic indicators at continuous time points, such as monthly GDP growth rate, quarterly inflation rate, etc. The analysis module uses a preset algorithm model to process these time series data to extract the cyclical components (e.g., seasonal fluctuations) and trend components (e.g., long-term growth or decline trends). This extraction usually uses some classic methods, such as Fourier transform to identify cyclical characteristics, or uses moving average method or Holt-Winters model to capture trend components. After successfully separating the cyclical and trend components of the time series, the analysis module further analyzes the change patterns of these components. This can be done by observing the rate of change, amplitude of the components and their impact on the overall time series. For example, the intensification of cyclicality or the obvious deviation of the trend may indicate abnormal economic conditions. Finally, based on these change patterns, the analysis module calculates the standard deviation value of the time series. As a statistic, the standard deviation can reflect the stability and variability of economic data fluctuations. A larger standard deviation may indicate a greater risk of fluctuations.

[0067] Through the above processing steps, the analysis module can accurately separate and analyze the cyclical and trend components in economic data, providing a reliable basis for subsequent risk detection. The calculation of standard deviation further quantifies the fluctuation characteristics of economic data, which helps the system to more accurately identify abnormal fluctuation signals when facing complex economic data, and improve the accuracy and timeliness of early warning.

[0068] In another embodiment, the analysis module can use different algorithm models to extract the periodicity and trend components in the time series. For example, more complex machine learning algorithms such as long short-term memory networks (LSTM) can be used to capture nonlinear changes in time series. The calculation method of the standard deviation can also be adjusted as needed, such as calculating the local standard deviation by sliding windows to better capture the fluctuation characteristics in the short term. In addition, the analysis module can also combine external economic factors or other data sources to more comprehensively analyze the fluctuation patterns of economic indicators, thereby further improving the system's prediction ability and alarm accuracy.

[0069] The detection and alarm module generates abnormal alarm information when it detects that the fluctuation of the indicator exceeds the preset threshold, including setting standard upper and lower thresholds for measuring abnormalities, triggering an alarm after detecting that the latest reading exceeds the range, and starting the machine learning model to analyze whether the current event is a true abnormality rather than a normal variation in a statistical sense. If it is determined to be an abnormality, an email or text message is sent to the relevant personnel, a detailed report is generated and archived simultaneously, and an alarm interface is pushed to the management end for manual inspection when necessary.

[0070] First, the detection and alarm module sets a set of standard upper and lower thresholds, which are set based on historical data and expert experience, with the aim of identifying abnormal fluctuations in economic indicator data. After the latest economic data is processed by the analysis module, the detection and alarm module compares the data with the preset thresholds. If a data point exceeds the set upper and lower limits, the system immediately triggers a preliminary alarm. The triggering of this alarm indicates that there may be abnormal fluctuations, but the system does not directly take it as the final conclusion. To further confirm the anomaly, the system immediately starts a machine learning model for in-depth analysis. The model has been trained with a large amount of historical data and can distinguish between true abnormal events and normal fluctuations in a statistical sense (i.e., normal variation). During the model analysis process, the system considers a variety of factors, such as the amplitude, frequency, and context of the fluctuations, to improve the accuracy of the judgment. If the model ultimately determines that the current event is indeed abnormal, the system will take a series of follow-up measures. First, the system will automatically generate a detailed report containing information such as the time, amplitude, and relevant data analysis results of the abnormal fluctuations, and archive it. Then, the system will send abnormal alarm notifications to pre-set relevant personnel via email or SMS to ensure timely response. In addition, the system will push an alert interface to the management side, which intuitively displays relevant information about the abnormal event and provides an entry for further manual inspection to ensure that the abnormal event is properly handled.

[0071] In this way, the detection and alarm module can not only quickly detect abnormal fluctuations in economic indicators, but also avoid false alarms caused by normal deterioration through auxiliary analysis of machine learning models. This not only improves the accuracy of anomaly detection, but also reduces unnecessary interference and waste of resources. The automatically generated detailed reports and the warning interface pushed to the management end provide sufficient basis and convenience for subsequent manual processing, ensuring comprehensive tracking and management of abnormal events.

