CN-121981755-A - Vehicle exit analysis method and system based on machine learning and big data

CN121981755ACN 121981755 ACN121981755 ACN 121981755ACN-121981755-A

Abstract

The invention relates to the technical field of vehicle outlet analysis and discloses a vehicle outlet analysis method and system based on machine learning and big data, wherein the method comprises the steps of constructing a multi-source data acquisition system, acquiring multi-source data related to a vehicle outlet, and preprocessing to obtain a standard analysis data set; constructing a multidimensional association knowledge graph, carrying out association mining to obtain a plurality of conduction paths of each core index, determining a plurality of weighted feature sets according to all the conduction paths, training a vehicle outlet analysis model according to all the weighted feature sets and an integrated learning algorithm, analyzing data acquired in real time by using the trained vehicle outlet analysis model, outputting a prediction result of a vehicle outlet trend, generating a vehicle outlet decision according to the prediction result and a pre-constructed decision rule base, and improving the prediction accuracy of the key index by effectively fusing and relation mining of multi-source heterogeneous data so as to realize accurate prediction and decision support of the vehicle outlet trend.

Inventors

WU SHOUXI
LIU YAN
ZHANG FAN
WANG HAIYANG
KANG KAI
FENG QIANLONG
LIU DECHENG

Assignees

中汽信息科技(天津)有限公司

Dates

Publication Date: 20260505
Application Date: 20260128

Claims (10)

