Search

CN-122027501-A - Digital economic network flow prediction method and system based on artificial intelligence

CN122027501ACN 122027501 ACN122027501 ACN 122027501ACN-122027501-A

Abstract

The invention discloses a digital economic network flow prediction method and a system based on artificial intelligence, which relate to the technical field of digital economic monitoring and comprise the steps of acquiring historical network flow time sequence data of a plurality of digital economic entity nodes, corresponding digital economic characteristic data and economic index data aligned with the historical network flow time sequence data, constructing a dynamic adjacency graph evolving along with time based on the digital economic characteristic data, the economic index data and the historical network flow time sequence data, processing the historical network flow time sequence data through a multi-scale time encoder to obtain node time characterization containing multi-granularity time dependent characteristics, merging the node time characterization with the digital economic characteristic data, carrying out graph convolution aggregation by combining the dynamic adjacency graph, introducing a joint attention mechanism for generating bias items based on economic index vectors, generating node characterization of merging space-time and economic semantics, and carrying out initial flow prediction based on the node characterization of merging the space-time and the economic semantics.

Inventors

  • SONG HONGJUN
  • MA CHUANXIANG

Assignees

  • 中邮建技术有限公司

Dates

Publication Date
20260512
Application Date
20260312

Claims (10)

  1. 1. The artificial intelligence-based digital economic network flow prediction method is characterized by comprising the following steps: acquiring historical network flow time sequence data of a plurality of digital economic entity nodes, corresponding digital economic characteristic data and economic index data aligned with the historical network flow time sequence data in time; Constructing a dynamic adjacency graph evolving with time based on the digital economic characteristic data, the economic index data and the historical network flow time sequence data; Processing the historical network flow time sequence data through a multi-scale time encoder to obtain node time characterization containing multi-granularity time dependence characteristics; after the node time representation and the digital economic feature data are fused, carrying out graph convolution aggregation by combining a dynamic adjacency graph, introducing a joint attention mechanism for generating bias items based on economic index vectors, and generating node representation fusing space-time and economic semantics; Carrying out initial flow prediction based on node characterization fusing space-time and economic semantics, constructing a correction model based on statistical characteristics of historical residual errors in economic state segmentation according to time sequence distribution characteristics of the historical predicted residual errors in different economic states, and driving the economic sensitivity correction model to generate dynamic bias correction amount by using economic index data; and adding the dynamic bias correction to the initial flow prediction, outputting a network flow prediction result, and adopting a continuous optimization type loop-free constraint causal discovery algorithm to identify causal driving factors so as to carry out continuous flow deduction.
  2. 2. The method for predicting digital economic network traffic based on artificial intelligence according to claim 1, wherein the method for constructing dynamic adjacency graph evolving with time based on digital economic characteristic data, economic index data and historical network traffic time sequence data comprises the following specific steps: aiming at any two digital economic entity nodes, calculating the synergy degree of the two digital economic entity nodes on the basis of the historical network flow sequence by adopting a mutual information algorithm as the association strength of a data layer; the method comprises the steps of combining respective digital economic attributes of two nodes, evaluating the economic homogeneity of the nodes by a distance measurement method, wherein the more similar the economic features are, the stronger the connection tendency is, and the digital economic attributes comprise the industries, platform types and user scales; meanwhile, introducing a macroscopic economic index at the current moment, and generating a global adjustment factor by using a scenic sensing module to reflect the influence of the current global economic environment on the association between nodes, wherein the macroscopic economic index comprises digital consumption activity, platform transaction index and regional digital economic scenery; Weighting and fusing the flow cooperativity, the economic homogeneity and the macroscopic scenic state, and generating the connection weight among the nodes in the time map through normalization processing; the dynamic adjacency graph can automatically adjust the topological structure along with the change of the economic environment, thereby reflecting the dynamic interaction relationship between the digital economic entities more truly.
  3. 3. The method for predicting digital economic network traffic based on artificial intelligence according to claim 2, wherein the step of processing the historical network traffic time series data by a multi-scale time encoder to obtain a node time representation comprising multi-granularity time dependent features comprises the following specific steps: Inputting a historical network flow sequence of a node into three parallel causal expansion convolution layers, wherein the expansion rates of the three parallel causal expansion convolution layers are respectively set to be 1, 7 and 28, respectively capturing local fluctuation of an hour level, a daily level cycle rule and a Zhou Jichang period trend, and outputting three groups of characteristic diagrams; Splicing the three groups of feature graphs along the channel dimension to form a multi-scale time feature tensor; Inputting a multi-scale temporal feature tensor into a sparse self-attention layer, wherein attention is paid only at time steps meeting a log-sparse constraint Calculation, namely: ; Wherein, the For the length of the historical time window, Is a sparse control constant; applying feed-forward network and layer normalization to sparse attention output, generating node time characterization 。
  4. 4. The method for predicting digital economic network traffic based on artificial intelligence according to claim 3, wherein the introducing the joint attention mechanism modulated by the economic index data comprises the following specific steps: Node hidden state after convolution aggregation of graph And corresponding digital economic feature vector Splicing to form an enhanced representation ; Based on And (3) with Generating query vectors And key vector Dot product similarity is calculated ; Vector the current economic index Non-linear mapping Obtaining economic context vectors ; Digital economic characteristic splicing result of pair node pair Inputting into a multi-layer perceptron, extracting and embedding united economic semantics ; Will be And (3) with Multiplying the inner product by a learnable scalar Obtaining economic semantic bias term The expression is: ; the economic semantic bias term The economic attribute is interacted in a high order by the macroscopic view signal and the microscopic node instead of simple weighting, so that quantitative guidance on economy is formed; Will be And (3) with After addition, softmax is applied to obtain attention weight ; By means of Weighted summation of value vectors, and output of node characterization fusing spatio-temporal and economic semantics 。