[0072] In another embodiment, the detection and alarm module can use different machine learning algorithms for abnormal determination, such as support vector machines, random forests or deep learning models, to adapt to different types of economic data and abnormal characteristics. In addition, the threshold setting can adopt a dynamic adjustment mechanism to continuously optimize the upper and lower limits according to the fluctuation of real-time data, thereby improving the sensitivity and accuracy of the system. For notification methods, in addition to e-mail and text messages, instant messaging tools such as WeChat, Slack, etc. can also be integrated to convey abnormal alarm information more quickly. The warning interface on the management side can also be expanded into a comprehensive management platform, providing more abundant analysis tools and report generation functions for managers to conduct in-depth abnormal event analysis and decision-making.

[0073] The data collection module collects economic indicator related data including:

[0074] Receive transaction data from financial exchanges in real time;

[0075] Extract historical financial reports;

[0076] Get information on economic indices and market news;

[0077] Get the latest money supply indicators;

[0078] Crawl macroeconomic data published on the Internet.

[0079] The data collection module first receives real-time transaction data from financial exchanges, such as stock prices, trading volumes, currency exchange rates, etc. These data can reflect the real-time dynamics of the market and are an important basis for the fluctuation of economic indicators. Next, the module extracts key information from historical financial reports, which include annual or quarterly financial statements of enterprises, providing a longer-term economic data background. At the same time, the data collection module also obtains economic indexes (such as consumer price index CPI, purchasing managers index PMI) and market news information, which can reflect economic trends and market sentiment. The latest money supply indicator data is collected by collecting information released by the central bank to monitor the impact of monetary policy on the economy. Finally, the module further enriches the data source by capturing macroeconomic-related data published on the Internet, such as economic research reports, government announcements, expert analysis, etc. This multi-channel data collection method ensures that the system can obtain comprehensive and timely economic data, providing a solid foundation for subsequent analysis and early warning. By integrating real-time data, historical reports and the latest economic information, the system can better capture the trends and potential risks of economic changes, and improve the system's sensitivity and early warning capabilities.

[0080] In another embodiment, the data collection module can further expand the data source. For example, it can cooperate with more financial information service platforms to obtain more refined data; or integrate social media data analysis modules to mine clues of changes in market sentiment. In addition, the data collection module can also use big data technology to pre-process the acquired data, such as data cleaning, deduplication and formatting, to improve data quality and analysis efficiency. The data capture part can also use more advanced web crawler technology to automatically collect more high-frequency data, especially in application scenarios in the global market, which will make the system more international and diversified.

[0081] The analysis module analyzes the changing patterns of the time series components of economic data, including:

[0082] The improved Hurst index and autoregressive moving average model are used to jointly analyze the changing rules of time series components and calculate the Hurst index H of each k , the specific formula is:

[0083]

[0084] Among them, H k is the Hurst exponent, R / S is the relative fluctuation range, and n is the number of samples;

[0085] The decomposed trend component r(t) is fitted with an autoregressive moving average model. The specific formula is:

[0086] r(t)=φ1r(t-1)+φ2r(t-2)+…+-θ1∈(t-1)+θ2∈(t-2)+…+e∈(t),

[0087] where r(t) is the trend component at time t, t is time, □1,□2,…,□i are the autoregressive coefficients in the model, θ1,θ2,…,θ j is the moving average coefficient in the model, and □(t) is the random error term.

