A Multi-Data Fusion Method and System Based on Edge Computing

Abstract

This invention relates to the field of multi-data fusion technology, specifically disclosing a multi-data fusion method and system based on edge computing, comprising the following steps: S1: The processing center connects to multiple sub-data sources, collects edge data, and generates processing results; simultaneously, it acquires all historical processing results and historical edge data, including all historical logs transmitted from the sub-data sources; S2: The historical logs are integrated and divided into changed segments and non-changed segments; scene segment features of changed segments are extracted, and changed segments with consistent features are categorized into a set of values ​​with the same indicator; S3: Based on the set of values ​​with the same indicator, the importance of each changed segment is analyzed, and it is determined whether it is a key segment; S4: Based on the key segments, the remaining segments are determined, and an edge data template is generated; the key edge data is normalized in format, matched with scene segment features, and then filled into the template, finally outputting the processing results.

Description

Technical Field

[0001] This invention relates to the field of multi-data fusion technology, specifically to a multi-data fusion method and system based on edge computing. Background Technology

[0002] In today's era of rapid development in information technology, the diversity and complexity of data are unprecedented. Among them, multi-source data fusion, as a common form of data, is widely present in many fields.

[0003] In existing technologies, the fusion of multi-source data often adopts a relatively traditional approach, which involves first unifying the data from different data sources into a standardized format before proceeding with subsequent processing. Specifically, since different data sources may have their own unique data structures, formats, encoding methods, and semantic connotations, in order to achieve data integration and collaborative processing, it is necessary to first sort out and standardize these differences, transforming them into a unified standardized format.

[0004] However, this approach has revealed many drawbacks in practical applications, the most prominent of which is its slow efficiency. On one hand, the data standardization process itself is quite cumbersome. Data from different data sources can vary significantly, and converting them all into a unified format requires a substantial investment of time and computing resources.

[0005] On the other hand, with the continuous growth of data volume, the challenges faced by this method of standardization before processing are becoming increasingly severe. In the big data environment, data is generated at an extremely rapid pace, and the data volume is growing explosively. If all data must be standardized before fusion and analysis each time, the time consumed by the entire process will be unbearable, failing to meet the real-time or near real-time application requirements. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-data fusion method and system based on edge computing, and to solve the following technical problems.

[0007] The objective of this invention can be achieved through the following technical solutions:

[0008] A multi-data fusion method based on edge computing includes the following steps:

[0009] S1: Establish a processing center, which is communicatively connected to several sub-data sources. The processing center obtains edge data through each sub-data source, processes the edge data, and generates processing results. It also obtains all historical processing results and historical edge data, including all historical logs transmitted from the sub-data sources to the processing center.

[0010] S2: Integrate the historical logs and divide them into changed segments and non-changed segments. Determine the scene segment characteristics of each changed segment. In different historical logs, obtain several changed segments with consistent scene segment characteristics and record them as the set of the same index values.

[0011] S3: Based on the set of values ​​for the same indicator, determine whether each changed segment is a key segment;

[0012] S4: Based on each key segment, determine the remaining segments in the historical log and generate an edge data template for the sub-data source; obtain key edge data and perform format normalization processing on the key edge data; determine the scene segment features of the key edge data, and fill the format normalized key edge data into the edge data template according to the scene segment features to finally obtain the processing result.

[0013] As a further aspect of the present invention, the process of integrating the historical logs includes:

[0014] A preset standard format is used to convert the format of all historical logs to the standard format and delete redundant data in the historical logs, including duplicate data; and the business logic of the sub-data source is obtained, and each piece of data in the historical logs is sorted according to the business logic.

[0015] As a further aspect of the present invention: the process of dividing historical logs into changed segments and non-changed segments includes:

[0016] Based on the business logic, several data types are obtained, including temperature data, time data, and device data; when the historical log is divided into several log segments according to the business logic, each log segment corresponds to a data type.

