Big data platform BI analysis method and system and computer equipment

Abstract

The invention relates to the technical field of data analysis, in particular to a big data platform BI analysis method and system and computer equipment. According to the method, the real-time data is acquired and processed into the stable sequence data, the dynamic change conditions of an enterprise in different time periods can be more comprehensively understood, the dependence on historical data can be reduced through the dynamic prediction model, the prediction result can be updated in real time, and a faster response can be made, so that the prediction efficiency is improved. By acquiring the data in the third stationary sequence data set and comparing the data with the first trend prediction value, the consistency between the current state and the prediction trend can be evaluated in real time, so that the prediction path can be adjusted in time, and when the stationary sequence data is inconsistent with the first trend prediction value, the prediction path can be adjusted in time. The third trend prediction value is obtained by using additional data processing, so that the prediction is finer and adapts to changes, and the multi-level data analysis method enables enterprises to comprehensively consider past, current and future multi-dimensional factors to form more comprehensive judgment.

Description

Technical Field

[0001] The present invention relates to the technical field, and in particular to a big data platform BI analysis method, system and computer equipment. Background Art

[0002] With the rapid development of the Internet, big data and artificial intelligence technologies, enterprises' reliance on data has exploded. Data has become an important driving factor for enterprise operations, strategic planning and competitiveness improvement. In this context, big data platforms, as core tools for data storage, processing and analysis, have become important infrastructure for enterprises to mine data value and support decision-making. Among them, BI (business intelligence), as a technical means of integrating data analysis and visualization, plays an indispensable role in enterprise data insight, trend prediction and operation optimization.

[0003] BI analysis is the process of extracting, cleaning and converting the data of the business system and loading it into the data warehouse. With the development of scientific and technological information, data quality issues have always been a close concern in the data mining process, and data analysis is the core part of the entire BI analysis process. It involves deep mining, modeling and statistical analysis of cleaned data, and is a key link in gaining insights and conclusions. Currently, data analysis is mainly done through descriptive analysis. Descriptive analysis mainly helps users understand what happened in the past and the current status through data aggregation, visualization and summary. However, this method can only provide data summaries for the current or past period of time and lacks a perspective on dynamic changes in data. Summary of the invention

[0004] The main purpose of the present invention is to provide a big data platform BI analysis method, aiming to solve the technical problems in the prior art.

[0005] The present invention proposes a big data platform BI analysis method, comprising: Acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; Preprocessing each of the continuous information data to obtain corresponding stationary sequence data; Splitting the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; Building a time series prediction model based on the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; Generate a visualized trend change table according to the plurality of first trend prediction values, and send the visualized trend change table to a visualization analysis system for display; Acquire the first stationary sequence data in the third stationary sequence data set; Determining whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

[0006] Preferably, the step of preprocessing each of the continuous information data to obtain corresponding stationary sequence data comprises: Acquire data period fluctuations according to the plurality of continuous information data, and acquire a smoothing factor according to the data period fluctuations; The corresponding smoothed information data is calculated according to the smoothing factor and each of the continuous information data, wherein the calculation formula is: ; in, Indicates Smoothed information data, represents the smoothing factor, Indicates Continuous information data, Indicates Smoothed information data, A serial number indicating continuous information data; Obtaining a corresponding data residual according to each of the smoothed information data and the corresponding continuous information data, and determining whether the data residual is greater than a preset value; If the data residual is greater than a preset value, the smoothing information data corresponding to the data residual is determined to be abnormal point data; The outlier data in the plurality of smooth information data are removed to obtain corresponding stable sequence data.

[0007] Preferably, the step of obtaining data period fluctuations according to the plurality of continuous information data, and obtaining a smoothing factor according to the data period fluctuations comprises: Performing Fourier transformation on each of the continuous information data to obtain corresponding frequency domain representation data; Acquire the amplitude of each of the frequency domain representation data, and acquire an amplitude spectrum according to the multiple amplitudes; Acquire multiple peak values ​​according to the amplitude spectrum, and sort the multiple peak values ​​in order of magnitude to obtain a peak sorting table; Setting a preset amplitude threshold, and truncating the peak sorting table according to the preset amplitude threshold to obtain a significant peak sorting table; Obtaining the frequency corresponding to each peak in the significant peak sorting table, and obtaining periodic fluctuations according to the multiple frequencies; Acquire a cycle length according to the cycle fluctuation, and determine whether the cycle length is greater than a preset length; If the period length is greater than a preset length, the smoothing method is determined to be a sliding average, and a larger smoothing factor is determined according to the sliding smoothing; If the cycle length is not greater than a preset length, the smoothing method is determined to be exponential smoothing, and a smaller smoothing factor is determined based on the exponential smoothing.

[0008] Preferably, the step of dividing the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set comprises: Obtaining total sequence data according to the plurality of said stationary sequence data, and obtaining a corresponding stationary sequence mean according to the total sequence data and each stationary sequence data; Sorting the plurality of stationary sequence means in order of magnitude to obtain a stationary sequence mean sorting table; Establish a preset mean range; The stationary sequence mean sorting table is truncated according to the lower limit value of the preset mean interval to obtain a first stationary sequence data set and a secondary stationary sequence mean sorting table; The secondary stationary sequence mean sorting table is truncated according to the upper limit value of the preset mean interval to obtain a second stationary sequence data set and a third stationary sequence data set.