1. A machine learning and big data based vehicle exit analysis method, comprising: constructing a multi-source data acquisition system, acquiring multi-source data related to a vehicle exit, and preprocessing to obtain a standard analysis data set; constructing a multidimensional association knowledge graph based on a standard analysis data set, carrying out association mining to obtain a plurality of preset conduction paths of each core index, and determining a plurality of weighted feature sets according to all the conduction paths; training a vehicle outlet analysis model according to all the weighted feature sets and an ensemble learning algorithm, analyzing data acquired in real time by using the trained vehicle outlet analysis model, and outputting a prediction result of the vehicle outlet trend; and generating a vehicle outlet decision according to the prediction result and a pre-constructed decision rule base.
2. The machine learning and big data based vehicle exit analysis method of claim 1, wherein obtaining a standard analysis dataset comprises: the multi-source data acquisition system comprises a plurality of data acquisition networks with dimensions, and each data acquisition network corresponds to a preset acquisition time interval; Preprocessing multi-source data acquired by a data acquisition network of each dimension in a preset time interval, and constructing a plurality of data sequences of each dimension according to the preprocessed data and the acquisition sequence; wherein the pretreatment comprises cleaning, de-duplication, format conversion and standardization treatment; setting corresponding check nodes according to the fluctuation degree of each data sequence, and carrying out quality check on each data sequence by combining with a preset check rule to obtain the quality coefficient of each data sequence; judging whether to carry out supplementary correction on the corresponding data sequence according to the quality coefficient, if not, incorporating the corresponding data sequence into a standard analysis data set, and if so, incorporating the data after the supplementary correction is qualified into the standard analysis data set in sequence; The standard analysis dataset comprises several data sequences of all dimensions.
3. The machine learning and big data based vehicle exit analysis method of claim 2, wherein constructing a multidimensional associated knowledge-graph based on a standard analysis dataset includes: Performing entity identification and extraction on the standard analysis data set to obtain four-dimensional entity nodes and corresponding attribute information, wherein the four-dimensional entity comprises market entity nodes, product entity nodes, policy entity nodes and competition entity nodes; Carrying out semantic association among entity nodes based on the influence relationship, the countermeasure relationship, the demand relationship and the conduction relationship, and constructing a multidimensional association knowledge graph according to semantic association results, attribute information and a graph neural network technology; The map comprises a plurality of entity nodes and edges, wherein the edges are association relations among the entities and corresponding weight values.
4. The machine learning and big data based vehicle exit analysis method of claim 3, wherein obtaining a plurality of conductive paths for each core index set in advance comprises: presetting a plurality of core indexes of vehicle outlet analysis; Screening out entity nodes directly associated with each core index from the multidimensional associated knowledge graph as a starting node set; randomly selecting one entity node in the initial node set as a target node, traversing the entity nodes in the knowledge graph from high to low according to the weight value of the edge, and generating a plurality of entity conduction links of the target node; Sequentially generating a plurality of entity conduction links of each entity node in the initial node set; Evaluating a plurality of entity conduction links of each entity node in the initial node set based on a preset path evaluation system to obtain a path evaluation value; generating a comprehensive path evaluation value of each entity conduction link according to the path evaluation value and the association coefficient of the entity node corresponding to the entity conduction link and the corresponding core index; sorting all the entity conduction links of the same core index according to the comprehensive path evaluation value, and selecting a plurality of entity conduction links corresponding to the core index by combining with the preset path proportion of the corresponding core index; And sequentially selecting a plurality of physical conduction links of each core index, and setting the physical conduction links as a plurality of conduction paths of each core index.
5. A machine learning and big data based vehicle exit analysis method as claimed in claim 3 wherein determining a plurality of weighted feature sets from all conductive paths comprises: Performing cluster analysis on all the conduction paths of the same core index to obtain a plurality of path clusters; extracting features of the same path cluster to obtain a plurality of feature entity nodes of the corresponding cluster, causal relations among the nodes and weight distribution of edges; carrying out serialization processing on attribute information of the feature entity nodes according to the sequence of the causal relationship, and constructing a feature matrix; The feature matrix comprises feature sequences of attribute information of each feature entity node in a conducting path, wherein the attribute information of each feature entity node changes along with time, and the feature sequences are sequentially arranged according to a causal relationship to form a matrix structure; the row vectors represent different time nodes, the column vectors correspond to attribute information of different characteristic entity nodes, and adjacent column vectors are connected through a causal relation direction; Carrying out standardization processing on attribute information in the feature matrix, and carrying out weighted fusion on each column of attribute information after the standardization processing based on weight distribution of edges to obtain weighted feature vectors corresponding to the path clustering clusters; The weighted feature vector comprises at least one independent variable vector, and the independent variable vector is mapped with a plurality of dependent variable vectors; constructing a weighted feature set corresponding to the core index according to weighted feature vectors of all path cluster clusters of the same core index; A number of weighted feature sets are constructed in sequence.
6. The machine learning and big data based vehicle outlet analysis method of claim 1, wherein training the vehicle outlet analysis model based on the full weighted feature set and the ensemble learning algorithm comprises: randomly selecting a core index as a target index; constructing a first training sample set according to a weighted feature set of the target index, wherein the first training sample set comprises a plurality of independent variable vectors of the target index and dependent variable vectors mapped by each independent variable vector; respectively training a plurality of preset learners according to the first training sample set, and collecting the prediction results of each preset learner on the verification set after training is finished; Sequentially generating the prediction results of each core index according to the steps, and constructing a second training sample set, wherein the second training sample set comprises the prediction results of all the core indexes and corresponding actual vehicle outlet result labels; Inputting the second training sample set to a meta learner, and carrying out weighted fusion on the prediction results of a plurality of preset learners through the meta learner to obtain a final prediction output result of the vehicle outlet analysis model; stopping model training when the reliability of the final prediction output result is greater than a preset reliability threshold; Wherein the weight parameters of the meta-learner are optimized by minimizing the loss function of the predicted output and the actual result label.
7. The machine learning and big data based vehicle exit analysis method of claim 6, wherein analyzing the data collected in real time using the trained vehicle exit analysis model, outputting the predicted result of the vehicle exit trend, comprises: collecting multi-source data related to a vehicle outlet in real time, and preprocessing; inputting the preprocessed data into a trained vehicle outlet analysis model to obtain a prediction result; the prediction result comprises real-time prediction values of all core indexes.
8. The machine learning and big data based vehicle outlet analysis method of claim 7, wherein generating a vehicle outlet decision from the prediction result and a pre-built decision rule base comprises: the pre-constructed decision rule base comprises a core index threshold rule, an index association rule, a risk early warning rule and a strategy recommendation rule, and each rule is mapped with a corresponding preset decision base; And judging whether to trigger rules in a preset decision rule base based on the real-time predicted values of the core indexes, and if so, generating a vehicle exit decision according to the preset decision base mapped by the trigger rules and the rule priority.
9. A machine learning and big data based vehicle exit analysis system comprising: The acquisition module is used for constructing a multi-source data acquisition system, acquiring multi-source data related to a vehicle outlet, and preprocessing the multi-source data to obtain a standard analysis data set; The construction module is used for constructing a multidimensional association knowledge graph based on the standard analysis data set, carrying out association mining to obtain a plurality of preset conduction paths of each core index, and determining a plurality of weighted feature sets according to all the conduction paths; The analysis module is used for training a vehicle outlet analysis model according to all the weighted feature sets and the integrated learning algorithm, analyzing the data acquired in real time by utilizing the trained vehicle outlet analysis model, and outputting a prediction result of the vehicle outlet trend; And the generation module is used for generating a vehicle outlet decision according to the prediction result and a pre-constructed decision rule base.
10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the machine learning and big data based vehicle outlet analysis method of any of claims 1 to 8 when the program is executed by the processor.

