Multi-source data fusion method and intelligent terminal

Abstract

The application discloses a multi-source data fusion method and an intelligent terminal, relates to the technical field of multi-source data fusion, and comprises the following steps: performing time sequence difference identification and hierarchical compensation processing on a standardized access data set, unifying the same fusion time, and outputting a time sequence alignment candidate data set; performing double-channel shunt processing on a marked result set, forming main fusion state content and conflict explanation information; performing result arrangement and real-time output processing on a target fusion set, reserving fusion credibility information, and obtaining a multi-source fusion output set. Through the parallel processing architecture of the main fusion channel and the conflict processing channel, the transparency and interpretability of the fusion result are enhanced, and direct and clear bases are provided for subsequent decision correction, source data quality evaluation and optimization.

Description

Technical Field

[0001] This invention relates to the field of multi-source data fusion technology, and in particular to a multi-source data fusion method and a smart terminal. Background Technology

[0002] With the rapid development of IoT, intelligent sensing, and collaborative computing technologies, multi-source data fusion technology has become a core support for achieving accurate situational awareness and intelligent decision-making in wide areas. Related research mainly focuses on data access standardization, temporal alignment algorithms, cross-source association matching, and multi-criteria fusion models. Among these, data preprocessing based on a unified spatiotemporal benchmark, object association based on feature similarity or spatial proximity, and fusion methods based on weighted averaging, Kalman filtering, or confidence theory constitute the current mainstream technical paths, aiming to improve the consistency of multi-dimensional descriptions of the same entity or event in dynamic environments.

[0003] However, existing methods still have significant limitations when dealing with complex real-world scenarios. When data from different sources contain a large amount of consistent information while also including some conflicting or contradictory data when describing the same object, traditional fusion frameworks often adopt a single processing mode. This mode is difficult to effectively distinguish and differentiate between the two different types of data, namely "consistent" and "conflicting," often leading to two results: first, in order to achieve a unified output, valuable conflicting information is over-smoothed or ignored, resulting in the loss of decision-making clues; second, because conflicting data cannot be effectively isolated and interpreted, the credibility of the overall fusion result is polluted, making the output unclear or less confident in key details. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a multi-source data fusion method to solve the problem that it is difficult to form a reliable fusion result for multi-source heterogeneous data under the conditions of temporal offset and cross-source conflict.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a multi-source data fusion method, comprising: collecting multi-source raw data and performing unified access processing and standardization before writing it into a real-time cache to form a standardized access dataset; performing temporal difference identification and hierarchical compensation processing on the standardized access dataset and unifying it to the same fusion time, outputting a temporal alignment candidate dataset; performing cross-source correspondence identification based on the temporal alignment candidate dataset to form object-level association clusters, and performing consistency judgment on the object-level association clusters to generate a labeled result set; performing dual-channel split processing on the labeled result set to form the main fusion state content and conflict explanation information; performing joint correction and conflict constraint correction on the main fusion state content and conflict explanation information to obtain a target fusion set; and performing result processing and real-time output processing on the target fusion set, while retaining fusion credibility information, to obtain a multi-source fusion output set.

[0007] As a preferred embodiment of the multi-source data fusion method of the present invention, the steps of collecting multi-source raw data, performing unified access processing and standardization, and then writing the data into a real-time cache to form a standardized access dataset are as follows: The system receives raw data from different data sources through the data access interface and attaches corresponding source identifiers, receiving time identifiers, and data type identifiers to form access data records. The access data records are standardized and organized to form standard access records. The standard access records are written into the corresponding real-time cache according to the source identifier and then summarized to form a standardized access dataset.

[0008] As a preferred embodiment of the multi-source data fusion method of the present invention, the specific steps of performing temporal difference identification and hierarchical compensation processing on the standardized access dataset and unifying it to the same fusion time, and outputting a temporally aligned candidate dataset, are as follows: Standard access records are extracted from the standardized access dataset, and temporal differences are identified for each standard access record to form temporal classification information. Based on the temporal classification information, the standard access records are divided into direct use records and compensation candidate records. Hierarchical compensation processing is performed on the compensation candidate records to generate compensation aligned records, which are then merged with the directly used records to form a time-series aligned candidate dataset.