[0009] Preferably, the step of generating a visualized trend change table according to the plurality of first trend prediction values ​​comprises: Acquire a corresponding time node according to each of the first trend prediction values; In the visual analysis system, the time node is used as the X-axis, the first trend prediction value is used as the Y-axis, and the trend change coordinate axis is established; Selecting a time node from a plurality of time nodes in chronological order as an initial time node, and selecting a corresponding first trend prediction value according to the initial time node as an initial first trend prediction value; The initial time node and the initial first trend forecast value are taken as the starting point, and each remaining time node and the corresponding first trend forecast value are drawn as the connection point on the trend change coordinate axis; The starting point and multiple connection points are connected in sequence through broken lines to obtain a visual trend change table.

[0010] Preferably, the step of obtaining the third trend prediction value according to the stationary sequence data and the first trend prediction value comprises: Obtaining a corresponding prediction error according to each of the stationary sequence data and the first trend prediction value; Selecting a prediction error from a plurality of prediction errors as a target error according to a preset rule, and adjusting the parameters of the first time series prediction model according to the target error to obtain a second time series prediction model; Inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; Determining whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of selecting a prediction error from multiple prediction errors as the target error according to the preset rules, and adjusting the parameters of the first time series prediction model according to the target error, until all prediction errors are traversed to obtain the third trend prediction value.

[0011] The present application also provides a big data platform BI analysis system, including: A first acquisition module is used to acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; A preprocessing module, used for preprocessing each of the continuous information data to obtain corresponding stable sequence data; A segmentation module, used for segmenting the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; An input module, used for constructing a time series prediction model according to the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; A second acquisition module is used to acquire the first stationary sequence data in the third stationary sequence data set; A judgment module, used for judging whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

[0012] Preferably, the judging module comprises: A first acquisition unit, used for acquiring a prediction error according to the stationary sequence data and a first trend prediction value; An adjustment unit, configured to adjust the parameters of the first time series prediction model according to the prediction error to obtain a second time series prediction model; An input unit, used for inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; A judging unit, used to judge whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of adjusting the parameters of the first time series prediction model according to the first prediction error until all the stationary series data are traversed to obtain the third trend prediction value.

[0013] The present invention also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the above-mentioned big data platform BI analysis method when executing the computer program.

[0014] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which implements the steps of the above-mentioned big data platform BI analysis method when executed by a processor.

[0015] The beneficial effects of the present invention are as follows: the present invention can help to more comprehensively understand the dynamic changes of enterprises in different time periods by acquiring real-time data and processing it into steady sequence data, can reflect the actual operation status and trend changes of enterprises, avoid the misleading caused by relying solely on static data, the dynamic analysis method can track data in real time according to the flow of time, capture changes that are not easily perceived, and make up for the shortcomings of static analysis, the generation of multiple first trend prediction values ​​means accurate quantitative prediction of data changes at multiple time points in the future, through this dynamic prediction model, it is possible to reduce dependence on historical data, update the prediction results in real time and make a faster and more accurate response, by acquiring data in the third steady sequence data set and comparing it with the first trend prediction value, it is possible to evaluate the consistency of the current state with the predicted trend in real time, so as to adjust the prediction path in time, and when the steady sequence data is inconsistent with the first trend prediction value, use additional data processing to obtain the third trend prediction value, so that the prediction is more refined and adaptable to changes, avoiding the limitation of relying on a single prediction path, not only providing analysis of static data, but also adding a feedback mechanism for dynamic data changes, this multi-level data analysis method enables enterprises to comprehensively consider multi-dimensional factors of the past, present and future, and form a more comprehensive judgment. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 The figure is a schematic diagram of a method flow according to an embodiment of the present invention.

[0017] Figure 2 FIG. 1 is a schematic diagram of a device structure according to an embodiment of the present invention.

[0018] Figure 3 A schematic diagram of the internal structure of a computer device according to an embodiment of the present application.

[0019] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0020] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0021] like Figure 1-Figure 3 As shown, the present application provides a big data platform BI analysis method, including: S1. Acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; S2, preprocessing each of the continuous information data to obtain corresponding stable sequence data; S3, dividing the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; S4, constructing a time series prediction model according to the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; S5. Generate a visualized trend change table according to the plurality of first trend prediction values, and send the visualized trend change table to a visualization analysis system for display; S6, obtaining the first stationary sequence data in the third stationary sequence data set; Determining whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

