Search

CN-122020495-A - Freight data management method and system based on artificial intelligence

CN122020495ACN 122020495 ACN122020495 ACN 122020495ACN-122020495-A

Abstract

The invention discloses a freight data management method and a freight data management system based on artificial intelligence, which relate to the technical field of logistics data management analysis and comprise the steps of collecting multi-source data for time correction, calculating quality scores of the multi-source data, constructing the multi-source data into fixed variables and splicing the fixed variables into vectors, calculating fusion reference values of each fixed variable to obtain data source consistency residual errors and consistency scores, fusing to obtain instant score updating fixed variable credibility, calculating fusion of the fixed variables based on the updating credibility and writing semantic knowledge graph, calculating statistics of each fixed variable and mapping the statistics into nodes to construct a variable dependency graph, constructing a window based on a time bucket, using two layers of full-connection encoders to process window values, training the two layers of full-connection encoders based on SAE loss functions, and outputting hidden variables based on the trained encoders. The invention realizes the probabilistic positioning from the abnormal phenomenon to the determinable root cause.

Inventors

  • ZHANG MAOFENG

Assignees

  • 厦门青风车软件有限公司

Dates

Publication Date
20260512
Application Date
20260415

Claims (10)

  1. 1. A freight data management method based on artificial intelligence is characterized by comprising the following steps of, Collecting multi-source data for time correction, calculating quality scores of the multi-source data, constructing the multi-source data into fixed variables and splicing the fixed variables into vectors; Calculating a fusion reference value of each fixed variable to obtain a data source consistency residual error and a consistency score, calculating a consistency mean value to obtain an instant score, updating the reliability of the fixed variable based on the instant score, calculating the fusion of the fixed variables based on the updated reliability of the fixed variable, and writing the fusion into a semantic knowledge graph; calculating statistics for each fixed variable and mapping the statistics to nodes to construct a variable dependency graph; Constructing a window based on a time bucket, processing window values by using two layers of full-connection encoders, training the two layers of full-connection encoders based on an SAE loss function, and outputting hidden variables based on the trained encoders; Constructing node characteristics based on hidden variables, calculating multi-head attention by adopting a graph Transformer to update the node characteristics and calculate the abnormal score of variable nodes, constructing an abnormal evidence subgraph according to the abnormal score and calculating a final symptom vector; defining a candidate root cause set, calculating prior probability and evidence likelihood based on the candidate root cause set, and outputting posterior probability to select the most probable root cause.
  2. 2. The method for managing freight data based on artificial intelligence according to claim 1, wherein the step of collecting multi-source data for time correction, calculating quality scores of the multi-source data, constructing the multi-source data into fixed variables and splicing the fixed variables into vectors is characterized in that a cloud deployment freight data gateway collects the multi-source data; calculating clock offsets for each data source s ; Calculating correction time of each piece of data according to clock offset And falls into a 10 second time bucket; Calculating an age-decay factor for each piece of data ; Quality scoring of data source s in time bin b based on age decay factor ; Constructing multi-source data in a time bucket as fixed variables Splicing fixed variables of multi-source data to form vector ; The data in the time barrel is deleted, the number of the continuous deletion barrels is less than or equal to G, the data of the previous time barrel is used for marking, if the number of the continuous deletion barrels is greater than G, the time barrel is marked to be empty, and a deletion mask is recorded ; Scoring the quality And miss mask Synchronous splicing to form quality vector And a miss mask vector 。
  3. 3. The method for managing freight data based on artificial intelligence according to claim 2, wherein the calculating the fusion reference value of each fixed variable obtains the consistency residual error and the consistency score of the data source, and the calculating the consistency mean value obtains the instant score, the updating the reliability of the fixed variable based on the instant score, the calculating the fusion of the fixed variable based on the reliability of the updated fixed variable, and the writing into the semantic knowledge graph is specifically the initializing the reliability Calculating a fused reference value for each fixed variable ; Computing source consistency residuals based on fused reference values Consistency score ; Calculating a consistency mean value by calculating a consistency score of a fixed variable covered by a data source s in a time bucket And score the consistency mean with the quality of time bucket b Weighted fusion to instant scoring ; EWMA-based updating of trustworthiness ; Weighting and fusing each fixed variable according to the updating credibility; And writing the weighted and fused fixed variables serving as facts into the semantic knowledge graph KG in an event form.
  4. 4. The freight data management method based on artificial intelligence according to claim 3, wherein the calculating statistics for each fixed variable and mapping the calculated statistics to a node construction variable dependency graph comprises dividing all the fixed variables into 5 groups including power oil consumption, brake chassis, tire vibration, carriage cold chain and visual text road condition scheduling, and calculating statistics for each fixed variable i; concatenating statistics to form an input vector And input vector Input a layer of transducer encoder to map and obtain node ; According to the node Calculating cosine similarity and introducing a reliability coefficient; And establishing a directed adjacency building variable dependency graph A for the Top-K with the maximum node selection.