[0017] Each log segment is binary encoded, converted into a binary sequence, and numbered sequentially. A two-dimensional coordinate system is established with the number as the horizontal axis and the binary sequence as the vertical axis. In the historical log, each numbered log segment and its corresponding binary sequence are converted into coordinate points on the two-dimensional coordinate system, and each coordinate point is fitted to obtain the sequence curve of the historical log.

[0018] Place all sequence curves in the same coordinate system. For any number, draw a perpendicular line from the number to the horizontal axis and obtain all intersection points of the perpendicular line with each sequence curve. If all intersection points coincide, the log segment corresponding to the number is recorded as a non-changed segment; if all intersection points do not coincide, the log segment corresponding to the number is recorded as a changed segment.

[0019] As a further aspect of the present invention: the process of determining the set of identical index values ​​includes:

[0020] Obtain the end point of the previous changed segment of the changed segment, denoted as the upper end point; obtain the start point of the next changed segment of the changed segment, denoted as the lower start point; obtain the start point of the changed segment, denoted as the segment start point; obtain the end point of the changed segment, denoted as the segment end point.

[0021] Obtain the set of log segments from the upper endpoint to the segment start point, denoted as the upper set, and obtain the set of log segments from the segment end point to the lower start point, denoted as the lower set; extract the text feature vectors of the upper set and the lower set respectively, denoted as the upper feature and the lower feature, to obtain the scene segment feature (Fv) of the changed segment. up Fv down ), where Fv up Representing the upper feature, Fv down Indicate the following features;

[0022] If two modified segments exist, obtain the scene segment features of each modified segment, and denote them as (Fv). up Fv down ) and (Fv up ´,Fv down If we calculate the scene similarity between the two modified segments, we get the scene similarity between them. If the scene similarity between two modified segments exceeds a preset scene similarity threshold, then the scene segment features of the two modified segments are considered to be consistent.

[0023] In different historical logs, obtain several change segments with consistent scene segment characteristics and denot them as the same indicator value set.

[0024] As a further aspect of the present invention: the process of determining whether each modified segment is a key segment includes:

[0025] All changed segments are binary encoded to obtain the numerical code of each changed segment, and all historical processing results are binary encoded to obtain the result numerical code; any historical log is selected as the standard log, and all numerical codes in the standard log are recorded as the standard value; and the historical processing result corresponding to the standard log is obtained and recorded as the standard result.

[0026] For any numerical code in the historical log, determine the standard and index value of the numerical code, and obtain the deviation value of the numerical code; determine the historical processing result corresponding to the historical log, and obtain the result deviation value of the historical log; based on all the deviation values ​​of the historical log and the result deviation value, obtain a sample data;

[0027] Any deviation value in the sample data is recorded as the deviation value to be measured. All deviation values ​​other than the deviation value to be measured are recorded as the remaining deviation values. The total remaining deviation of all remaining deviation values ​​is obtained. Based on the deviation value to be measured, the total remaining deviation, and the result deviation value in the sample data, a sample data of the deviation value to be measured is obtained.

[0028] A regression model is established. Based on the sample data of all historical logs, the deviation value to be measured is obtained. All sample data to be measured are obtained. The regression model is trained using all sample data to be measured. The weight coefficients of the regression model are adjusted to obtain the deviation weight of the deviation value to be measured.

[0029] Obtain the deviation weight of each numerical code in the historical log to get the deviation weight of each changed segment; set a weight threshold, select the changed segments whose deviation weight exceeds the weight threshold, and record them as key segments.

[0030] As a further aspect of the present invention: the process of obtaining the deviation weight includes:

[0031] Establish a regression model W1Dv0 + W2ODv = RDv, where W1 and W2 are preset weight coefficients, Dv0 is the deviation value to be tested, ODv is the total deviation of the remaining values, and RDv is the result deviation value. Split all the sample data to be tested into several training sets and test sets. Train the regression model using the training sets and adjust the size of W1 and W2 to obtain the initial regression model.