[0022] As described in the above steps S1-S2, with the rapid development of the Internet, big data and artificial intelligence technologies, the dependence of enterprises on data has exploded. Data has become an important driving factor for enterprise operations, strategic planning and competitiveness improvement. In this context, big data platforms, as core tools for data storage, processing and analysis, have become important infrastructure for enterprises to mine data value and support decision-making. Among them, BI (business intelligence), as a technical means of integrating data analysis and visualization, plays an indispensable role in enterprise data insight, trend prediction and operation optimization. BI analysis is the process of loading the data of the business system into the data warehouse after extraction, cleaning and conversion. With the development of scientific and technological information, data quality issues have always been a problem of close attention in the data mining process. Data analysis is the core part of the entire BI analysis process. It involves deep mining, modeling and statistical analysis of cleaned data, and is a key link in obtaining insights and conclusions. At present, data analysis is mainly through descriptive analysis. Descriptive analysis mainly helps users understand what happened in the past and the current status through data aggregation, visualization and summary. However, this method can only provide data aggregation for the current or past period of time and lacks a perspective on the dynamic changes of data. The present invention obtains real-time dynamic information data of the enterprise and historical information data of the enterprise including multiple continuous information data within a preset time period, and obtains corresponding stable sequence data by preprocessing each continuous information data, wherein the stable sequence data refers to a data sequence whose statistical characteristics remain unchanged over time, and divides the multiple stable sequence data into a first stable sequence data set, a second stable sequence data set and a third stable sequence data set. Since traditional descriptive analysis methods usually focus on data aggregation over a period of time in the past and mainly give static trends and summaries, the present invention can help to more comprehensively understand the dynamic changes of the enterprise in different time periods by obtaining real-time data and processing it into stable sequence data, which can reflect the actual operating status and trend changes of the enterprise, avoid the misleading caused by relying solely on static data, and The method can more effectively mine the potential trends and cyclical changes in the data by segmenting the rows and forming multiple stationary sequence data sets. The processing of each stationary sequence helps to identify the long-term laws and short-term fluctuations of the data, so as to accurately identify potential trend changes or abnormal phenomena, which is crucial for corporate decision-making. Converting data into a stationary sequence can eliminate non-stationary components in the data (such as trends, seasonality, etc.), making the analysis results more accurate. Through such preprocessing, subsequent modeling and prediction (time series prediction model) will be more effective, and can provide more reliable future prediction results, helping companies make better strategic decisions. The data is segmented into different stationary sequence sets, so that the changing trends of different time periods can be independently identified, and the data changes of each time period (such as the past period, current dynamic changes and future trends) can be deeply analyzed.Avoid simply summarizing and analyzing all data. Such multi-dimensional analysis helps to evaluate corporate performance from different perspectives and time scales. The introduction of real-time dynamic information enables this method to more quickly discover abnormal fluctuations or emergencies in corporate operations. Traditional static analysis usually lacks sensitivity to data changes and trends, and tends to ignore instantaneous changes and short-term fluctuations. The dynamic analysis method can track data in real time according to the flow of time and capture changes that are not easily perceived, making up for the shortcomings of static analysis. A time series prediction model is constructed based on the first stationary sequence data set, and the second stationary sequence data set is input into the first time series prediction model to obtain multiple first trend prediction values. A visual trend is generated based on the multiple first trend prediction values. The visual trend change table is generated and sent to the visual analysis system for display. By constructing a time series prediction model based on the first stationary series data set and inputting the second stationary series data set into the model, future trends can be dynamically predicted, thereby providing a forward-looking perspective on future data changes. The generation of multiple first trend prediction values ​​means accurate quantitative prediction of data changes at multiple time points in the future. Traditional descriptive analysis can only make a static conclusion based on existing data, while this method can continuously predict future trend changes through the time series prediction model. By generating a visual trend change table from multiple first trend prediction values ​​and sending it to the visual analysis system for display, data can be The traditional descriptive analysis method usually presents static summary data, while the dynamic trend forecast results are displayed visually, which not only helps decision makers understand the data intuitively, but also makes it easier to identify fluctuations, anomalies, seasonal changes or long-term trends in the data. By inputting the second stationary sequence data set into the time series forecasting model, the lag problem of traditional methods in data aggregation can be reduced. Traditional methods usually need to summarize and analyze the data of a past time period each time they perform descriptive analysis, so there is a problem of data lag. Through this dynamic forecasting model, the dependence on historical data can be reduced, the forecast results can be updated in real time, and a faster and more accurate response can be made. By obtaining the first The first stationary sequence data in the three stationary sequence data sets is used to determine whether the stationary sequence data is the same as the corresponding first trend prediction value. If they are the same, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain the second trend prediction value. If they are different, the third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value. Traditional descriptive analysis methods are usually limited to static summaries. They can only make analysis based on known historical data and lack the ability to respond to future trends and data changes in real time. In contrast, the present invention obtains data in the third stationary sequence data set and compares it with the first trend prediction value. It can evaluate the consistency between the current state and the predicted trend in real time, thereby adjusting the prediction path in time.When the stationary sequence data is inconsistent with the first trend forecast value, additional data processing can be used to obtain the third trend forecast value, making the forecast more refined and adaptable to changes, avoiding the limitation of relying on a single forecast path. When the new data does not match the original forecast value, it can be adjusted through this comparison mechanism to obtain a more accurate and reliable trend forecast. In contrast, traditional descriptive analysis only relies on past data and lacks the ability to dynamically adjust the forecast based on the data. By comparing the consistency of the stationary sequence data and the forecast value, this method can timely discover anomalies or deviations in the data, thereby reducing the forecast errors that may occur in traditional descriptive analysis methods. If the stationary sequence data is consistent with the forecast value, real-time data is used for secondary forecasting; if not, a new forecast value is obtained based on the actual data situation. This dynamic comparison and adjustment mechanism improves the accuracy of the forecast results. By comparing the first trend forecast value with the stationary sequence data, this method not only provides analysis of static data, but also adds a feedback mechanism for dynamic data changes. This multi-level data analysis method enables enterprises to comprehensively consider multi-dimensional factors of the past, present and future to form a more comprehensive judgment. ,

[0023] In one embodiment, the step S2 of preprocessing each of the continuous information data to obtain corresponding stationary sequence data includes: S21, obtaining data period fluctuations according to the plurality of continuous information data, and obtaining a smoothing factor according to the data period fluctuations; S22, calculating corresponding smoothed information data according to the smoothing factor and each of the continuous information data, wherein the calculation formula is: ; in, Indicates Smoothed information data, represents the smoothing factor, Indicates Continuous information data, Indicates Smoothed information data, A serial number indicating continuous information data; S23, obtaining a corresponding data residual according to each of the smoothed information data and the corresponding continuous information data, and determining whether the data residual is greater than a preset value; If the data residual is greater than a preset value, the smoothing information data corresponding to the data residual is determined to be abnormal point data; The outlier data in the plurality of smooth information data are removed to obtain corresponding stable sequence data.