[0088] First, the analysis module processes the data in the time series and determines the long-term dependence of the time series by calculating the Hurst index. The Hurst index Hk reflects the strength of the trend of the time series. Usually, the index value is between 0.5 and 1. The closer the index is to 1, the stronger the trend is. Next, the module separates the trend component r(t) in the time series and uses the ARMA model to fit and analyze it. The ARMA model combines autoregression (AR) and moving average (MA) to effectively capture the linear relationship and short-term fluctuations in the time series. Through this joint analysis method, the system can more accurately identify the changing laws of economic data, especially when dealing with complex economic data, with high prediction ability and analysis accuracy. By combining the Hurst index with the ARMA model, the analysis module can deeply analyze the changing laws of the time series of economic data, so as to more accurately predict future trends. The Hurst index helps to understand the long-term trend of the data, while the ARMA model can handle the short-term fluctuations of the data. The combination of the two makes the analysis results more comprehensive and accurate. This method not only improves the system's ability to identify economic fluctuations, but also provides a more reliable basis for subsequent early warning and decision-making.

[0089] In another embodiment, the Hurst exponent can be replaced with other types of long memory models, such as fractional Brownian motion, to capture more complex time series characteristics. In addition, the ARMA model can also be replaced with more advanced models such as ARI MA (autoregressive integrated moving average model) or generalized autoregressive conditional heteroskedasticity model to handle economic data with higher volatility and complexity. These replacement schemes can further improve the adaptability and analysis effect of the system in different application scenarios.

[0090] The analysis module calculates the standard deviation of economic data based on the law of change, including:

[0091] Statistical historical data sets of economic indicators within a set period of time;

[0092] Determine the ratio of the volatility component in the historical data set to the overall level;

[0093] The standard deviation is calculated based on the proportion and overall level of the volatility components;

[0094] If the standard deviation of economic indicator data in a certain period exceeds the set threshold T, it indicates abnormal fluctuations.

[0095] First, the analysis module extracts the historical data set of economic indicators from the set time window. These data usually contain the values ​​of economic indicators in several periods in the past. Next, the module decomposes these historical data to distinguish the volatility component from the overall level. The volatility component usually reflects the short-term changes of economic indicators, while the overall level represents its long-term trend. By calculating the ratio of the volatility component to the overall level, the system can understand the fluctuation characteristics of economic indicators more accurately. Finally, the module obtains the standard deviation value based on these calculation results. As a statistical measure, the standard deviation reflects the degree of dispersion of data. If the standard deviation exceeds the preset threshold T, the system will determine that there is an abnormal fluctuation in the economic indicator during the period, thereby triggering the early warning mechanism. This method can more accurately identify abnormal fluctuations in economic indicators by quantifying the relationship between the volatility component and the overall level. The calculation of the standard deviation provides an objective measurement standard, which enables the system to effectively filter out possible abnormal situations and reduce false alarms. This method is not only suitable for the detection of short-term fluctuations, but also can provide strong support for the analysis of long-term trends.

[0096] In another embodiment, the analysis module may use different time window lengths to statistically analyze historical data sets to adapt to different economic environments or analysis needs. For example, in a highly volatile market environment, a shorter time window may be more effective, while in a stable economic environment, a longer time window is more appropriate. In addition, the calculation method of the standard deviation can also be adjusted, such as using a weighted standard deviation to better reflect the impact of recent data, or using other statistical indicators such as skewness and kurtosis to supplement the analysis to provide a more comprehensive anomaly detection mechanism.

[0097] The standard deviation is calculated based on the proportion and overall level of the volatility components:

[0098] The proportion of the quantitative volatility component, B, is a parameter in the standard deviation formula;

[0099] Use the overall mean M to adjust the sensitivity of the calculation formula;

[0100] To obtain the standard deviation σ, use the formula:

[0101]

[0102] Where σ is the standard deviation, x i is the i-th data point, M is the average value, B is the coefficient of the volatility component, and n is the number of data points

[0103] The obtained standard deviation result is used to compare with the predefined abnormal fluctuation limit.

