Search

CN-122022897-A - Big data-based intelligent E-commerce prediction management method and system

CN122022897ACN 122022897 ACN122022897 ACN 122022897ACN-122022897-A

Abstract

The invention relates to the technical field of electronic commerce, in particular to an electronic commerce intelligent prediction management method and system based on big data, comprising the following steps: the method comprises the steps of obtaining a real-time message of a buffer area, generating a throughput rate sequence based on a sliding window, performing difference and mapping on throughput rates of adjacent periods to construct a gradient slope sequence, extracting transient mutation ratios by means of a moving average model to generate a pre-trigger grade mark, extracting set boundary extremum quantization evaluation to generate discrete parameters, correcting a transfer matrix, and performing steady-state iterative calculation to obtain prediction confidence coefficient to construct an elastic management and control instruction. According to the invention, through accurately capturing the abrupt change trend of the time domain dimension of the service and eliminating the static reference deviation interference, the deep analysis and the real-time correction of the dynamic transfer characteristic of the core transaction link are realized, the dilemma of research and judgment distortion and scheduling hysteresis existing in the conventional rule facing the sudden flow is thoroughly avoided, and the full-link prediction precision and the elastic resource management and control efficiency in the complex transaction scene are obviously improved.

Inventors

  • ZHANG ZELAN
  • YAO SHUXIANG

Assignees

  • 厦门软件职业技术学院

Dates

Publication Date
20260512
Application Date
20260414

Claims (10)