[0024] As described in the above steps S21-S23, the present invention obtains data periodic fluctuations through multiple continuous information data, obtains a smoothing factor based on the data periodic fluctuations, and calculates corresponding smoothed information data based on the smoothing factor and each of the continuous information data, wherein data periodic fluctuations refer to the regular fluctuations or periodic fluctuations of data within a certain time period, and the smoothing factor is a parameter used to smooth the time series data, the purpose of which is to eliminate noise or random fluctuations in the data, thereby revealing the basic trend or law of the data. By introducing the calculation of data periodic fluctuations and smoothing factors, the changing law of data can be captured in real time, the periodic fluctuations of data can be identified, and then the changing trend of data can be dynamically tracked, which can provide more flexible analysis and recognition. Identifying and responding to long-term and short-term cyclical fluctuations in data can help better predict future trends or abnormal changes. Data is usually affected by noise, especially in continuous time data, which often contains random fluctuations or external interference. Traditional methods may not be able to effectively remove these noises, while smoothing factors can help eliminate these noises, make data more stable, remove extreme fluctuations and unnecessary interference, and extract the true trend of the data. Through smoothing, the stability of the data can be improved, the impact of short-term random fluctuations on the analysis results can be reduced, and long-term trends or cyclical changes can be more prominent. For the analysis of cyclical data, traditional descriptive analysis may only rely on statistics such as averages, which cannot effectively reveal the cyclical fluctuation characteristics in the data. By analyzing the data According to periodic fluctuations, the periodic characteristics of data can be identified, so as to model and predict periodic trends. Periodic analysis can more accurately capture the regular fluctuations of data and help analysts understand the impact of long-term trends and seasonal factors. The smoothed data calculated by the smoothing factor can provide decision makers with a more stable and reliable analysis basis. Compared with traditional descriptive analysis methods, the combination of periodic fluctuation analysis and smoothing factors can reduce the misleading caused by data fluctuations and improve the accuracy of decision-making. Smoothed data can filter out unnecessary short-term fluctuations and outliers, thereby providing decision makers with clearer and more predictable dynamic change trends. The corresponding data residuals are obtained through each smoothed information data and the corresponding continuous information data. Data residual refers to the difference between the actual observation value and the model prediction value in the time series prediction model, and determines whether the data residual is greater than the preset value. If it is greater, the smooth information data corresponding to the data residual is determined to be an outlier data, and then the corresponding stable series data is obtained by removing the outlier data from multiple smooth information data. Unlike the traditional descriptive analysis method that only provides static data summaries for the current or past period of time, this method can dynamically capture the changing trends and fluctuations of the data, identify and process outliers in real time, and improve the ability to respond to dynamic changes in data. By removing outliers through residual analysis, noise and interference are effectively removed, making the data more stable, and avoiding the impact of short-term fluctuations and extreme values ​​on the results that may be ignored in static analysis methods.It can identify long-term trends and cyclical fluctuations instead of relying solely on data from a single time window, providing more accurate trend analysis and better reflecting the dynamic evolution of data in a complex environment. It can avoid misjudgments caused by data noise by removing outliers and smoothing data, which helps improve the stability and accuracy of model predictions, especially when processing time series data.

[0025] In one embodiment, the step S21 of obtaining data period fluctuations according to the plurality of continuous information data and obtaining a smoothing factor according to the data period fluctuations comprises: S211, performing Fourier transform on each of the continuous information data to obtain corresponding frequency domain representation data; S212, obtaining the amplitude of each of the frequency domain representation data, and obtaining an amplitude spectrum according to the multiple amplitudes; S213, obtaining a plurality of peak values ​​according to the amplitude spectrum, and sorting the plurality of peak values ​​in order of magnitude to obtain a peak sorting table; S214, setting a preset amplitude threshold, and truncating the peak sorting table according to the preset amplitude threshold to obtain a significant peak sorting table; S215, obtaining the frequency corresponding to each peak in the significant peak sorting table, and obtaining periodic fluctuations according to the multiple frequencies; S216, obtaining a cycle length according to the cycle fluctuation, and determining whether the cycle length is greater than a preset length; If the period length is greater than a preset length, the smoothing method is determined to be a sliding average, and a larger smoothing factor is determined according to the sliding smoothing; If the cycle length is not greater than a preset length, the smoothing method is determined to be exponential smoothing, and a smaller smoothing factor is determined based on the exponential smoothing.