  5. 5. The method for artificial intelligence based freight data management as defined in claim 4, wherein the time bucket based window is constructed and window values are processed using a two-layer fully connected encoder, the two-layer fully connected encoder is trained based on an SAE loss function, and the trained encoder output hidden variables are specifically time bucket based vectors Window of construction length w ; From window Window value of extracting fixed variable i Window value using a two-layer fully concatenated encoder Mapping to hidden vectors ; The SAE loss function is used to train the two-layer full-concatenated encoder and the hidden variable is output by the trained full-concatenated encoder.
  6. 6. The method for managing freight data based on artificial intelligence according to claim 5, wherein the constructing node features based on hidden variables uses graph converters to calculate multi-headed attention to update node features and calculate anomaly scores of variable nodes is characterized by integrating hidden variables of all fixed vectors Stacking to obtain a sparse hidden representation matrix h, and stacking nodes Stacking to obtain variable time embedding v, linearly projecting the sparse hidden representation matrix h, and then splicing the sparse hidden representation matrix h with the variable time embedding v to obtain node initial characteristics z, and defining edge characteristics according to directed edges ; Calculating multi-head attention by adopting a graph Transformer to update node characteristics; node characteristics to be updated And (3) with Performing element-by-element multiplication and then inputting the linear layer to obtain a prediction vector Calculating a prediction vector and an actual vector Prediction loss between; Obtaining a total loss function according to the predicted loss and the SAE loss function weighting, training the graph converter parameters based on the total loss function, and outputting a predicted vector based on the trained graph converter; calculating anomaly scores for variable nodes based on predictive vectors 。
  7. 7. The method for artificial intelligence based shipment data management as set forth in claim 6, wherein constructing the anomaly evidence subgraph from the anomaly scores and computing the final symptom vector is performed by selecting the first 5 variables in descending order of anomaly scores For a pair of Each node i of the list selects a directed neighbor set Forming evidence edge sets and constructing abnormal evidence subgraphs ; Defining an abnormal evidence subgraph Symptom intensity of each variable i in (a) ; Defining propagation intensity for each evidence edge ; The symptom intensity and the propagation intensity are formed into a final symptom vector F.
  8. 8. The method for managing freight data based on artificial intelligence according to claim 1, wherein a candidate root cause set is defined, and the candidate root cause set is manually formulated and fixed as a root cause template library.
  9. 9. The method for managing freight data based on artificial intelligence according to claim 8, wherein the step of calculating prior probability and evidence likelihood based on the candidate root cause set and outputting a posterior probability to select the most probable root cause comprises obtaining prior of the root cause by historical statistics for each root cause r in the candidate root cause combination ; At the same time calculate evidence likelihood from root cause ; Based on priori And evidence likelihood Output posterior probability ; According to posterior probability The root cause with the highest probability is selected as the most probable root cause output.
  10. 10. An artificial intelligence-based freight data management system, based on the artificial intelligence-based freight data management method as set forth in any one of claims 1 to 9, characterized by comprising, The data acquisition module is used for acquiring multi-source data to perform time correction, calculating quality scores of the multi-source data, constructing the multi-source data into fixed variables and splicing the fixed variables into vectors; The credibility analysis module is used for calculating a fusion reference value of each fixed variable to obtain a data source consistency residual error and a consistency score, calculating a consistency mean value to obtain an instant score, updating the credibility of the fixed variable based on the instant score, calculating the fusion of the fixed variables based on the updated credibility of the fixed variable, and writing the fusion reference value into a semantic knowledge graph; The dependency graph construction module is used for calculating statistics for each fixed variable and mapping the statistics to nodes to construct a variable dependency graph; The variable mapping module is used for constructing a window based on a time bucket, processing window values by using the two-layer full-connection encoder, training the two-layer full-connection encoder based on an SAE loss function, and outputting hidden variables based on the trained encoder; The node anomaly analysis module is used for constructing node characteristics based on hidden variables, calculating multi-head attention by adopting a graph Transformer to update the node characteristics and calculate anomaly scores of variable nodes, constructing an anomaly evidence subgraph according to the anomaly scores and calculating final symptom vectors; and the root cause analysis module is used for defining a candidate root cause set, calculating prior probability and evidence likelihood based on the candidate root cause set, and outputting posterior probability to select the most probable root cause.