  1. 1. The intelligent E-commerce prediction management method based on big data is characterized by comprising the following steps of: S1, acquiring a real-time order message in a streaming data buffer area of an electronic commerce platform, analyzing and extracting an order time stamp and an order type identifier, executing time span segmentation based on a sliding window algorithm, and carrying out classification statistics by combining the order type identifier to generate an order throughput rate sequence; s2, acquiring the order throughput rate sequence, performing differential operation on the order throughput rates of adjacent sampling periods to extract throughput rate difference and performing time domain change rate mapping to construct a throughput rate gradient slope sequence; S3, acquiring the throughput rate gradient slope sequence, inputting the throughput rate gradient slope sequence into a weighted moving average model, performing baseline deviation analysis to extract transient mutation ratio values, and performing matching screening based on a preset trigger threshold interval to generate a peak pre-trigger grade identifier; S4, acquiring the peak pre-trigger level mark to construct a level aggregation set, extracting the frequency highest mark as a period leading trigger level, extracting the boundary extremum of the level aggregation set, performing state span quantization evaluation, and generating a trigger fluctuation discrete parameter; and S5, correcting an initial state transition matrix constructed based on a preset full-link log by utilizing the trigger fluctuation discrete parameter, performing steady-state probability distribution iterative calculation, obtaining flow prediction confidence coefficient, extracting a corresponding elasticity quota parameter, and constructing an E-commerce prediction management and control instruction.
  2. 2. The big data based e-commerce intelligent prediction management method of claim 1, wherein the order throughput rate sequence comprises a household appliance order throughput rate, a clothing order throughput rate and a digital order throughput rate, the throughput rate gradient slope sequence comprises a first order difference value, a second order derivative and a change rate, the peak pre-trigger level identification comprises a normal release identification, a current limiting queuing identification and a fusing degradation identification, the trigger fluctuation discrete parameters comprise a total variance, a sample standard deviation and a quarter bit distance, and the e-commerce prediction management instruction comprises an elastic telescoping instruction, a token bucket current limiting instruction and a service degradation instruction.
  3. 3. The big data-based e-commerce intelligent prediction management method of claim 1, wherein the specific steps of S1 are as follows: S101, acquiring a real-time service message of a business platform buffer zone, extracting an order time stamp and an order type identifier in a message header node field, acquiring a time stamp representation value and an identifier character code, and performing association by taking the time stamp representation value as a main key and the identifier character code as a value to generate a temporal feature mapping table; S102, calling the temporal characteristic mapping table, acquiring window stepping parameters and span tolerance parameters, comparing the timestamp characterization values with the span tolerance parameters, screening data items falling within the limit range of the span tolerance parameters, and repeatedly screening according to the transition range of the window stepping parameters to acquire a time domain segmentation state set; And S103, performing classification operation based on the time domain segmentation state set, extracting intra-interval identification character codes, combining similar data items, summing the numbers to generate interval order classification total numbers, combining the interval order classification total numbers into a one-dimensional array according to time lapse sequences, and establishing an order throughput rate sequence.
  4. 4. The big data-based e-commerce intelligent prediction management method of claim 3, wherein the specific steps of S2 are as follows: s201, calling the order throughput rate sequence, sequentially reading the throughput rates corresponding to discrete nodes in the sequence, extracting the values of two adjacent nodes according to the time arrangement sequence, performing differential operation to obtain throughput rate difference, arranging the difference into a one-dimensional linear array structure according to the time-lapse unidirectional sequence, and establishing a periodic throughput difference set; s202, acquiring a preset sampling duration parameter based on the periodic throughput differential set, performing division operation on a plurality of differential values in the periodic throughput differential set and the sampling duration parameter, calculating the internal change rate of a sampling interval, and sequentially splicing the change rates according to the node sequence of throughput differential values to obtain a time domain change mapping column; s203, setting a gradient slope plane reference space for the time domain change mapping column, extracting the change rate in the time domain change mapping column to serve as a longitudinal coordinate scalar, taking the relative displacement step length of the corresponding node as a transverse coordinate scalar, constructing a two-dimensional coordinate mapping point, connecting the points along a time axis, and generating a throughput rate gradient slope sequence.
  5. 5. The big data-based e-commerce intelligent prediction management method of claim 4, wherein the specific steps of S3 are as follows: S301, acquiring the throughput gradient slope sequence, extracting node original slope magnitude, reading a sliding time window span variable and node attenuation weight factors, multiplying and summing the original slope magnitude and the node attenuation weight factors in the sliding time window span variable, dividing the multiplied and summed value by the node attenuation weight factors, and establishing a dynamic weighting reference uniform line column; S302, calling the dynamic weighting reference average line column, subtracting a reference mean value in the dynamic weighting reference average line column from an original slope value at a synchronous sequence position, extracting a time domain absolute deviation parameter, dividing the time domain absolute deviation parameter by a corresponding reference mean value, extracting a forward mutation item, and generating a transient mutation ratio set; S303, for the transient mutation ratio set, the judgment parameters of the upper and lower limiting degrees in the preset trigger state threshold value interval set are called, the judgment ratio falls into the target state threshold value interval, mapping classification characters of the target state threshold value interval are read, and the classification characters are spliced in sequence, so that the peak pre-trigger level identification is obtained.