[0026] As described in the above steps S211-S216, the present invention obtains corresponding frequency domain representation data by performing Fourier transform on each continuous information data, obtains the amplitude of each frequency domain representation data, and obtains the amplitude spectrum according to multiple amplitudes, obtains multiple peaks through the amplitude spectrum, and sorts the multiple peaks in order of size to obtain a peak sorting table, truncates the peak sorting table by setting a preset amplitude threshold, and obtains a significant peak sorting table, obtains the frequency corresponding to each peak in the significant peak sorting table, and obtains periodic fluctuations according to multiple frequencies. Traditional descriptive analysis methods usually focus on static characteristics of the data such as the mean and variance, and often ignore the potential periodicity and frequency domain characteristics in the data. Through Fourier transform, continuous time series can be converted into frequency domain representation, revealing the periodic fluctuations and frequency components in the data, which helps to discover long-term trends and periodic changes that traditional methods cannot identify. By converting to the frequency domain, Fourier transform makes the analysis of data not limited to the current or past time window, but can be observed from a global dynamic perspective. By extracting significant peaks and frequencies in the frequency domain, information about the data change cycle can be obtained, thereby identifying the fluctuation characteristics of the data on different time scales. Compared with static analysis methods, this frequency domain analysis method is more adaptable to the dynamic changes of data in time. By setting a preset amplitude threshold and truncating the peak sorting table, the method can effectively filter out low-amplitude noise in the frequency domain and retain only significant frequency components. This processing can remove unimportant fluctuation components in the data, thereby helping to more clearly identify important periodic components. By extracting the frequency of significant peaks, multiple periodic fluctuations in the data can be obtained, which not only helps to understand the short-term and long-term trends of the data, but also can reveal the fluctuation patterns at multiple levels in complex data.For example, the data may have both daily fluctuations and seasonal fluctuations. The Fourier transform can help separate these fluctuations of different scales. By extracting periodic components from the frequency domain, it can better identify and predict future fluctuation trends. Especially in systems with periodic fluctuations, the frequency information provided by the Fourier transform can provide valuable input for the prediction model, improve the prediction ability based on historical summary data in traditional methods, obtain the cycle length through periodic fluctuations, and determine whether the cycle length is greater than the preset length. If it is greater than the preset length, the smoothing method is determined to be sliding average, and a larger smoothing factor is determined based on sliding smoothing. If it is not greater than the preset length, the smoothing method is determined to be exponential smoothing, and a larger smoothing factor is determined based on exponential smoothing. Smaller smoothing factor. Traditional descriptive analysis methods usually use fixed statistical methods, such as simple mean or variance, to describe data characteristics. These methods can usually only reflect the data summary of the current or past period of time, and lack adaptability to dynamic changes in data. The present invention dynamically selects the smoothing method (sliding average or exponential smoothing) according to the periodicity and cycle length of the data, and adjusts the smoothing factor according to different periodic fluctuation characteristics, so that the smoothing process is more in line with the actual fluctuation of the data. This adaptability can effectively reflect the dynamic characteristics of the data rather than static summary. By analyzing the periodic fluctuations, it can capture the long-term periodicity and short-term fluctuations of the data, especially when the cycle length is large, selecting sliding The average can smooth out longer-term fluctuations and adapt to relatively stable data trends. When the cycle length is short, exponential smoothing will pay more attention to changes in recent data and can respond more sensitively to short-term fluctuations. This periodic analysis method helps to reveal the dynamic changes of data and avoids the limitations of descriptive analysis methods that only focus on fixed data summaries within the current or historical period and ignore long-term trends or short-term fluctuations. Sliding average and exponential smoothing each have different smoothing effects. Sliding average is suitable for smoothing longer cycles and can effectively remove noise in long-term trends and ensure that long-term cyclical components will not be over-smoothed. Exponential smoothing is suitable for smoothing shorter cycles and can better capture recent data changes and avoid Important information is lost in rapidly changing data. By selecting the smoothing method according to the cycle length, over-smoothing caused by using too large a smoothing factor is avoided, so that important fluctuation information in the data can be retained. By dynamically selecting the smoothing method and adjusting the smoothing factor, the method can flexibly analyze different types of data (long-term cyclical changes or short-term fluctuations), avoiding the one-size-fits-all approach for different types of data in descriptive analysis methods. For time series data with strong cyclical changes, the sliding average can effectively remove the impact of long-term fluctuations; for data that changes more frequently or is dominated by short-term fluctuations, exponential smoothing can better reflect short-term dynamic changes, thereby providing more accurate data analysis results.

[0027] In one embodiment, the step S3 of dividing the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set comprises: S31, obtaining total sequence data according to the plurality of stationary sequence data, and obtaining a corresponding stationary sequence mean according to the total sequence data and each stationary sequence data; S32, sorting the plurality of stationary sequence means in order of magnitude to obtain a stationary sequence mean sorting table; S33, setting a preset mean value interval; S34, truncating the stationary sequence mean sorting table according to the lower limit of the preset mean interval to obtain a first stationary sequence data set and a secondary stationary sequence mean sorting table; S35. Truncate the secondary stationary sequence mean sorting table according to the upper limit value of the preset mean interval to obtain a second stationary sequence data set and a third stationary sequence data set.

[0028] As described in the above steps S31-S35, the present invention obtains total sequence data through multiple stationary sequence data, and obtains the corresponding stationary sequence mean according to the total sequence data and each stationary sequence data, and sorts the multiple stationary sequence means in order of size to obtain a stationary sequence mean sorting table. Traditional descriptive analysis methods often focus on summarizing the data for the current or past period of time, and cannot effectively capture the dynamic changes of the data. By integrating the means of multiple stationary sequences and sorting them, the changing trend of the data in different time periods can be better reflected, thereby providing a more dynamic perspective for data analysis. Especially in the context of multiple stationary sequences, this method can reveal the fluctuation relationship and change pattern between different sequences. By merging multiple stationary sequence data, calculating the mean of each stationary sequence and sorting them, a more comprehensive global data view can be obtained. This method can transcend the limitations of a single time period, comprehensively consider the characteristics of different sequences, and provide a broader data observation angle, thereby revealing potential laws and trends. This global data analysis can make up for the static The analysis only focuses on the shortcomings of data aggregation at a certain moment or a period of time. The mean sorting of stationary series can not only reveal the relative importance of each series, but also further analyze the hierarchical structure or interdependence in the data through the size relationship of these series. By combining the information of multiple stationary series and sorting them according to their means, the trend characteristics of data in different periods can be better captured, which helps to improve the ability to predict future data changes. A preset mean interval is set, and the stationary series mean sorting table is truncated by the lower limit of the preset mean interval to obtain the first stationary series data set and the secondary stationary series mean sorting table. Then, the secondary stationary series mean sorting table is truncated by the upper limit of the preset mean interval to obtain the second stationary series data set and the third stationary series data set. By setting a preset mean interval and truncating the stationary series mean sorting table, the method can divide the data into multiple levels (such as the first stationary series, the secondary stationary series, the third stationary series, etc.), so as to better capture the changes of data at different stages.Each level can reflect different dynamic characteristics of data, helping analysts understand the changing trend of data from different angles. The present invention can dynamically adjust the analysis scope of data. Data sets at different levels can represent different trend stages (such as short-term fluctuations, long-term trends, etc.). Therefore, the method can adaptively capture various changing forms of data, providing a more flexible analysis method. By truncating the mean sorting table of the stationary sequence, the focus can be shifted from the entire data set to an in-depth analysis of a specific stage or a specific trend. Through this segmented truncation method, the changing pattern and trend of data in different intervals can be revealed. By setting up mean intervals and segmenting the data, abnormal fluctuations or drastic changes in the data can be more flexibly dealt with. Traditional methods may ignore abnormal fluctuations in the short term due to excessive reliance on historical data aggregation, and through dynamic truncation analysis, these changes can be quickly identified and responded to. For example, if the mean of a stationary sequence exceeds the upper or lower limit of the preset interval, it can be processed separately and potential anomalies can be identified. This is especially important for complex data sets. By dividing the data into multiple stationary sequences and analyzing each level independently, the short-term changes and long-term trends of the data can be captured more accurately. This is particularly beneficial for time series prediction. For example, a long-term stable data set can help identify long-term trends, while a short-term fluctuating data set can provide a reference for short-term forecasts. By analyzing the data in a hierarchical manner, the model's ability to predict future dynamic changes can be enhanced, thereby improving the accuracy of the forecast.

