CN-122022891-A - Real-time prediction method for big data change trend

CN122022891ACN 122022891 ACN122022891 ACN 122022891ACN-122022891-A

Abstract

The invention relates to the field of big data, in particular to a real-time prediction method for big data change trend, which is characterized by confirming a data stream and a public opinion perceived data stream by hysteresis, analyzing a time-lag coupling relation between a historical data change position and a sales peak, calculating a dynamic delay window, constructing a parameterized mapping model, generating a predictive complement sequence based on current public opinion perceived data, constructing a panoramic data sequence by combining the confirmed data, further carrying out real-time abnormal fluctuation verification on the public opinion perceived stream, dynamically adjusting the delay window according to matching reliability to reconstruct a mapping relation, and outputting the complement sequence passing verification as a prediction result. The method and the device realize the technical effects of multi-source heterogeneous data fusion, time-lag relation self-adaptive evolution and real-time calibration of the prediction process, improve the timeliness and accuracy of trend prediction, and provide scientific data support for medicine sales and medicine management.

Inventors

Xue Lintong
YANG SHAOJIE
HUANG XIN

Assignees

北京法伯宏业科技发展有限公司

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. The real-time prediction method for the big data change trend is characterized by comprising the following steps of: Acquiring a lag confirmation data stream and a public opinion perception data stream of a target area, wherein the lag confirmation data stream is target index data from a sales data system, and the public opinion perception data stream is based on target data acquired by a target platform in real time; Analyzing the corresponding relation between the data change position of the target data and the subsequent sales peak of the target index data in the historical statistics period, and calculating a dynamic delay window between the data change position and the subsequent sales peak of the target index data to construct a time-lag coupling mapping relation; Generating a predictive complement sequence for the missing part in the latest statistical period based on the time-lag coupling mapping relation and the public opinion perception data stream in the current statistical period; splicing the predictive complement sequence with the acquired lag confirmation data stream to construct a panoramic data sequence; based on panoramic data sequence prediction, carrying out real-time abnormal fluctuation verification on the public opinion perception data stream, including, Calculating the matching reliability of the current data fluctuation mode; if the matching reliability is greater than or equal to a preset threshold value, judging that the abnormal fluctuation verification is passed; If the matching reliability is smaller than a preset threshold, adjusting the size of a dynamic delay window, reconstructing a time-lag coupling mapping relation based on the adjusted dynamic delay window, and regenerating a predictive alignment sequence based on the updated mapping relation; And outputting the predictive complement sequence passing the abnormal fluctuation verification as a trend prediction result.
2. The method for real-time prediction of big data trend according to claim 1, wherein the obtaining the lag behind the target region comprises, Obtaining structured statistics from the sales data system having a first fixed period as the lag validation data stream; Screening user release content data with continuous time stamps from an Internet platform based on a preset related keyword library to serve as the public opinion perception data stream; Wherein the real-time perceptual data update frequency is higher than the first fixed period.
3. The method for predicting trend of big data in real time according to claim 1, wherein analyzing the correspondence between the data change position of the target data and the subsequent sales peak of the target index data in the historical statistics period, calculating the dynamic delay window therebetween to construct the time-lag coupling mapping relationship comprises, Analyzing the target data sequence and the target index data sequence in a history window, and identifying the time shift amount for enabling the correlation of the two sequences to reach a peak value; Determining a baseline delay amount based on the time shift amount average; calculating a dynamic adjustment coefficient based on the fluctuation rate of the public opinion perceived data stream; Taking the product of the dynamic adjustment coefficient and the baseline delay as a dynamic delay window of a final application; And constructing a parameterized time-lag coupling mapping relation based on the corresponding relation between the two sequences of data points in the dynamic delay window.
4. The method for real-time prediction of big data trend according to claim 3, wherein calculating the dynamic adjustment coefficient based on the fluctuation rate of the public opinion perceived data stream comprises, Extracting public opinion perceived data streams in a current time period and a plurality of preset historical time periods, and calculating the fluctuation rate of the public opinion perceived data streams; Comparing the fluctuation rate measurement value with a preset historical reference fluctuation rate, and calculating the fluctuation rate deviation degree; Inputting the fluctuation rate deviation degree into a preset monotonic mapping function, and outputting a dynamic adjustment coefficient larger than 1 if the fluctuation rate deviation degree is larger than or equal to a first threshold value; outputting a dynamic adjustment coefficient smaller than 1 if the fluctuation rate deviation is smaller than or equal to a second threshold value; Outputting a dynamic adjustment coefficient equal to 1 if the fluctuation rate deviation is smaller than the first threshold and larger than the second threshold; wherein the first threshold is greater than 0 and the second threshold is less than 0.
5. The method for predicting big data trend in real time according to claim 3, wherein the parameterized time-lag coupling mapping relationship is constructed based on the correspondence between two sequences of data points in the dynamic delay window, According to the dynamic delay window, the public opinion perception characteristic sequence in the historical time sequence data is aligned with the hysteresis confirmation index sequence in time; determining a mapping function based on a plurality of groups of time-aligned public opinion perception feature sequences and hysteresis confirmation index sequence data pairs, and taking the mapping function as a time-lag coupling mapping relation; Wherein the mapping function defines a calculation relation from the public opinion perception characteristic value to the future lag confirmation index prediction value.
6. The method of claim 1, wherein generating a predictive complement sequence for missing parts in the latest statistical period based on the time-lag coupling mapping relationship and the public opinion perceived data stream in the current statistical period comprises, Extracting public opinion perceived data stream in the current statistical period, and calculating corresponding public opinion perceived characteristics according to an input form defined by the mapping function; Inputting the calculated public opinion perception characteristics into the mapping function to output a numerical value prediction result of the missing part hysteresis confirmation data stream in the latest statistics period; and taking the numerical prediction result as an estimated value of the predictive complement sequence to construct the predictive complement sequence.
7. The method for predicting trend of large data in real time as recited in claim 1, wherein said calculating the matching reliability of the current data fluctuation pattern includes, Extracting public opinion perceived data stream in a current time window, and calculating a fluctuation mode feature vector of the public opinion perceived data stream; acquiring a historical reference feature vector of a public opinion perception data stream fluctuation mode in a historical contemporaneous time window; And calculating a correlation coefficient between the current fluctuation mode feature vector and the historical reference feature vector, and taking the calculated correlation coefficient as the matching reliability.
8. The method of claim 7, wherein the obtaining the historical reference feature vector of the public opinion data stream fluctuation pattern in the historical contemporaneous time window comprises, Retrieving a plurality of historical contemporaneous time windows from a historical database; Calculating the average value of the public opinion perception data stream fluctuation mode feature vectors in the history contemporaneous time window; And taking the calculated average value as the historical reference feature vector.
9. The method for real-time prediction of big data change trend according to claim 1, wherein said adjusting the size of the dynamic delay window comprises, Calculating a window adjustment coefficient based on the matching reliability; multiplying the window adjustment coefficient with the current dynamic delay window to obtain the adjusted dynamic delay window size; Wherein the window adjustment coefficient is inversely related to the matching reliability.
10. The method for predicting trend of big data in real time according to claim 9, wherein reconstructing the time-lag coupling mapping relationship based on the adjusted dynamic delay window comprises, Using the adjusted dynamic delay window to align the public opinion perception characteristic sequence and the lag confirmation index sequence in the historical time sequence again in time; Based on the realigned public opinion perception feature sequence and lag confirmation index sequence data pair, updating parameters of the mapping function through a parameter estimation algorithm to generate an updated time-lag coupling mapping relation.