[0032] All results in the test set are numerically encoded and recorded as actual numerical values. The test set is then substituted into the initial regression model to obtain several simulated numerical values, which yields the absolute error. Where N is the total number of simulated numerical values ​​of the result, RDv pred，t RDv represents the numerical encoded simulated value of the t-th result. true，t Let t represent the actual value of the numerical encoding of the t-th result, where t ∈ [1, N] and t is a positive integer;

[0033] The sizes of W1 and W2 are adjusted sequentially through each training set to minimize the final absolute error Loss, and the W1 at this point is obtained, which is denoted as the deviation weight of the deviation value to be tested.

[0034] As a further aspect of the present invention: the process of generating edge data templates for sub-data sources includes:

[0035] In all log segments, log segments other than key segments are recorded as the remaining segments. Standard logs are obtained, and the key segments in the standard logs are replaced with empty strings to obtain the edge data template.

[0036] A multi-data fusion system based on edge computing includes:

[0037] Data acquisition module: Establish a processing center, which is communicatively connected to several sub-data sources. The processing center acquires edge data through each sub-data source, processes the edge data, and generates processing results; acquires all historical processing results and historical edge data, including all historical logs transmitted from the sub-data sources to the processing center;

[0038] Template generation module: Integrates various historical logs and divides them into changed segments and non-changed segments. It determines the scene segment characteristics of each changed segment and obtains several changed segments with consistent scene segment characteristics from different historical logs, which are recorded as the same index value set.

[0039] Based on the set of values ​​for the same indicator, determine whether each change segment is a key segment; based on each key segment, determine the remaining segments in the historical log and generate edge data templates for the sub-data source;

[0040] Fusion module: acquires key edge data and performs format normalization processing on the key edge data; determines the scene segment features of the key edge data, and fills the format normalized key edge data into the edge data template according to the scene segment features, and finally obtains the processing result.

[0041] The beneficial effects of this invention are:

[0042] This invention utilizes edge computing technology to perform preliminary processing near the data source (sub-data source), reducing data transmission latency and central node load, significantly improving the real-time performance of multi-source data fusion. It normalizes and fuses only key segments, avoiding the waste of computational resources caused by full data standardization and optimizing overall processing speed. Based on scene segment features and sets of identical indicator values, it intelligently identifies key change segments (key segments), ensuring the fusion process focuses on data components that significantly impact the results. It quantifies the weight of change segments using regression models, scientifically selecting key data, reducing noise interference, and improving the reliability of results. It supports differentiated formats from different data sources (such as temperature, time, and device data), achieving unified processing of heterogeneous data through binary encoding and feature matching. Its template-based filling mechanism is compatible with diverse data scenarios and has strong scalability. This invention, through the combination of edge computing and intelligent data filtering, achieves efficient, accurate, and low-redundancy multi-source data fusion, suitable for complex scenarios requiring real-time performance, high reliability, and heterogeneous data processing capabilities. Attached Figure Description

[0043] The invention will now be further described with reference to the accompanying drawings.

[0044] Figure 1This is a schematic diagram illustrating the steps of a multi-data fusion method based on edge computing according to the present invention;

[0045] Figure 2 This is a schematic diagram of the structure of a multi-data fusion system based on edge computing according to the present invention. Detailed Implementation

[0046] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0047] Please see Figure 1 As shown, this invention is a multi-data fusion method based on edge computing, comprising the following steps:

[0048] S1: Establish a processing center, which is communicatively connected to several sub-data sources. The processing center obtains edge data through each sub-data source, processes the edge data, and generates processing results. It also obtains all historical processing results of the processing center and historical edge data of the sub-data sources, including all historical logs transmitted from the sub-data sources to the processing center.

[0049] When structuring the processing center, the processing center needs to have edge computing capabilities, including a data receiving module, a preliminary processing unit, a storage unit, and a communication interface; among them, the data receiving module supports multi-protocol access (such as MQTT, HTTP, CoAP, etc.) to adapt to the communication needs of different sub-data sources; the preliminary processing unit can complete real-time data reception, verification, and temporary caching; the storage unit is used to persistently store historical data and processing results;

[0050] Sub-data sources include, but are not limited to, various sensors (such as temperature, humidity, and pressure sensors), IoT devices (such as smart terminals and industrial controllers), and business systems (such as production execution systems and equipment management platforms). The processing center establishes a stable connection with the sub-data sources through wired communication (such as Ethernet and RS485) or wireless communication (such as Wi-Fi, LoRa, and 5G), and completes identity authentication (such as device ID verification and encrypted transmission) through preset communication protocols to ensure the security and uniqueness of data transmission.