[0104] First, the analysis module quantifies the volatility component in the economic data and incorporates its proportion B into the calculation of the standard deviation. The proportion of the volatility component reflects the intensity of changes in economic indicators in a certain period. Next, the system uses the overall average value M of historical data to adjust the sensitivity of the calculation formula, thereby improving the accuracy and stability of the standard deviation calculation. Finally, by obtaining the standard deviation σ, the system can quantify the fluctuation range of economic indicators and compare this result with the predefined upper limit of abnormal fluctuations. If the calculated standard deviation exceeds this upper limit, it means that there is a potential volatility risk. The system will further trigger the corresponding early warning mechanism. By introducing the proportion B of the volatility component as a parameter for the standard deviation calculation, the system can more accurately quantify the volatility of economic data. This method not only improves the sensitivity of the standard deviation calculation, but also enables the system to detect abnormal fluctuations more effectively. By comparing with the predefined upper limit of abnormal fluctuations, the system can identify potential economic risks in a timely manner, thereby providing early warnings.

[0105] In another embodiment, the quantification method of the volatility component can be optimized, for example, by using a weighted average method to better reflect the recent data fluctuations. In addition, the calculation formula of the standard deviation can be further adjusted to adapt to different types of economic data or volatility characteristics. For example, a rolling window standard deviation calculation method can be used to dynamically capture the changing trend of the volatility component, thereby improving the adaptability and detection accuracy of the system in different market environments.

[0106] Use the obtained standard deviation results to compare the predefined abnormal fluctuation limits including:

[0107] Let the standard deviation result be SD;

[0108] Compare the standard deviation SD value with the previously set upper limit of the safety fluctuation range T;

[0109] If SD>T, the judgment result is 1, which means there is a potential volatility risk;

[0110] If SD≤T, the result is 0, which means it is within the safe range;

[0111] If SD exceeds the upper limit T, it is marked as potential volatility risk.

[0112] The analysis module first calculates the standard deviation SD, which reflects the fluctuation range of economic indicators. Then, the system compares SD with the upper limit of the safe fluctuation range T set in advance. The T value is usually set based on historical data and expert judgment, and is intended to serve as a benchmark for judging abnormal fluctuations. When SD exceeds the T value, the system determines that there is a potential risk of fluctuation in the current economic indicators, and marks this status as "1", which means that further attention and processing are required. If SD is less than or equal to T, it is judged as "0", indicating that the current economic situation is within the safe range and no additional measures are required. By comparing the standard deviation results with the predefined threshold, the system can quickly identify the risk of abnormal fluctuations. This not only improves the accuracy of anomaly detection, but also reduces the false alarm rate, making the system more reliable. At the same time, based on clear judgment criteria, the system's early warning process becomes more transparent and explainable, which helps users understand and trust the system's alarm decisions.

[0113] In another embodiment, the system can introduce a dynamic threshold adjustment mechanism to better adapt to the volatility characteristics under different market environments. For example, the T value can be dynamically adjusted according to the volatility of real-time data, so that the system can maintain efficient anomaly detection capabilities in different economic cycles. In addition, a multi-level early warning mechanism can be introduced. When the standard deviation approaches but does not exceed the T value, the system can issue a warning in advance so that users can foresee potential risks and make corresponding preparations.

[0114] Determining the ratio of volatility components in a historical data set to the overall level involves:

[0115] In the historical data set obtained, the volatility component and the overall level are first separated, and the volatility component is represented by calculating the deviation between the economic indicator value at each time point and its trend level, specifically:

[0116] V t =X t -T t ;

[0117] Among them, V t is the volatility component, X t is the value of the economic indicator at time t, T t It is the trend component;

[0118] The ratio of the volatility component to the overall level is expressed as:

[0119] R t =V t / T t ;

[0120] Among them, R t It is the ratio of the volatility component to the overall level.