Description

Real-time prediction method for big data change trend Technical Field The invention relates to the field of big data, in particular to a real-time prediction method for big data change trend. Background The big data change trend prediction is used as a core analysis task of scenes such as smart cities, financial wind control, industrial Internet of things and the like, and aims to extract evolution rules from massive, high-dimensional and heterogeneous data streams and study and judge future trends. With the popularization of sensing equipment, mobile internet and government information systems, traffic flow, environmental quality and energy consumption in urban management are continuously generated, and a real-time perception basis required by a prediction model is formed. Under the background, various prediction methods based on time series modeling and statistical analysis have been developed, especially in the field of medical data, and can predict sales data of medicines, thereby scientifically providing decision support. For example, chinese patent publication No. CN118710194A discloses a dynamic optimizing method of medicine inventory based on deep learning, comprising the steps of S1 collecting medicine sales record data, medicine supply chain data and external environment data, S2 cleaning and standardizing the collected medicine sales record data, medicine supply chain data and external environment data, S3 establishing an inventory prediction model according to the cleaned and standardized medicine sales record data, medicine supply chain data and external environment data, and S4 capturing time series characteristics of inventory change in the inventory prediction model by using a circulating neural network and analyzing unstructured data by combining the convolutional neural network. According to the invention, through the collection of multi-source data, including the drug sales record data, the drug supply chain data and the external environment data, various factors influencing the drug inventory change can be reflected more accurately, so that the accuracy of inventory prediction is improved. However, the following problems exist in the prior art: 1. in the prior art, the time lag relation between the official monitoring data and the target data of the target platform is not considered, so that the prediction model cannot capture the signal characteristics released in advance by the public behaviors, and response lag is caused. 2. In the prior art, the dynamic influence of the change of the data fluctuation rate on the time delay window is not considered, a mapping relation is constructed by adopting fixed delay parameters, the drift change of the data association strength under different periods is difficult to adapt, and the prediction precision is attenuated along with time. 3. In the prior art, feedback calibration of a predictive result is not considered, a dynamic evaluation mechanism is lacked, and mapping relation parameters cannot be adaptively adjusted according to prediction deviation. Disclosure of Invention Therefore, the invention provides a real-time prediction method for big data change trend, which is used for solving the problems that in the prior art, the time lag relation between official monitoring data and internet target data is not considered, the dynamic influence of the fluctuation rate change of the data on a time lag window is not considered, the feedback calibration on a predictive result is not considered, the response lag caused by a dynamic evaluation mechanism is lacked, the prediction precision decays along with time, and the mapping relation parameter cannot be adaptively adjusted according to the prediction deviation. In order to achieve the above object, the present invention provides a real-time prediction method for big data change trend, comprising: Obtaining a lag confirmation data stream and a public opinion perception data stream of a target area, wherein the lag confirmation data stream is target index data from a sales data system, and the public opinion perception data stream is based on target data obtained by a target platform in real time; Analyzing the corresponding relation between the data change position of the target data and the subsequent sales peak of the target index data in the historical statistics period, and calculating a dynamic delay window between the data change position and the subsequent sales peak of the target index data to construct a time-lag coupling mapping relation; Generating a predictive complement sequence for the missing part in the latest statistical period based on the time-lag coupling mapping relation and the public opinion perception data stream in the current statistical period; splicing the predictive complement sequence with the acquired lag confirmation data stream to construct a panoramic data sequence; based on panoramic data sequence prediction, carrying out real-time abnormal fluctuation verification