[0051] The processing center can send collection instructions to the sub-data sources at a preset frequency, which is suitable for periodic data; the sub-data sources actively push data to the processing center when specific conditions are met (such as changes in equipment status or data reaching a threshold), which is suitable for event-based data (such as equipment fault alarms and abnormal status reports).

[0052] After receiving edge data, the processing center first performs basic cleaning, and then performs preliminary integration based on the business logic of the sub-data sources (such as sorting by timestamp and associated device identifier) ​​to generate preliminary processing results.

[0053] Based on the business logic of the sub-data source, the conclusive information obtained after targeted processing of edge data is the processing result; for example, by analyzing temperature data and equipment operating parameters, a judgment result of "equipment A is at risk of failure due to excessive temperature" is generated.

[0054] S2: Integrate all historical logs and divide them into several log segments. Compare the historical logs with each other and divide all log segments into changed segments and non-changed segments based on the comparison results. Determine the scene segment characteristics of each changed segment. In different historical logs, obtain several changed segments with consistent scene segment characteristics and record them as the same index value set.

[0055] In a preferred embodiment of the present invention, the process of integrating the historical logs includes:

[0056] A preset standard format is used to convert the format of all historical logs to the standard format and delete redundant data in the historical logs, including duplicate data; and the business logic of the sub-data source is obtained, and each piece of data in the historical logs is sorted according to the business logic.

[0057] Specifically, the historical logs of all sub-data sources are converted into a standard format, and redundant data is identified and deleted through hash verification or field matching, such as duplicate logs generated by the same device at the same timestamp, or multiple records with completely identical content.

[0058] Based on the business processes of the sub-data sources, such as data collection, preprocessing, transmission, and storage, the logs are reordered according to time sequence or business node order. For example, industrial equipment logs need to be arranged in the process order of equipment startup, parameter collection, operating status, and shutdown to ensure that the log sequence is consistent with the business logic.

[0059] In a preferred embodiment of the present invention, the process of dividing historical logs into several log segments includes:

[0060] Based on the business logic, several data types are obtained, including temperature data, time data, and device data; when the historical log is divided into several log segments according to the business logic, each log segment corresponds to a data type.

[0061] For example, continuous logs containing "temperature=25℃" and "temperature=26℃" are divided into "temperature data segments", and logs containing "start time=10:00" and "down time=18:00" are divided into time data segments.

[0062] In a preferred embodiment of the present invention, the process of comparing historical logs includes:

[0063] Each log segment is binary encoded, converted into a binary sequence, and numbered sequentially. A two-dimensional coordinate system is established with the number as the horizontal axis and the binary sequence as the vertical axis. In the historical log, each numbered log segment and its corresponding binary sequence are converted into coordinate points on the two-dimensional coordinate system, and each coordinate point is fitted to obtain the sequence curve of the historical log.

[0064] Place all sequence curves in the same coordinate system. For any number, draw a perpendicular line from the number to the horizontal axis and obtain all intersection points of the perpendicular line with each sequence curve. If all intersection points coincide, the log segment corresponding to the number is recorded as a non-changed segment; if all intersection points do not coincide, the log segment corresponding to the number is recorded as a changed segment.

[0065] Specifically, each log segment is binary encoded, for example, "temperature=25℃" is converted into the binary sequence "10101", "temperature=26℃" is converted into "10110"; and the segments are numbered according to their order of appearance in the log, for example, the first segment is numbered 1, the second is numbered 2, and so on.

[0066] A two-dimensional coordinate system is established with the serial number as the horizontal axis and the binary sequence as the vertical axis. The encoded value of each segment is mapped to the coordinate point in the two-dimensional coordinate system, and the sequence curve of each historical log is generated by curve fitting.