  5. 5. The method for predicting digital economic network traffic based on artificial intelligence according to claim 4, wherein the method for constructing the economic sensitivity correction model according to the time sequence distribution characteristics of the historical prediction residual errors in different economic states comprises the following specific steps: Continuously recording each moment in training phase Prediction residual of (2) Corresponding economic index vector ; Will be Projected as scalar scenic index Dividing the scene index into three economic states of low, medium and high according to the quantiles of 33% and 66% of the scene index; statistical mean shift of residual sequences under each class of state And variance of ; When on-line prediction is carried out, calculating the similarity degree between the current scenic index and the centers of the three states, and carrying out weighted fusion on the mean shift under various states by adopting a Gaussian kernel function to obtain a dynamic bias correction quantity; and finally, superposing the dynamic bias correction amount to an initial predicted value to generate a corrected result.
  6. 6. The method for predicting digital economic network traffic based on artificial intelligence according to claim 5, wherein the step of adding the dynamic bias correction to the initial traffic prediction to output the network traffic prediction result comprises the following steps: Calculating a corrected predicted value ; Synchronous acquisition of prediction variance Constructing uncertainty intervals ; Judging whether the interval width exceeds the adaptive threshold If the current value exceeds the current value, triggering on-line fine adjustment; The adaptive threshold Dynamically adjusting according to recent economic fluctuation intensity by calculating variance of economic index vectors of the last N time steps Quantifying instability of an external environment, taking variance as an increment proportion for adjusting the size of a threshold value, improving tolerance of the model when economy fluctuates severely, tightening sensitivity in a stationary period, and forming an adaptive maintenance mechanism dynamically coupled with an external system; When the economic index is severely oscillated, the tolerance of the model is automatically improved, so that frequent training due to normal disturbance is avoided; conversely, the threshold is tightened in the stationary phase, the sensitivity to abnormal signals is enhanced, and an adaptive maintenance mechanism dynamically coupled with an external system is formed.
  7. 7. The method for predicting digital economic network traffic based on artificial intelligence according to claim 6, wherein the causal discovery algorithm identifies causal driving factors from digital economic characteristic data and economic index data, and performs traffic deduction by the specific steps of: Organizing historical observation data into structured matrices ; Solving directed acyclic graph adjacency matrix using continuous optimization causal discovery algorithm The objective function comprises a data reconstruction term and an acyclic constraint term, and the expression is: ; Objective function pass trace index term Accurately delineate the acyclic nature only The term is zero when the directed acyclic graph is corresponding, so that the combination optimization problem is converted into continuous micro-optimizable, and the large-scale variable causal structure learning is possible; From the slave Extracting non-zero father node index set pointing to target flow variable As a causal driving factor; fixing during the deduction of the counterfactual The intermediate variable takes the value as an intervention value, the other variables remain unchanged, and the intermediate variable is substituted into the forward calculation of the trained model to obtain the inverse fact flow prediction result ; The causal deduction mechanism described above is based on a true causal structure, as distinguished from approximate intervention simulations based on relevance or on grange causality.
  8. 8. An artificial intelligence based digital economic network traffic prediction system, based on the artificial intelligence based digital economic network traffic prediction method according to any one of claims 1 to 7, characterized by comprising: The system comprises a data fusion module, a dynamic diagram construction module, a time sequence coding module, an economic perception aggregation module, a residual error correction module and a causal deduction module; The data fusion module is used for acquiring historical network flow time sequence data of a plurality of digital economic entity nodes, corresponding digital economic characteristic data and macroscopic economic index data aligned with the historical network flow time, and carrying out alignment and pretreatment; The dynamic graph construction module is used for constructing a dynamic adjacency graph evolving along with time based on flow cooperativity, node economic homogeneity and current macroscopic scenic state, so that the graph structure can adaptively reflect dynamic association among digital economic entities; The time sequence coding module is used for extracting multi-granularity time characterization containing hour level, day level and week level dependence from the historical flow sequence through multi-scale causal expansion convolution and a sparse self-attention mechanism; the economic perception aggregation module is used for fusing node time representation with digital economic characteristics, carrying out graph convolution by combining a dynamic adjacency graph, introducing a joint attention mechanism modulated by economic indexes, and generating node representation fusing space-time and economic semantics; The residual error correction module is used for constructing an economic sensitivity correction model according to the statistical distribution characteristics of the historical predicted residual errors in different economic states, generating a dynamic bias correction amount by using the current economic index, and carrying out self-adaptive correction on an initial predicted result; The causal deduction module is used for identifying causal driving factors affecting the flow from the multi-source data by adopting a continuous optimization type acyclic constraint causal discovery algorithm and supporting continuous flow deduction under the condition of counterfactual intervention.
  9. 9. The computer equipment comprises a memory and a processor, wherein the memory stores a computer program, and the computer program is characterized in that the processor realizes the steps of the artificial intelligence-based digital economic network flow prediction method according to any one of claims 1 to 7 when executing the computer program.
  10. 10. A computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor implements the steps of the artificial intelligence based digital economic network traffic prediction method according to any one of claims 1 to 7.