  6. 6. The intelligent predictive management method of electronic commerce based on big data according to claim 5, wherein the determining that the ratio falls within the target state threshold interval means that a plurality of continuous value determining intervals are established based on the upper and lower limiting degree determining parameters, ratio values in the transient mutation ratio set are respectively compared with a lower limit boundary and an upper limit boundary in the value determining intervals, the value determining intervals in which the ratio values are within the lower limit boundary and the upper limit boundary are extracted, and the target state threshold interval is set; the trigger state threshold interval set is used for calculating a data expected mean value and a standard deviation parameter of the historical mutation ratio sample set by extracting the historical mutation ratio sample set, multiplying the standard deviation parameter and a fixed proportionality constant to obtain a tolerance range parameter, respectively accumulating and deducting the tolerance range parameter on the basis of the data expected mean value to obtain a fluctuation upper limit boundary and a fluctuation lower limit boundary, equally dividing the span between the fluctuation lower limit boundary and the fluctuation upper limit boundary, extracting a dividing node value as an upper limiting degree judgment parameter and a lower limiting degree judgment parameter, pairing and combining the adjacent upper limiting degree judgment parameters and the adjacent lower limiting degree judgment parameter to establish a value judgment interval, and integrating the value judgment interval to determine.
  7. 7. The big data-based e-commerce intelligent prediction management method of claim 5, wherein the specific steps of S4 are as follows: s401, acquiring the peak pre-trigger grade identification, extracting an aggregation period parameter, collecting the peak pre-trigger grade identification in an aggregation period parameter time span, counting and recording the occurrence number parameter, associating each grade identification with the occurrence number parameter, and establishing a grade frequency mapping set; S402, traversing the appearance frequency parameters by calling the grade frequency mapping set, extracting a maximum value corresponding mark as a period leading triggering grade, retrieving upper and lower limit boundary marks in the grade frequency mapping set, extracting a preset dangerous quantization scale, and converting the upper and lower limit boundary marks by combining the dangerous quantization scale to obtain a boundary extremum state vector; s403, extracting a distribution upper limit value and a distribution lower limit value according to the boundary extremum state vector, subtracting the distribution lower limit value from the distribution upper limit value to extract a state span difference value, reading a dominant quantization value of a period dominant trigger level, dividing the state span difference value by the dominant quantization value, and generating a trigger fluctuation discrete parameter.
  8. 8. The big data-based e-commerce intelligent prediction management method of claim 7, wherein the specific steps of S5 are as follows: s501, acquiring the trigger fluctuation discrete parameter, extracting a page jump sequence in a preset full-link log, counting the jump frequency among page nodes in the page jump sequence, dividing the jump frequency by the total outflow frequency of the corresponding source page nodes, calculating the initial transition probability among the nodes, and constructing an initial state transition matrix; S502, calling the initial state transition matrix, extracting a preset core transaction link node as a target correction object, executing nonlinear amplification mapping processing on transition probability corresponding to the target correction object according to a trigger fluctuation discrete parameter, and executing corresponding deduction and line vector normalization processing on transition probability of a non-core transaction link node to generate a dynamic transition feature matrix; S503, calling the dynamic transfer feature matrix, acquiring a real-time online user distribution vector as an initial state vector, executing continuous iterative multiplication calculation, extracting a converged steady-state probability vector as a flow prediction confidence coefficient, matching a quota association matrix, extracting a corresponding elasticity quota parameter, and constructing an E-commerce prediction management and control instruction.
  9. 9. The big data-based e-commerce intelligent prediction management method of claim 8, wherein the performing nonlinear amplification mapping processing on transition probabilities corresponding to target correction objects according to trigger fluctuation discrete parameters, performing corresponding deduction and line vector normalization processing on transition probabilities of non-core transaction link nodes refers to performing division with a constant 1 as a dividend and trigger fluctuation discrete parameters as divisors, extracting reciprocal as a mapping index parameter, performing power function operation with the transition probabilities directed to core transaction link nodes in an initial state transition matrix as a base, taking the mapping index parameter as an exponent, obtaining amplified correction core probabilities, extracting original transition probabilities of all non-core transaction link nodes in the same line, performing running-in summation with correction core probabilities to obtain amplified probability sums, taking the corrected core probabilities and the original transition probabilities of all non-core transaction link nodes as dividends respectively, performing division normalization scaling with the amplified probability sums as divisors, and updating probability values of current lines.
  10. 10. The big data-based e-commerce intelligent prediction management system is characterized in that the system is used for realizing the big data-based e-commerce intelligent prediction management method as claimed in any one of claims 1 to 9, and the system comprises: The order stream aggregation module is used for acquiring real-time order messages in the e-commerce platform stream data buffer zone, analyzing and extracting order time stamps and order type identifiers, executing time span segmentation based on a sliding window algorithm, and carrying out classification statistics in combination with the order type identifiers to generate an order throughput rate sequence; The change rate extraction module is used for obtaining the order throughput rate sequence, performing differential operation on the order throughput rates of adjacent sampling periods to extract throughput rate difference and performing time domain change rate mapping to construct a throughput rate gradient slope sequence; The flow mutation perception module is used for acquiring the throughput rate gradient slope sequence, inputting the throughput rate gradient slope sequence into a weighted moving average model, executing baseline deviation analysis to extract transient mutation ratio values, executing matching screening based on a preset trigger threshold interval, and generating a peak pre-trigger level identifier; The level quantitative evaluation module is used for acquiring the peak pre-trigger level identification to construct a level aggregation set, extracting the frequency highest identification as a period leading trigger level, extracting the boundary extremum of the level aggregation set, performing state span quantitative evaluation and generating a trigger fluctuation discrete parameter; And the quota elastic management and control module corrects an initial state transition matrix constructed based on a preset full-link log by utilizing the trigger fluctuation discrete parameter, performs steady-state probability distribution iterative calculation, acquires flow prediction confidence coefficient, extracts a corresponding elasticity quota parameter and constructs an E-commerce prediction management and control instruction.