[0067] In a preferred embodiment of the present invention, the process of determining the scene segment features of the changed segment includes:

[0068] Obtain the end point of the previous changed segment of the changed segment, denoted as the upper end point; obtain the start point of the next changed segment of the changed segment, denoted as the lower start point; obtain the start point of the changed segment, denoted as the segment start point; obtain the end point of the changed segment, denoted as the segment end point.

[0069] For example, if change segment B in the log is after change segment A and before change segment C, then the end point of A is the upper end point of B, the start point of C is the lower start point of B, and the start and end points of B itself are the start and end points of the segment.

[0070] Obtain the set of log segments from the upper endpoint to the segment start point, denoted as the upper set, and obtain the set of log segments from the segment end point to the lower start point, denoted as the lower set; extract the text feature vectors of the upper set and the lower set respectively, denoted as the upper feature and the lower feature, to obtain the scene segment feature (Fv) of the changed segment. up Fv down ), where Fv up Representing the upper feature, Fv down Indicate the following features;

[0071] In a preferred embodiment of the present invention, the process of determining whether scene fragment features are consistent includes:

[0072] If two modified segments exist, obtain the scene segment features of each modified segment, and denote them as (Fv). up Fv down ) and (Fv up ´,Fv down If we calculate the scene similarity between the two modified segments, we get the scene similarity between them. If the scene similarity between two modified segments exceeds a preset scene similarity threshold, then the scene segment features of the two modified segments are considered to be consistent.

[0073] The matching degree of the scene segments is quantified by obtaining the average cosine similarity of the context features before and after the changes. The closer the Sim value is to 1, the more similar the context before and after the changes are, that is, the more consistent the scenes are.

[0074] S3: Encode all changed segments into binary to obtain the numerical code of each changed segment, and encode all historical processing results into binary to obtain the result numerical code; select any historical log as the standard log, record all the numerical codes in the standard log as the standard value; and obtain the historical processing result corresponding to the standard log as the standard result.

[0075] For any numerical code in the historical log, determine the standard and index value of the numerical code, and obtain the deviation value of the numerical code; determine the historical processing result corresponding to the historical log, and obtain the result deviation value of the historical log; based on all the deviation values ​​of the historical log and the result deviation value, obtain a sample data;

[0076] Any deviation value in the sample data is recorded as the deviation value to be measured. All deviation values ​​other than the deviation value to be measured are recorded as the remaining deviation values. The total remaining deviation of all remaining deviation values ​​is obtained. Based on the deviation value to be measured, the total remaining deviation, and the result deviation value in the sample data, a sample data of the deviation value to be measured is obtained.

[0077] A regression model is established. Based on the sample data of all historical logs, the deviation value to be measured is obtained. All sample data to be measured are obtained. The regression model is trained using all sample data to be measured. The weight coefficients of the regression model are adjusted to obtain the deviation weight of the deviation value to be measured.

[0078] Obtain the deviation weight of each numerical code in the historical log to get the deviation weight of each change segment; set a weight threshold, select the change segments whose deviation weight exceeds the weight threshold, and record them as key segments;

[0079] In a preferred embodiment of the present invention, the process of obtaining the deviation value and the result deviation value includes:

[0080] The deviation value of the numerical encoding is Dv=|Nc-nc|, where nc is the standard index value and Nc is the numerical encoding; the standard result is binary encoded to obtain the standard result numerical encoding of the standard result, and the historical processing result corresponding to the historical log is binary encoded to obtain the result numerical encoding, where the result deviation value is RDv=|RNc-Rnc|, where RNc is the result numerical encoding and Rnc is the standard result numerical encoding;

[0081] In a preferred embodiment of the present invention, the remaining total deviation of all deviation values Where m is the total number of other deviation values, Dv k ´ represents the kth remaining deviation value, where k∈[1,m] and k is a positive integer;

[0082] In a preferred embodiment of the present invention, the process of obtaining the deviation weight of the deviation value to be measured includes:

[0083] Establish a regression model W1Dv0+W2ODv=RDv, where W1 and W2 are preset weight coefficients and Dv0 is the deviation value to be tested; split all the sample data to be tested into several training sets and test sets, train the regression model through the training set, adjust the size of W1 and W2, and obtain the initial regression model.