Description

Digital economic network flow prediction method and system based on artificial intelligence Technical Field The invention relates to the technical field of digital economic monitoring, in particular to a digital economic network flow prediction method and system based on artificial intelligence. Background The digital economic monitoring technology is a technical system for carrying out dynamic sensing, quantitative evaluation and trend prejudgment on the digital economic operation situation by collecting, integrating and analyzing multi-source real-time data generated by a digital platform, a network infrastructure and an economic main body and combining macroscopic economic indexes and digital industrial characteristics. The method has the core aims of realizing observability, measurability and early warning of digital industrialization and industrial digital processes and providing data support and intelligent research and judgment basis for government supervision, enterprise decision making and policy making. Therefore, how to improve the intelligentization level and the safety of digital economic monitoring by using advanced technical means is one of the problems to be solved in the current urgent need. In the existing analysis system, the sampling frequency of the network flow time sequence data is often inconsistent with that of the macro economic index, the time granularity is not matched, so that an accurate time sequence corresponding relation cannot be established, the reliability of causal inference and dynamic association analysis is affected. Disclosure of Invention The present invention has been made in view of the above-described problems occurring in the prior art. Therefore, the invention provides a digital economic network flow prediction method based on artificial intelligence, which solves the problems that in the existing analysis system, the network flow time sequence data and the macroscopic economic index are often inconsistent in sampling frequency and unmatched in time granularity, so that an accurate time sequence corresponding relation cannot be established, the reliability of causal inference and dynamic association analysis is affected, in addition, in the prior art, a statistical method with high static or hysteresis is mostly adopted, the rapid change of digital economic activity is difficult to capture in real time, prospective prediction of economic trend cannot be realized, and the timeliness of policy formulation and market response is restricted. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the present invention provides an artificial intelligence based digital economic network traffic prediction method comprising: acquiring historical network flow time sequence data of a plurality of digital economic entity nodes, corresponding digital economic characteristic data and economic index data aligned with the historical network flow time sequence data in time; Constructing a dynamic adjacency graph evolving with time based on the digital economic characteristic data, the economic index data and the historical network flow time sequence data; Processing the historical network flow time sequence data through a multi-scale time encoder to obtain node time characterization containing multi-granularity time dependence characteristics; after the node time representation and the digital economic feature data are fused, carrying out graph convolution aggregation by combining a dynamic adjacency graph, introducing a joint attention mechanism for generating bias items based on economic index vectors, and generating node representation fusing space-time and economic semantics; Carrying out initial flow prediction based on node characterization fusing space-time and economic semantics, constructing a correction model based on statistical characteristics of historical residual errors in economic state segmentation according to time sequence distribution characteristics of the historical predicted residual errors in different economic states, and driving the economic sensitivity correction model to generate dynamic bias correction amount by using economic index data; and adding the dynamic bias correction to the initial flow prediction, outputting a network flow prediction result, and adopting a continuous optimization type loop-free constraint causal discovery algorithm to identify causal driving factors so as to carry out continuous flow deduction. The invention relates to a digital economic network flow prediction method based on artificial intelligence, which is a preferable scheme, wherein the method is based on digital economic characteristic data, economic index data and historical network flow time sequence data to construct a dynamic adjacency graph evolving along with time, and comprises the following specific steps: aiming at any two digital economic entity nodes, calculating the synergy degree of the two d