[0009] As a preferred embodiment of the multi-source data fusion method of the present invention, the specific steps of identifying cross-source correspondences based on time-series aligned candidate datasets to form object-level association clusters are as follows: Temporal alignment candidate data from different sources are extracted from the temporal alignment candidate dataset, and a candidate correspondence set is generated by cross-source correspondence identification. The candidate corresponding relationship sets are combined and merged to form data groups of the same object, and invalid associations of the data groups of the same object are removed to form object-level association clusters.

[0010] As a preferred embodiment of the multi-source data fusion method of the present invention, the specific steps for performing consistency judgment on object-level association clusters and generating a labeled result set are as follows: For data from different sources in an object-level association cluster, consistency judgments are made based on time state information, spatial state information, and object feature information to obtain consistency judgment information. Based on the consistency determination information, each associated data in the object-level association cluster is assigned a consistency label and a conflict label, forming a label result set.

[0011] As a preferred embodiment of the multi-source data fusion method of the present invention, the specific steps of performing dual-channel splitting processing on the labeled result set to form the main fusion state content and conflict explanation information are as follows: Based on consistent and conflicting labels, the labeling result set is divided into consistent datasets and conflicting datasets. The consistent dataset is input into the main fusion channel, and the temporal state information, spatial state information, and object feature information in the consistent dataset are weighted and fused to generate the main fusion state content. Input the conflict dataset into the conflict processing channel, perform conflict analysis on the conflict dataset, extract conflict source information, conflict direction information and conflict degree information, and generate conflict explanation information.

[0012] As a preferred embodiment of the multi-source data fusion method of the present invention, the dual channels include a main fusion channel and a conflict processing channel. The main fusion channel refers to the processing path that performs fusion processing on consistent data, and the conflict processing channel refers to the processing path that performs conflict resolution processing on conflicting data.

[0013] As a preferred embodiment of the multi-source data fusion method of the present invention, the steps of jointly correcting the main fusion state content and conflict interpretation information and correcting conflict constraints to obtain the target fusion set are as follows: The main fusion state content is matched with the conflict interpretation information to form an object-level correction data set. Based on the conflict source, conflict direction and conflict degree in the object-level correction data set, the corresponding state content in the main fusion state content is corrected for deviation and the expression is adjusted to form the corrected fusion content. Identify conflict-affected content in the fusion content and perform scope restriction, direction constraint and source correction processing on the conflict-affected content based on the conflict description information to form a target fusion set.

[0014] As a preferred embodiment of the multi-source data fusion method of the present invention, the specific steps of processing and real-time outputting the target fusion set and retaining the fusion credibility information to obtain the multi-source fusion output set are as follows: The constraint correction results in the target fusion set are aggregated according to the object-level association relationship to form an object-level output group; The state expression content and conflict constraint correction content of the object-level output group are organized to form object-level result data, and the corresponding fusion credibility information is extracted from the constraint correction results. The fusion credibility information is correlated with the object-level result data to form a multi-source fusion output set.

[0015] Secondly, this invention provides a multi-source data fusion intelligent terminal, comprising: a data access module for collecting multi-source raw data and performing unified access processing and standardization before writing it into a real-time cache to form a standardized access dataset; a time-series alignment processing module for performing time-series difference identification and hierarchical compensation processing on the standardized access dataset and unifying it to the same fusion time, outputting a time-series alignment candidate dataset; a consistency judgment module for identifying cross-source correspondences based on the time-series alignment candidate dataset, forming object-level association clusters, and performing consistency judgment on the object-level association clusters to generate a labeled result set; a dual-channel module for performing dual-channel split processing on the labeled result set to form the main fusion state content and conflict explanation information; a constraint correction module for performing joint correction and conflict constraint correction on the main fusion state content and conflict explanation information to obtain a target fusion set; and an output management module for processing the target fusion set results and performing real-time output processing, while retaining fusion credibility information to obtain a multi-source fusion output set.