[0084] All results in the test set are numerically encoded and recorded as actual numerical values. The test set is then substituted into the initial regression model to obtain several simulated numerical values, which yields the absolute error. Where N is the total number of simulated numerical values ​​of the result, RDv pred，t RDv represents the numerical encoded simulated value of the t-th result. true，t Let t represent the actual value of the numerical encoding of the t-th result, where t ∈ [1, N] and t is a positive integer;

[0085] The values ​​of W1 and W2 are adjusted sequentially using gradient descent on each training set to minimize the final absolute error Loss, i.e., iterative training until the error converges; and W1 at this point is obtained and denoted as the deviation weight of the deviation value to be tested.

[0086] It should be noted that converting the changed fragments and historical processing results into binary numerical codes facilitates quantitative analysis and comparison. By comparing the current numerical codes with the standard indicator values ​​of the standard logs, the deviation value is calculated to measure the degree of data fluctuation.

[0087] Establish a regression equation, train the model using historical data, determine the weight of each deviation value on the final result, and based on the deviation weights obtained from the training, filter out the change segments whose weights exceed the threshold and identify them as key segments;

[0088] Identify segments from massive amounts of change that have a significant impact on the final processing results (such as decision-making, alarms, and predictions), avoid wasting resources on full data processing, and only perform subsequent normalization and template filling on key segments to reduce invalid calculations and accelerate real-time response in edge computing scenarios.

[0089] S4: In all log segments, log segments other than key segments are recorded as the remaining segments, and the edge data template of the sub-data source is generated;

[0090] The processing center acquires key edge data from the sub-data source using the acquisition method and performs format normalization on the key edge data. It then determines the scene segment features of the key edge data and fills the format-normalized key edge data into the edge data template based on these scene segment features, ultimately obtaining the processing result.

[0091] In a preferred embodiment of the present invention, the process of generating the edge data template includes:

[0092] Obtain the standard log, replace the empty strings with the key segments in the standard log to obtain the edge data template;

[0093] In a preferred embodiment of the present invention, the key segment collection method includes log collection tools, function calls, and regular expressions;

[0094] In a preferred embodiment of the present invention, the format normalization process is to convert the format of all key edge data into the standard format;

[0095] In a preferred embodiment of the present invention, the process of filling the edge data template with key edge data includes:

[0096] Obtain scene fragment features of each empty string in the edge data template, and record them as empty features; match the scene fragment features of the key edge data with each empty feature, select the empty feature with the highest scene similarity to the scene fragment features of the key edge data, record it as the target empty, and replace the target empty with the key edge data in the edge data template;

[0097] Specifically, the key segments in the standard logs are replaced with empty strings (placeholders), while retaining fixed structures (such as timestamps, device IDs, and other static fields); key segments collected from different sub-data sources are uniformly converted into a preset standard format, scene segment features of key data are extracted, and similarity matching is performed with the empty features in the template to determine the filling position; the normalized key data is inserted into the corresponding empty position in the template according to the feature matching results to generate the final processing result.

[0098] The template retains only the necessary static structure, and dynamic data (key segments) are filled in as needed, reducing repeated transmission and storage. Through format normalization, key data from different sources and formats are processed uniformly to avoid fusion failures caused by structural differences. Templated processing avoids full data reconstruction and only performs real-time calculations on key dynamic parts (key segments) to meet low latency requirements. The template can be dynamically adjusted (such as updating non-key segments according to business logic) to adapt to changing edge scenarios (such as adding sensors or protocols).