[0029] In one embodiment, the step S5 of generating a visualized trend change table according to the plurality of first trend prediction values ​​comprises: S51, acquiring a corresponding time node according to each of the first trend prediction values; S52, in the visual analysis system, with the time node as the X-axis and the first trend prediction value as the Y-axis, a trend change coordinate axis is established; S53, selecting a time node from a plurality of time nodes in chronological order as an initial time node, and selecting a corresponding first trend prediction value according to the initial time node as an initial first trend prediction value; S54, taking the initial time node and the initial first trend prediction value as the starting point, and taking each remaining time node and the corresponding first trend prediction value as a connection point and drawing them on the trend change coordinate axis; S55, connecting the starting point and multiple connection points in sequence through broken lines to obtain a visualized trend change table.

[0030] As described in the above steps S51-S55, the present invention obtains the corresponding time node through each first trend prediction value, and establishes a trend change coordinate axis in the visual analysis system with the time node as the X-axis and the first trend prediction value as the Y-axis, and selects a time node as the initial time node from multiple time nodes in chronological order, and selects the corresponding first trend prediction value as the initial first trend prediction value according to the initial time node, and takes the initial time node and the initial first trend prediction value as the starting point, and draws each remaining time node and the corresponding first trend prediction value as a connection point on the trend change coordinate axis, and finally connects the starting point and multiple connection points in turn through broken lines to obtain a visual Traditional descriptive analysis methods generally provide statistical information by summarizing data from the current or past period of time, but these methods cannot intuitively display the changing trend and dynamic fluctuation of data over time. By taking the time node as the X-axis and the trend prediction value as the Y-axis, and connecting different time nodes in the form of a broken line, the present invention can intuitively display the changes of data over time. This visualized trend change table can dynamically reflect the fluctuation pattern of the data instead of static summary values, thereby providing users with a complete time series analysis perspective. By establishing a trend change coordinate axis and a broken line chart, it is possible to not only focus on short-term data changes, but also capture changes in long-term trends. For example, in multiple Between time nodes, users can observe the rising, falling, and stable patterns of trend forecast values, which helps to identify long-term growth or recession trends. Traditional descriptive analysis methods may only focus on the summary data at a specific moment and cannot reveal these long-term changes. The dynamic changes of data usually need to be understood and analyzed in a more intuitive way, especially when the amount of data is large or the changes are complex. The summary of text or numbers often makes it difficult for users to quickly understand the dynamic fluctuations of data. By connecting time nodes and forecast values ​​in the form of line graphs, trend change charts make complex data sequences more concise and easy to understand. Users can quickly capture key trend change points and improve their understanding of data. The efficiency of understanding and decision-making can be improved by selecting one of multiple time nodes as the starting point and drawing a trend change graph in chronological order. It can adapt to the needs of different data time periods. Whether it is short-term daily fluctuations or long-term seasonal trends, it can be flexibly displayed by adjusting the initial time node and trend forecast value. In this way, customized analysis can be performed for the change cycle of different data. Traditional descriptive analysis rules can often only provide statistical data limited to the past period of time and cannot flexibly respond to multi-period and multi-changing data characteristics. By dynamically displaying the relationship between time nodes and trend forecast values, it can help analysts and decision makers respond to data changes more quickly. For example, when the trend of a certain time period changes dramatically, the line chart can instantly display this change, helping users quickly identify the change in trend, while traditional static data aggregation methods may lag behind and cannot provide such rapid feedback.

[0031] In one embodiment, the step S6 of obtaining the third trend prediction value according to the stationary sequence data and the first trend prediction value comprises: S61, obtaining a corresponding prediction error according to each of the stationary sequence data and the first trend prediction value; S62, selecting a prediction error from multiple prediction errors as a target error according to a preset rule, and adjusting the parameters of the first time series prediction model according to the target error to obtain a second time series prediction model; S63, inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; S64, determining whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of selecting a prediction error from multiple prediction errors as the target error according to the preset rules, and adjusting the parameters of the first time series prediction model according to the target error, until all prediction errors are traversed to obtain the third trend prediction value.