[0016] The beneficial effects of this invention are as follows: by establishing a parallel processing architecture of the main fusion channel and the conflict processing channel, it achieves refined processing of consistent data and conflict data, extracts the specific source, manifestation direction and degree of contradiction of the conflict, and generates structured conflict explanation information, which enhances the transparency and interpretability of the fusion results and provides a direct and clear basis for subsequent decision correction and source data quality assessment and optimization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a multi-source data fusion method.

[0019] Figure 2 This is a schematic diagram of a smart terminal for multi-source data fusion.

[0020] Figure 3 A flowchart for obtaining consistency determination information.

[0021] Figure 4 A flowchart for generating conflict explanation information.

[0022] Figure 5 The graph shows the fusion error curves under different conflict occurrence ratios.

[0023] Figure 6 Radar chart for conflict interpretation capabilities. Detailed Implementation

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0025] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0027] Reference Figures 1-6 As an embodiment of the present invention, this embodiment provides a multi-source data fusion method, including the following steps: S1. Collect raw data from multiple sources, perform unified access processing and standardization, and then write the data into a real-time cache to form a standardized access dataset.

[0028] S1.1. Receive raw data sent from different data sources through the data access interface, and attach the corresponding source identifier, receiving time identifier and data type identifier to form an access data record.

[0029] Specifically, the system receives raw data from different data sources through the data access interface. Raw data refers to data content sent directly to the data access interface from different data sources that has not yet undergone standardization processing. This includes spatial location data, object feature data, time status data, image data, text data, sensor detection data, positioning data, or status feedback data. A source identifier is added according to the source information corresponding to the raw data, a receiving time identifier is added according to the arrival time of the raw data, and a data type identifier is added according to the content category of the raw data, forming identified raw data. The identified raw data undergoes unified access processing, including unified field names, unified content sorting methods, and unified time expression formats. The source identifier, receiving time identifier, and data type identifier are then integrated with the corresponding raw data content to form access data records with consistent format and structure.

[0030] S1.2. Standardize and organize the access data records to form standard access records. Write the standard access records into the corresponding real-time cache area according to the source identifier, and summarize them to form a standardized access dataset.

[0031] Specifically, when standardizing access data records, the process involves checking whether the fields in the access data records are complete, whether the field order is consistent, and whether the time and type content correspond. Access data records with missing fields are marked as missing, access data records with duplicate fields are deduplicated, and access data records with abnormal formats are marked as abnormal, thus forming standard access records with standardized content and uniform structure. After forming standard access records, they are written to the corresponding real-time buffers according to the source identifier and arranged in the writing order according to the receiving time identifier. The standard access records in each real-time buffer are collected and organized, and the standard access records within the same processing cycle are uniformly summarized to form a standardized access dataset.

[0032] S2. Perform temporal difference identification and hierarchical compensation on the standardized access dataset and unify it to the same fusion time, outputting a temporal alignment candidate dataset.

[0033] S2.1. Extract standard access records from the standardized access dataset, identify temporal differences in each standard access record to form temporal classification information, and divide the standard access records into direct use records and compensation candidate records based on the temporal classification information.

[0034] Specifically, the source identifier, acquisition time information, reception time identifier, and record content corresponding to each standard access record are read. Using the current fusion time as the time alignment benchmark, the time difference between the acquisition time corresponding to each standard access record and the current fusion time is calculated. Combining the distribution of historical standard access records relative to the current fusion time under the same source identifier, each standard access record is classified: when the acquisition time corresponding to a standard access record is consistent with the current fusion time, or when the corresponding record content can directly participate in the fusion processing under the current fusion time, the corresponding standard access record is marked as a directly used record; when there is an offset between the acquisition time corresponding to a standard access record and the current fusion time, and the corresponding record content needs to be time-compensated before it can participate in the fusion processing under the current fusion time, the corresponding standard access record is marked as a compensation candidate record. The classification and marking results corresponding to each standard access record are summarized to form time-series classification information to characterize the time-series processing category of each standard access record, and the standard access records are divided into directly used records and compensation candidate records based on the time-series classification information.