[0099] A multi-data fusion system based on edge computing includes:

[0100] Data acquisition module: Establish a processing center, which is communicatively connected to several sub-data sources. The processing center acquires edge data through each sub-data source, processes the edge data, and generates processing results; acquires all historical processing results and historical edge data, including all historical logs transmitted from the sub-data sources to the processing center;

[0101] Template generation module: Integrates various historical logs and divides them into changed segments and non-changed segments. It determines the scene segment characteristics of each changed segment and obtains several changed segments with consistent scene segment characteristics from different historical logs, which are recorded as the same index value set.

[0102] Based on the set of values ​​for the same indicator, determine whether each change segment is a key segment; based on each key segment, determine the remaining segments in the historical log and generate edge data templates for the sub-data source;

[0103] Fusion module: acquires key edge data and performs format normalization processing on the key edge data; determines the scene segment features of the key edge data, and fills the format normalized key edge data into the edge data template according to the scene segment features, and finally obtains the processing result.

[0104] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for multi-data fusion based on edge computing, characterized in that, The method comprises the following steps: S1: establishing a processing center, the processing center being communicatively connected to a plurality of sub-data sources, the processing center acquiring edge data from each sub-data source and processing the edge data to generate a processing result; all historical processing results and historical edge data are acquired, the historical edge data including all historical logs transmitted by the sub-data sources to the processing center; S2: each historical log is integrated and divided into changed segments and unchanged segments, the scene segment features of each changed segment are determined, a plurality of changed segments with consistent scene segment features are acquired in different historical logs, and the changed segments are recorded as a same-index value set; S3: whether each changed segment is a key segment is determined according to the same-index value set; S4: according to each key segment, the remaining segments in the historical log are determined, an edge data template of the sub-data source is generated, key edge data is acquired, and the key edge data is subjected to format normalization processing; the scene segment features of the key edge data are determined, the key edge data subjected to the format normalization processing is filled into the edge data template according to the scene segment features, and finally a processing result is obtained. 2.The method of claim 1, wherein, In S2, the process of integrating the historical logs comprises: a standard format is preset, the formats of all historical logs are converted into the standard format, redundant data in the historical logs is deleted, the redundant data including repeated data, and the business logic of the sub-data source is acquired, each piece of data in the historical log is sorted according to the business logic. 3.The method of claim 2, wherein, In S2, the process of dividing the historical logs into changed segments and unchanged segments comprises: a plurality of data types are obtained according to the business logic, the data types including temperature data, time data and device data, each log segment corresponds to a data type when the historical logs are divided into a plurality of log segments according to the business logic; each log segment is subjected to binary coding, each log segment is converted into a binary sequence, and each log segment is numbered in sequence; a two-dimensional coordinate system is established with the number as the horizontal coordinate and the binary sequence as the vertical coordinate; in the historical logs, each numbered log segment and the binary sequence corresponding to each numbered log segment are converted into coordinate points at corresponding positions on the two-dimensional coordinate system, and each coordinate point is fitted to obtain a sequence curve of the historical logs; all sequence curves are placed in the same coordinate system, for any number, a vertical line perpendicular to the horizontal coordinate axis is drawn through the number, all intersection points of the vertical line and each sequence curve are acquired, if all intersection points are a coincident point, the log segment corresponding to the number is recorded as an unchanged segment, and if all intersection points are not a coincident point, the log segment corresponding to the number is recorded as a changed segment. 4.The method of claim 1, wherein, In S2, the determination process of the same-index value set comprises: an end point of a previous changed segment of the changed segment is acquired and recorded as an upper end point, a start point of a next changed segment of the changed segment is acquired and recorded as a lower start point, a start point of the changed segment is acquired and recorded as a segment start point, and an end point of the changed segment is acquired and recorded as a segment end point; obtain a set of log fragments between the upper end point and the fragment start point, denoted as an upper set, and obtain a set of log fragments between the fragment end point and the lower start point, denoted as a lower set; extract text feature vectors of the upper set and the lower set respectively, denoted as an upper feature and a lower feature respectively, to obtain a scenario fragment feature (Fv up , Fv down ) of the change fragment, wherein Fv up represents the upper feature, and Fv down represents the lower feature; If there are two changed clips, the scene clip features of the two changed clips are obtained respectively, and are denoted as (Fv up , Fv down ) and (Fv up , Fv down ) respectively, scene similarity of the two changed clips is obtained, and if the scene similarity of the two changed clips exceeds a preset scene similarity threshold, it is considered that the scene clip features of the two changed clips are consistent; In different historical logs, obtain several change fragments with consistent scene fragment features, denoted as a same index value set.