Description

Vehicle exit analysis method and system based on machine learning and big data Technical Field The application relates to the technical field of vehicle outlet analysis, in particular to a vehicle outlet analysis method and system based on machine learning and big data. Background In recent years, the automobile export market scale is continuously enlarged, but increasingly complex market demands and competing environments are faced, and accurate prediction of the automobile export trend and optimization of the export strategy become key to industry development. In the prior art, the traditional vehicle outlet analysis method relies on manual experience or single-dimension data statistics, the vehicle outlet business relates to massive and multi-source data, and comprehensive influences of multiple factors such as market dynamics, policy changes, competitor strategies and the like on the outlet business are difficult to capture comprehensively, so that the data cannot be integrated and processed effectively, dynamic influence relations among all the links are modeled accurately, analysis results are lagged, prediction accuracy is insufficient, and timely and effective decision support cannot be provided for enterprises. Disclosure of Invention In order to solve the technical problems, the application provides a vehicle outlet analysis method and a vehicle outlet analysis system based on machine learning and big data, standard analysis data is obtained by collecting multi-source data, a multi-dimensional associated knowledge graph is constructed, a plurality of conducting paths influencing the vehicle outlet are excavated, effective fusion and relation excavation of multi-source heterogeneous data are realized, a plurality of weighting feature sets are determined based on all the conducting paths, a vehicle outlet analysis model is trained by combining an integrated learning algorithm, and the accuracy of key index prediction is improved, so that the accurate prediction and decision support of the vehicle outlet trend are realized, and the comprehensiveness, the accuracy and the timeliness of vehicle outlet analysis are improved. In some embodiments of the present application, there is provided a vehicle outlet analysis method based on machine learning and big data, comprising: constructing a multi-source data acquisition system, acquiring multi-source data related to a vehicle exit, and preprocessing to obtain a standard analysis data set; constructing a multidimensional association knowledge graph based on a standard analysis data set, carrying out association mining to obtain a plurality of preset conduction paths of each core index, and determining a plurality of weighted feature sets according to all the conduction paths; training a vehicle outlet analysis model according to all the weighted feature sets and an ensemble learning algorithm, analyzing data acquired in real time by using the trained vehicle outlet analysis model, and outputting a prediction result of the vehicle outlet trend; and generating a vehicle outlet decision according to the prediction result and a pre-constructed decision rule base. In some embodiments of the application, obtaining a standard analytical dataset includes: the multi-source data acquisition system comprises a plurality of data acquisition networks with dimensions, and each data acquisition network corresponds to a preset acquisition time interval; Preprocessing multi-source data acquired by a data acquisition network of each dimension in a preset time interval, and constructing a plurality of data sequences of each dimension according to the preprocessed data and the acquisition sequence; wherein the pretreatment comprises cleaning, de-duplication, format conversion and standardization treatment; setting corresponding check nodes according to the fluctuation degree of each data sequence, and carrying out quality check on each data sequence by combining with a preset check rule to obtain the quality coefficient of each data sequence; judging whether to carry out supplementary correction on the corresponding data sequence according to the quality coefficient, if not, incorporating the corresponding data sequence into a standard analysis data set, and if so, incorporating the data after the supplementary correction is qualified into the standard analysis data set in sequence; The standard analysis dataset comprises several data sequences of all dimensions. In some embodiments of the application, constructing a multidimensional associated knowledge-graph based on a standard analytical dataset includes: Performing entity identification and extraction on the standard analysis data set to obtain four-dimensional entity nodes and corresponding attribute information, wherein the four-dimensional entity comprises market entity nodes, product entity nodes, policy entity nodes and competition entity nodes; Carrying out semantic association among entity nodes based on the influen