Description

Big data-based intelligent E-commerce prediction management method and system Technical Field The invention relates to the technical field of electronic commerce, in particular to an electronic commerce intelligent prediction management method and system based on big data. Background The technical field of electronic commerce mainly relates to core matters such as commodity information display transaction matching order processing payment settlement logistics distribution, user behavior analysis and the like, which cover a plurality of links such as data acquisition, such as user browsing, recording of click track purchase history evaluation information, data storage, such as distributed database and data warehouse, data processing, such as cleaning integration and feature construction, business application, such as commodity recommendation inventory management price adjustment, marketing strategy formulation and the like, and integrally form an online transaction and operation management system driven by data. The traditional intelligent forecasting and management method for the electronic commerce comprises the steps of summarizing technical matters such as inventory turnover, sales trend and the like of purchasing trend of a commodity demand change user through a pointer, counting historical order records according to time sequence, generating time sequence data in a fixed period dividing mode by combining holiday mark promotion activity records and commodity classification labels, estimating future sales through average value calculation sliding window accumulation or simple regression fitting, judging inventory replenishment time and quantity according to a manually set threshold value, and grading the access times and purchasing frequency of the user through a preset rule table so as to predict and manage the commodity sales and operation rhythms. In the prior art, generally, historical order records are summarized according to time sequence, time series data are generated in a fixed period dividing mode, future sales volume is estimated by means of mean value calculation or simple regression fitting, meanwhile, the replenishment volume is judged according to a manually set threshold value, the user access purchasing frequency is classified based on a preset fixed rule, the operation mode is limited by a fixed time span and a static evaluation rule, burst flow characteristics and dynamic demand changes are difficult to accurately capture, serious hysteresis exists between low flow prediction precision and resource scheduling, and traffic processing congestion or system resource idle waste is extremely easy to cause. Disclosure of Invention In order to solve the technical problems that in the prior art, a time series data is generated by adopting a fixed period dividing mode aiming at historical order records in time sequence summarization, future sales volume is estimated by means of mean value calculation or simple regression fitting, meanwhile, the replenishment volume is judged according to a manually set threshold value, the user access purchasing frequency is classified based on a preset fixed rule, the operation mode is limited by a fixed time span and a static evaluation rule, sudden flow characteristics and dynamic demand changes are difficult to accurately capture, the flow prediction precision is low, serious hysteresis exists in resource scheduling, and traffic processing congestion or idle waste of system resources is extremely easy to cause. In order to achieve the purpose, the invention adopts the intelligent E-commerce prediction management method based on big data, which comprises the following steps: S1, acquiring a real-time order message in a streaming data buffer area of an electronic commerce platform, analyzing and extracting an order time stamp and an order type identifier, executing time span segmentation based on a sliding window algorithm, and carrying out classification statistics by combining the order type identifier to generate an order throughput rate sequence; s2, acquiring the order throughput rate sequence, performing differential operation on the order throughput rates of adjacent sampling periods to extract throughput rate difference and performing time domain change rate mapping to construct a throughput rate gradient slope sequence; S3, acquiring the throughput rate gradient slope sequence, inputting the throughput rate gradient slope sequence into a weighted moving average model, performing baseline deviation analysis to extract transient mutation ratio values, and performing matching screening based on a preset trigger threshold interval to generate a peak pre-trigger grade identifier; S4, acquiring the peak pre-trigger level mark to construct a level aggregation set, extracting the frequency highest mark as a period leading trigger level, extracting the boundary extremum of the level aggregation set, performing state span quantization evaluation, and generating