5. The method of claim 1, wherein, In S3, the process of determining whether each change fragment is a key fragment includes: Binary encode all change fragments to obtain numerical encodings of each change fragment, and binary encode all historical processing results to obtain result numerical encodings; select any historical log as a standard log, record all numerical encodings in the standard log as standard values, and obtain the historical processing result corresponding to the standard log, denoted as a standard result; For any numerical encoding in the historical log, determine the standard same index value of the numerical encoding, and obtain the deviation value of the numerical encoding; and determine the historical processing result corresponding to the historical log, and obtain the result deviation value of the historical log; obtain a sample data according to all deviation values and result deviation values of the historical log; Record any deviation value in the sample data as a to-be-tested deviation value, record the deviation values other than the to-be-tested deviation value as remaining deviation values, obtain the remaining total deviation of all remaining deviation values, and obtain a to-be-tested sample data of the to-be-tested deviation value according to the to-be-tested deviation value, the remaining total deviation, and the result deviation value in the sample data; Establish a regression model, obtain all to-be-tested sample data of the to-be-tested deviation value according to the sample data of all historical logs, train the regression model through all to-be-tested sample data, adjust the weight coefficient of the regression model, and obtain the deviation weight of the to-be-tested deviation value. Obtain the deviation weight of each numerical encoding in the historical log to obtain the deviation weight of each change fragment; set a weight threshold, and select the change fragments with deviation weights exceeding the weight threshold, denoted as key fragments. 6.The method of claim 5, wherein, In S3, the process of obtaining the deviation weight includes: Establish a regression model W1Dv0+W2ODv=RDv, where W1 and W2 are preset weight coefficients, Dv0 is a to-be-tested deviation value, ODv is a remaining total deviation, and RDv is a result deviation value; split all to-be-tested sample data into several training sets and test sets, train the regression model through the training sets, adjust the sizes of W1 and W2, and obtain an initial regression model; encoding the all result numerical values in the test set as result numerical value encoding actual values; and substituting the test set into the initial regression model to obtain a plurality of result numerical value encoding simulation values according to the initial regression model, to obtain absolute errors wherein N is the total number of result numerical value encoding simulation values, RDv pred，t represents the tth result numerical value encoding simulation value, RDv true，t represents the tth result numerical value encoding actual value, t∈[1, N] and t is a positive integer; Adjust the sizes of W1 and W2 through each training set in turn, so that the final absolute error Loss reaches a minimum, and obtain W1 at this time, denoted as the deviation weight of the to-be-tested deviation value. 7.The method of claim 5, wherein, In S4, the process of generating the edge data template of the sub-data source includes: In all log fragments, record the log fragments other than the key fragments as remaining fragments, obtain the standard log, replace the key fragments in the standard log with empty strings to obtain an edge data template.

8. An edge computing based multi-data fusion system, characterized by, It includes: A data acquisition module: establish a processing center, the processing center is communicatively connected with several sub-data sources, the processing center obtains edge data through each sub-data source, and processes the edge data to generate a processing result; Obtain all historical processing results and historical edge data, the historical edge data includes all historical logs transmitted by the sub-data sources to the processing center; The template generation module integrates each historical log, divides the historical log into change segments and non-change segments, determines scene segment features of each change segment, obtains a plurality of change segments with consistent scene segment features in different historical logs, and records the change segments as a same-index value set; According to the same-index value set, it is determined whether each change segment is a key segment; according to each key segment, the remaining segments in the historical log are determined, and an edge data template of the sub-data source is generated; The fusion module obtains key edge data, performs format normalization processing on the key edge data, determines scene segment features of the key edge data, fills the key edge data after the format normalization processing into the edge data template according to the scene segment features, and finally obtains a processing result.

Images