[0121] The analysis module first extracts the trend component T in the historical data through trend analysis method. t , which reflects the long-term trend of economic indicators. Then, by calculating the actual economic indicator value X at each time point t The trend component T at that time point t The difference between the volatility component V t This volatility component represents the short-term fluctuation range of economic indicators. Next, the system calculates the volatility component V t With the trend component T t The ratio R t This ratio can be used to measure the degree of volatility of economic indicators in a certain period of time, relative to the abnormality of its long-term trend. If the Rt value is high, it indicates that the economic fluctuations in this period are abnormal, and there may be potential economic risks. By calculating the ratio of the volatility component to the overall level, the system can more accurately identify abnormal fluctuations. t The ratio provides a relative indicator that enables the system to accurately judge the severity of fluctuations in different economic environments. This method not only improves the sensitivity of fluctuation detection, but also reduces false alarms caused by normal market fluctuations.

[0122] In another embodiment, the calculation method of the volatility component can introduce smoothing technology, such as using exponential smoothing to reduce the impact of short-term fluctuations, so as to more accurately capture long-term trends. In addition, the system can also combine other volatility indicators, such as relative standard deviation or coefficient of variation, to provide richer volatility analysis information to enhance the overall risk detection capability.

[0123] The detection and alarm module starts the machine learning model to analyze whether the current event is a true anomaly rather than a normal variation in a statistical sense, including:

[0124] Use known abnormal events and normal events in historical data to supervise the model. The loss function during training can use weighted cross entropy to cope with the situation where there are fewer abnormal event samples. Specifically:

[0125]

[0126] Where L is the loss function, m is the total number of training samples, wi is the weight of the i-th sample, yi is the actual label of the i-th sample, is the model prediction value of the ith sample.

[0127] The detection and alarm module first constructs a training data set by collecting and annotating abnormal events and normal events in historical data. Since abnormal events are usually rare and the identification of abnormal events has a greater impact on model performance, a weighted cross entropy loss function is used in model training. This loss function assigns a higher weight w to abnormal event samples. i , so that the model can pay more attention to the correct identification of abnormal events, thereby improving the accuracy of anomaly detection. During the training process, the model continuously adjusts its parameters to minimize the value of the weighted cross entropy loss function, and finally obtains an effective model that can distinguish normal fluctuations from true abnormal events. By using the weighted cross entropy loss function, the system can effectively solve the problem of fewer abnormal event samples and improve the model's sensitivity and accuracy to abnormal events. This training method ensures that the system can reduce the risk of underreporting when detecting abnormal events, and can still maintain a high early warning effect when facing complex economic data.

[0128] In another embodiment, model training can use other techniques adapted to unbalanced data sets, such as oversampling or undersampling strategies, to further balance the ratio of abnormal to normal samples. In addition, the loss function can be replaced with other more complex functions, such as focal loss, to further improve the performance of the model when dealing with a small number of abnormal events.

[0129] Make predictions for the model, specifically:

[0130] For the current input feature matrix F t Make a prediction and output a probability value P anomaly , indicating the possibility that the current event is abnormal, by setting the dynamic threshold T dynamic To determine whether to trigger the final alarm,

[0131] If P anomaly >T dynamic , it is considered abnormal.

[0132] After receiving real-time economic data, the detection and alarm module will first convert this data into a feature matrix Ft and input it into the trained machine learning model. The model outputs a probability value P based on previous training experience. anomaly , represents the possibility that the current event is abnormal. Next, the module compares this probability value with the dynamically adjusted threshold T dynamic For comparison. dynamic It is a threshold value that is dynamically adjusted according to the actual situation and is used to balance the sensitivity and accuracy of the alarm. anomaly More than T dynamic , the system determines that the event is abnormal and triggers the corresponding alarm mechanism; on the contrary, if P anomaly Not reached T dynamic, the system considers that the event is within the normal fluctuation range and does not trigger an alarm. By outputting the probability value P anomaly By comparing with the dynamic threshold, the system can flexibly respond to different degrees of economic fluctuations. The use of dynamic thresholds ensures the adaptability and flexibility of the alarm system, and can reduce false alarms and missed alarms while ensuring accuracy. This mechanism is particularly suitable for situations where the economic environment changes greatly, making the system's alarm more accurate and timely.