[0032] As described in the above steps S61-S64, the present invention obtains the corresponding prediction error through each stationary series data and the first trend prediction value, selects a prediction error as the target error from multiple prediction errors according to preset rules, and adjusts the parameters of the first time series prediction model according to the target error to obtain the second time series prediction model. Traditional descriptive analysis methods mainly rely on the summary of historical data, and usually can only provide data conditions for the current or past period of time, and cannot reflect the dynamic changes of data in a timely manner. In contrast, the present invention uses a dynamic feedback mechanism of prediction error to adjust the parameters of the time series prediction model according to the target error, so that the model can respond to data changes in real time and enhance the adaptability to data dynamic trends. By selecting a target error from multiple prediction errors and adjusting the model parameters accordingly, the method has a "self-correction" ability, which means that even if there is a deviation in the preliminary prediction model, as more time series data are input, the model can be continuously adjusted to reduce the prediction error and optimize the prediction effect. The present invention can optimize the model parameters based on the error conditions of current and historical data. Compared with traditional static models, it can be corrected in real time according to new data changes, thereby improving the prediction accuracy. Especially in the face of complex or highly volatile time series data, the adjusted prediction model can more accurately capture the trends and patterns of the data and reduce prediction deviations. Therefore, the model not only analyzes current and past data, but also adapts to future data fluctuations, thereby providing more accurate and effective predictions in a changing environment. By inputting the third stationary series data set into the second time series prediction model to obtain a second trend prediction value, and judging whether the second trend prediction value is consistent with the real-time dynamic information data of the enterprise The same, if the same, the enterprise real-time dynamic information data is input into the second time series prediction model to obtain the third trend prediction value. If different, return to the step of selecting a prediction error as the target error from multiple prediction errors according to the preset rules, and adjusting the parameters of the first time series prediction model according to the target error, until all prediction errors are traversed to obtain the third trend prediction value. The present invention can dynamically respond to data changes by judging the difference between the prediction value and the enterprise dynamic information in real time and adjusting the model according to the difference feedback. Whenever the actual data (such as the enterprise real-time dynamic information) does not match the prediction value, the system will adjust the prediction model according to the error feedback to optimize the prediction accuracy, so that the model can adapt to new information changes in real time instead of relying on past static data. By comparing the real-time dynamic information with the trend prediction value, the error is judged and the prediction model is self-corrected. This adaptive mechanism enables the model to quickly adjust the model parameters when the enterprise environment changes, thereby continuously providing accurate predictions and avoiding the problem of lagging or outdated prediction models. Traditional descriptive analysis methods cannot dynamically update or adjust the model. Once the model or prediction algorithm is determined, it usually remains unchanged for a long time and lacks automatic update and optimization mechanisms.The present invention can ensure continuous optimization of the model by continuously selecting the target error from multiple prediction errors and adjusting the model. This dynamic adjustment mechanism can continuously reduce the prediction error during data changes, thereby gradually improving the accuracy of the prediction. By inputting real-time enterprise dynamic information into the second time series prediction model, it can directly affect and optimize trend prediction. This method enables the model to combine the latest enterprise dynamic information to obtain more accurate trend predictions, avoiding the lag caused by relying solely on historical data. By combining real-time dynamic data with the prediction model, the method can provide more real-time and accurate trend predictions. When the enterprise dynamic information is inconsistent with the prediction result, the system will adjust the model until a prediction value consistent with the real-time dynamic information is generated. This method can greatly improve the timeliness of the data and make the prediction results more accurate and forward-looking.

[0033] The present application also provides a big data platform BI analysis system, including: A first acquisition module is used to acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; A preprocessing module, used for preprocessing each of the continuous information data to obtain corresponding stable sequence data; A segmentation module, used for segmenting the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; An input module, used for constructing a time series prediction model according to the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; A second acquisition module is used to acquire the first stationary sequence data in the third stationary sequence data set; A judgment module, used for judging whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

[0034] In one embodiment, the judgment module includes: A first acquisition unit, used for acquiring a prediction error according to the stationary sequence data and a first trend prediction value; An adjustment unit, configured to adjust the parameters of the first time series prediction model according to the prediction error to obtain a second time series prediction model; An input unit, used for inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; A judging unit, used to judge whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of adjusting the parameters of the first time series prediction model according to the first prediction error until all the stationary series data are traversed to obtain the third trend prediction value.

[0035] It should be noted that each module and unit in the online defect detection system in the PCB production process corresponds one-to-one to the steps in the online defect detection method in the PCB production process.

[0036] like Figure 3 As shown, the present application also provides a computer device, which may be a server, and its internal structure may be as shown in Figure 3 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor designed by the computer is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store all data required for the process of the online defect detection method in the PCB production process. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, the online defect detection method based on the PCB production process is implemented.

[0037] Those skilled in the art will understand that Figure 3 The structure shown in is merely a block diagram of a portion of the structure related to the present application solution and does not constitute a limitation on the computer device to which the present application solution is applied.

[0038] An embodiment of the present application also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned online defect detection methods based on the PCB production process is implemented.

[0039] Those of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media provided in this application and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0040] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, device, article or method including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, device, article or method. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, device, article or method including the element.

[0041] The above description is only a preferred embodiment of the present application, and does not limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made using the contents of the present application specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present application.

Claims

1. A big data platform BI analysis method, applied to a visual analysis system, characterized in that: include: Acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; Preprocessing each of the continuous information data to obtain corresponding stationary sequence data; Splitting the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; Building a time series prediction model based on the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; Generate a visualized trend change table according to the plurality of first trend prediction values, and send the visualized trend change table to a visualization analysis system for display; Acquire the first stationary sequence data in the third stationary sequence data set; Determining whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

2. The big data platform BI analysis method according to claim 1, characterized in that: The step of preprocessing each of the continuous information data to obtain corresponding stationary sequence data includes: Acquire data period fluctuations according to the plurality of continuous information data, and acquire a smoothing factor according to the data period fluctuations; The corresponding smoothed information data is calculated according to the smoothing factor and each of the continuous information data, wherein the calculation formula is: ； in, Indicates Smoothed information data, represents the smoothing factor, Indicates Continuous information data, Indicates Smoothed information data, A serial number indicating continuous information data; Obtaining a corresponding data residual according to each of the smoothed information data and the corresponding continuous information data, and determining whether the data residual is greater than a preset value; If the data residual is greater than a preset value, the smoothing information data corresponding to the data residual is determined to be abnormal point data; The outlier data in the plurality of smooth information data are removed to obtain corresponding stable sequence data.