Description

Freight data management method and system based on artificial intelligence Technical Field The invention relates to the technical field of logistics data management and analysis, in particular to a freight data management method and system based on artificial intelligence. Background In recent years, along with the rapid popularization of the Internet of vehicles (V2X), the mobile Internet and an intelligent logistics platform, a large-scale and multi-source heterogeneous data system is formed by freight enterprises in links of vehicle operation, cargo state, order transfer, station operation and the like, wherein the data system comprises continuous sensing data such as vehicle-mounted terminals, OBD/CAN, GPS, tire pressure/temperature and humidity and the like, and unstructured data such as electronic freight notes, loading and unloading codes, handover signing, abnormal handling records, video/image and text logs and the like. Meanwhile, the development of cloud edge cooperative computing and machine learning technology enables real-time perception and management of in-transit risks to gradually evolve from 'post statistics' to 'on-line monitoring-abnormal alarming-tracing treatment-responsibility tracing' in a closed-loop paradigm. In order to support the paradigm, the industry introduces methods of time alignment, missing repair, data cleaning and quality assessment at a data layer, explores methods of unsupervised anomaly detection, graphic neural network and knowledge graph at a model layer to improve adaptability and interpretation of complex transportation scenes (multi-line, multi-vehicle, multi-carrier and multi-environment disturbance), and especially faces to data management requirements of cross-system collaboration, the reasoning capacity of semantic data organization and relationship structure-based reasoning gradually becomes an important direction of freight data management, however, the prior related technology still has defects, the prior knowledge graph scheme often deviates from a static entity relationship modeling, high-frequency writing mechanism and evidence-level confidence expression of on-road state events are lacked, so that numerical symptoms-business semantics-evidence sources are difficult to be communicated when anomalies occur, and interpretation and audit capacity are insufficient. 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 freight data management method and system based on artificial intelligence, which solve the problems of the prior art that the high-frequency writing mechanism and the evidence-level confidence expression of the in-transit state event are lack, and the interpretation and audit capabilities are insufficient. 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 shipping data management method, comprising, Collecting multi-source data for time correction, calculating quality scores of the multi-source data, constructing the multi-source data into fixed variables and splicing the fixed variables into vectors; Calculating a fusion reference value of each fixed variable to obtain a data source consistency residual error and a consistency score, calculating a consistency mean value to obtain an instant score, updating the reliability of the fixed variable based on the instant score, calculating the fusion of the fixed variables based on the updated reliability of the fixed variable, and writing the fusion into a semantic knowledge graph; calculating statistics for each fixed variable and mapping the statistics to nodes to construct a variable dependency graph; Constructing a window based on a time bucket, processing window values by using two layers of full-connection encoders, training the two layers of full-connection encoders based on an SAE loss function, and outputting hidden variables based on the trained encoders; Constructing node characteristics based on hidden variables, calculating multi-head attention by adopting a graph Transformer to update the node characteristics and calculate the abnormal score of variable nodes, constructing an abnormal evidence subgraph according to the abnormal score and calculating a final symptom vector; defining a candidate root cause set, calculating prior probability and evidence likelihood based on the candidate root cause set, and outputting posterior probability to select the most probable root cause. The invention relates to an optimal scheme of a freight data management method based on artificial intelligence, wherein the acquisition of multi-source data is performed with time correction, the quality score of the multi-source data is calculated, the multi-source data is constructed as fixed variables and spliced into vectors, and the cloud deployment of a freight data g