[0133] In another embodiment, the system can introduce a multi-stage threshold judgment mechanism, that is, the abnormal probability value is divided into multiple levels, such as mild abnormality, moderate abnormality and severe abnormality, and different response measures are set for different levels. This multi-level threshold mechanism can provide users with more refined early warning services. In addition, T dynamic Adjustments can also be combined with multi-factor analysis, such as dynamic optimization based on external factors such as economic cycles and policy changes to improve the overall early warning capability of the system.

[0134] For the threshold T dynamic Make dynamic adjustments, specifically:

[0135] Introduce a dynamic threshold adjustment mechanism based on Bayesian optimization to continuously optimize the threshold T through online learning dynamic , to minimize the risk of false positives and false negatives, Bayesian optimization updates the prior distribution of the model through historical detection results and actual feedback, and adjusts the threshold at each detection to balance the accuracy and timeliness of the alarm. The formula is:

[0136]

[0137] Among them, Tdynamic is the dynamic threshold, p(T|data) is the prior distribution updated according to historical data, and Loss(T) is the loss function after the alarm is triggered.

[0138] During operation, the detection and alarm module will continuously receive new economic data and generate new abnormal detection results. In order to ensure that the alarm threshold T dynamic To ensure the rationality and accuracy of the model, the module uses Bayesian optimization technology for dynamic adjustment. Bayesian optimization combines historical detection results with actual feedback to update the prior distribution of the model so that T can be dynamically adjusted when new data arrives. dynamic After each detection, the system calculates a loss function Loss(T), which measures the accuracy of the alarm (for example, the number of false positives and false negatives). By minimizing the loss function, the system can optimize T dynamic, so that it can show the best alarm performance in the current economic environment. By introducing the dynamic threshold adjustment mechanism of Bayesian optimization, the system can adaptively adjust the alarm threshold, so as to maintain the best alarm effect in different economic environments. This method effectively reduces false alarms and missed alarms, and improves the overall performance of the system and user trust. In addition, the online learning mechanism enables the system to continuously optimize its alarm threshold over time, ensuring its long-term adaptability and stability.

[0139] In another embodiment, the system can be combined with other optimization algorithms, such as genetic algorithm or particle swarm optimization algorithm, to dynamically adjust T dynamic These algorithms perform well in dealing with complex multi-dimensional optimization problems and can provide the system with more optimization options. In addition, the threshold adjustment process can be combined with a user feedback mechanism to allow users to manually intervene or adjust the threshold to further improve the personalization and flexibility of the system.

[0140] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An economic indicator abnormality alarm system based on intelligent data analysis, characterized in that: The system comprises: Data collection module, used to collect data related to economic indicators; The analysis module is used to analyze the fluctuation characteristics of economic indicator data based on a preset algorithm model; A detection and alarm module, used to generate abnormal alarm information when it is detected that the fluctuation of the indicator exceeds a preset threshold; The analysis module analyzes the fluctuation characteristics of economic indicator data based on a preset algorithm model, including collecting time series information of economic data, extracting periodicity and trend components in the time series, analyzing the change law of the time series components of economic data, and calculating the standard deviation value of economic data based on the change law; The detection and alarm module generates abnormal alarm information when it detects that the fluctuation of the indicator exceeds the preset threshold, including setting standard upper and lower thresholds for measuring abnormalities, triggering an alarm after detecting that the latest reading exceeds the range, and starting the machine learning model to analyze whether the current event is a true abnormality rather than a normal variation in a statistical sense. If it is determined to be an abnormality, an email or text message is sent to relevant personnel, a detailed report is generated and archived simultaneously, and an alarm interface is pushed to the management end for manual inspection when necessary.

2. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 1 is characterized by: The data acquisition module collects economic indicator related data including: Receive transaction data from financial exchanges in real time; Extract historical financial reports; Get information on economic indices and market news; Get the latest money supply indicators; Crawl macroeconomic data published on the Internet.

3. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 1 is characterized by: The analysis module analyzes the changing rules of the time series components of economic data including: The improved Hurst index and autoregressive moving average model are used to jointly analyze the changing rules of time series components and calculate the Hurst index H of each k , the specific formula is: Among them, H k is the Hurst exponent, R / S is the relative fluctuation range, and n is the number of samples; The decomposed trend component r(t) is fitted with an autoregressive moving average model. The specific formula is: r(t)=φ1r(t-1)+φ2r(t-2)+…+θ1∈(t-1)+θ2∈(t-2)+…+∈(t), where r(t) is the trend component at time t, t is time, □1, □2, …, □ i are the autoregressive coefficients in the model, θ1, θ2, …, θ j is the moving average coefficient in the model, and □(t) is the random error term.

4. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 1 is characterized by: The analysis module calculates the standard deviation value of economic data based on the change law, including: Statistical historical data sets of economic indicators within a set period of time; Determine the ratio of the volatility component in the historical data set to the overall level; The standard deviation is calculated based on the proportion and overall level of the volatility components; If the standard deviation of economic indicator data in a certain period exceeds the set threshold T, it indicates abnormal fluctuations.

5. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 4 is characterized by: The calculation of the standard deviation based on the proportion and overall level of the volatility components includes: The proportion of the quantitative volatility component, B, is a parameter in the standard deviation formula; Use the overall mean M to adjust the sensitivity of the calculation formula; To obtain the standard deviation σ, use the formula: Where σ is the standard deviation, x i is the i-th data point, M is the average value, B is the coefficient of the volatility component, and n is the number of data points The obtained standard deviation result is used to compare with the predefined abnormal fluctuation limit.

6. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 5 is characterized by: The use of the obtained standard deviation result to compare the predefined abnormal fluctuation upper limit includes: Let the standard deviation result be SD; Compare the standard deviation SD value with the previously set upper limit of the safety fluctuation range T; If SD>T, the judgment result is 1, which means there is a potential volatility risk; If SD≤T, the result is 0, which means it is within the safe range; If SD exceeds the upper limit T, it is marked as potential volatility risk.

7. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 4 is characterized by: Determining the ratio of the volatility component in the historical data set to the overall level includes: In the historical data set obtained, the volatility component and the overall level are first separated, and the volatility component is represented by calculating the deviation between the economic indicator value at each time point and its trend level, specifically: V t =X t -T t ; Among them, V t is the volatility component, X t is the value of the economic indicator at time t, T t It is the trend component; The ratio of the volatility component to the overall level is expressed as: R t =V t / T t ; Among them, R t It is the ratio of the volatility component to the overall level.

8. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 1 is characterized by: The detection and alarm module starts the machine learning model to analyze whether the current event is a true anomaly rather than a normal variation in a statistical sense, including: Use known abnormal events and normal events in historical data to supervise the model. The loss function during training can use weighted cross entropy to cope with the situation where there are fewer abnormal event samples. Specifically: Where L is the loss function, m is the total number of training samples, wi is the weight of the i-th sample, yi is the actual label of the i-th sample, is the model prediction value of the ith sample.

9. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 8 is characterized by: The model is predicted as follows: For the current input feature matrix F t Make a prediction and output a probability value P anomaly , indicating the possibility that the current event is abnormal, by setting the dynamic threshold T dynamic To determine whether to trigger the final alarm, If P anomaly >T dynamic , it is considered abnormal.

10. The economic indicator abnormality alarm system based on intelligent data analysis according to claim 9 is characterized by: For the threshold T dynamic Make dynamic adjustments, specifically: Introduce a dynamic threshold adjustment mechanism based on Bayesian optimization to continuously optimize the threshold T through online learning dynamic , to minimize the risk of false positives and false negatives, Bayesian optimization updates the prior distribution of the model through historical detection results and actual feedback, and adjusts the threshold at each detection to balance the accuracy and timeliness of the alarm. The formula is: Among them, T dynamic is the dynamic threshold, p(T|data) is the prior distribution updated according to historical data, and Loss(T) is the loss function after the alarm is triggered.

Images