3. The big data platform BI analysis method according to claim 2, characterized in that: The step of obtaining data period fluctuations according to the plurality of continuous information data, and obtaining a smoothing factor according to the data period fluctuations comprises: Performing Fourier transformation on each of the continuous information data to obtain corresponding frequency domain representation data; Acquire the amplitude of each of the frequency domain representation data, and acquire an amplitude spectrum according to the multiple amplitudes; Acquire multiple peak values ​​according to the amplitude spectrum, and sort the multiple peak values ​​in order of magnitude to obtain a peak sorting table; Setting a preset amplitude threshold, and truncating the peak sorting table according to the preset amplitude threshold to obtain a significant peak sorting table; Obtaining the frequency corresponding to each peak in the significant peak sorting table, and obtaining periodic fluctuations according to the multiple frequencies; Acquire a cycle length according to the cycle fluctuation, and determine whether the cycle length is greater than a preset length; If the period length is greater than a preset length, the smoothing method is determined to be a sliding average, and a larger smoothing factor is determined according to the sliding smoothing; If the cycle length is not greater than a preset length, the smoothing method is determined to be exponential smoothing, and a smaller smoothing factor is determined based on the exponential smoothing.

4. The big data platform BI analysis method according to claim 1, characterized in that: The step of dividing the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set comprises: Obtaining total sequence data according to the plurality of said stationary sequence data, and obtaining a corresponding stationary sequence mean according to the total sequence data and each stationary sequence data; Sorting the plurality of stationary sequence means in order of magnitude to obtain a stationary sequence mean sorting table; Establish a preset mean range; The stationary sequence mean sorting table is truncated according to the lower limit value of the preset mean interval to obtain a first stationary sequence data set and a secondary stationary sequence mean sorting table; The secondary stationary sequence mean sorting table is truncated according to the upper limit value of the preset mean interval to obtain a second stationary sequence data set and a third stationary sequence data set.

5. The big data platform BI analysis method according to claim 1, characterized in that: The step of generating a visualized trend change table according to the plurality of first trend prediction values ​​comprises: Acquire a corresponding time node according to each of the first trend prediction values; In the visual analysis system, the time node is used as the X-axis, the first trend prediction value is used as the Y-axis, and the trend change coordinate axis is established; Selecting a time node from a plurality of time nodes in chronological order as an initial time node, and selecting a corresponding first trend prediction value according to the initial time node as an initial first trend prediction value; The initial time node and the initial first trend forecast value are taken as the starting point, and each remaining time node and the corresponding first trend forecast value are drawn as the connection point on the trend change coordinate axis; The starting point and multiple connection points are connected in sequence through broken lines to obtain a visual trend change table.

6. The big data platform BI analysis method according to claim 1, characterized in that: The step of obtaining the third trend prediction value according to the stationary sequence data and the first trend prediction value comprises: Obtaining a corresponding prediction error according to each of the stationary sequence data and the first trend prediction value; Selecting a prediction error from a plurality of prediction errors as a target error according to a preset rule, and adjusting the parameters of the first time series prediction model according to the target error to obtain a second time series prediction model; Inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; Determining whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of selecting a prediction error from multiple prediction errors as the target error according to the preset rules, and adjusting the parameters of the first time series prediction model according to the target error, until all prediction errors are traversed to obtain the third trend prediction value.

7. A big data platform BI analysis system, characterized in that: include: A first acquisition module is used to acquire enterprise historical information data and enterprise real-time dynamic information data, wherein the enterprise historical information data includes a plurality of continuous information data within a preset time period; A preprocessing module, used for preprocessing each of the continuous information data to obtain corresponding stable sequence data; A segmentation module, used for segmenting the plurality of stationary sequence data into a first stationary sequence data set, a second stationary sequence data set and a third stationary sequence data set; An input module, used for constructing a time series prediction model according to the first stationary series data set, and inputting the second stationary series data set into the first time series prediction model to obtain a plurality of first trend prediction values; A second acquisition module, used to acquire the first stationary sequence data in the third stationary sequence data set; A judgment module, used for judging whether the stationary sequence data is the same as the corresponding first trend prediction value; If the stationary sequence data is the same as the first trend prediction value, the enterprise real-time dynamic information data is input into the first time series prediction model to obtain a second trend prediction value; If the stationary sequence data is different from the first trend prediction value, a third trend prediction value is obtained according to the stationary sequence data and the first trend prediction value.

8. The big data platform BI analysis system according to claim 7, characterized in that: The judging module comprises: A first acquisition unit, used for acquiring a prediction error according to the stationary sequence data and a first trend prediction value; An adjustment unit, configured to adjust the parameters of the first time series prediction model according to the prediction error to obtain a second time series prediction model; An input unit, used for inputting the third stationary series data set into a second time series prediction model to obtain a second trend prediction value; A judging unit, used to judge whether the second trend prediction value is the same as the enterprise real-time dynamic information data; If the second trend prediction value is the same as the enterprise real-time dynamic information data, inputting the enterprise real-time dynamic information data into the second time series prediction model to obtain a third trend prediction value; If the second trend prediction value is different from the enterprise real-time dynamic information data, return to the step of adjusting the parameters of the first time series prediction model according to the first prediction error until all the stationary series data are traversed to obtain the third trend prediction value.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images