[0035] S2.2. Perform hierarchical compensation processing on the compensation candidate records to generate compensation aligned records, and merge them with the directly used records to form a time-series aligned candidate dataset.

[0036] Specifically, for each compensation candidate record, historical standard access records related to the current fusion time under the same source identifier are read, and the compensation method is determined based on the temporal distribution relationship of the read historical standard access records relative to the current fusion time. When adjacent standard access records located before and after the current fusion time are read, and the current fusion time is located between the collection times of the adjacent standard access records, the adjacent time interpolation method is used for compensation. The collection time information and record content of the adjacent standard access records are read, and the record content is interpolated according to the relative position of the current fusion time between the adjacent standard access records to obtain the compensation method corresponding to the current fusion time. The interpolation compensation results at the previous fusion time; when no adjacent standard access records located before and after the current fusion time are read, but a nearby historical standard access record under the same source identifier is read, compensation is performed using the historical extension method. The record content of the nearby historical standard access record is read, and the record content is extended in combination with the temporal relationship between the nearby historical standard access record and the current fusion time to obtain the extension compensation result corresponding to the current fusion time; the interpolation compensation result and the extension compensation result are uniformly recorded as compensation alignment record, and then the compensation alignment record and the directly used record are collected, organized and summarized to form a temporal alignment candidate dataset.

[0037] S3. Based on the time-series aligned candidate dataset, cross-source correspondence is identified to form object-level association clusters. Consistency judgment is performed on the object-level association clusters to generate a labeled result set.

[0038] S3.1. Extract time-series alignment candidate data from different sources from the time-series alignment candidate dataset, and generate a candidate correspondence set by identifying cross-source correspondences.

[0039] Specifically, after reading the candidate data in the time-series alignment candidate dataset, the source identifier, spatial location identifier, and object feature information are extracted from each candidate data. The candidate data are then distinguished by source identifier. Candidate data from different sources are compared item by item to compare the spatial proximity and object feature similarity between candidate data from different sources. Correspondence is then identified based on the correspondence under the same fusion time condition. After the correlation identification is completed, the candidate data with corresponding relationships are classified accordingly, and the corresponding relationship content is recorded to generate a candidate correspondence set.

[0040] It should be noted that the expression for calculating the cross-source correspondence score is as follows:

[0041] ; in, Representing candidate data With candidate data Scoring of cross-source correspondence between them Representing candidate data Corresponding spatial location identifier, Representing candidate data Corresponding spatial location identifier, Representing candidate data Corresponding object feature information, Representing candidate data Corresponding object feature information, This represents the weighting coefficient corresponding to spatial location factors. This represents the weight coefficient corresponding to the characteristic factors of the object. Indicates the candidate data index variable. Indicates and Different candidate data index variables.

[0042] It should be noted that object feature information refers to descriptive information used to characterize the attribute features of the object corresponding to the candidate data, including at least one of object category information, object identification information, morphological feature information, size feature information, color feature information, texture feature information, state feature information, motion feature information, or semantic attribute information; The weighting coefficients for spatial location factors and object feature factors can be dynamically determined based on the relative discriminative power of the current candidate data in the spatial location and object feature dimensions. Specifically, this involves statistically analyzing the candidate data separately. With candidate data The contribution of spatial location factors is determined by the completeness of information in spatial location identifiers, location discrimination capability, and stability of location similarity; candidate data are also statistically analyzed. With candidate data The contribution of object feature factors is obtained by evaluating the completeness of information on object feature information, feature discrimination ability, and feature similarity stability. Then, according to the relative size of the two types of contribution, the weight coefficients corresponding to spatial location factors and object feature factors are normalized, so that the weight coefficients corresponding to strong spatial location discrimination ability and strong object feature discrimination ability are increased.

[0043] S3.2. Combine and merge the candidate corresponding relation sets to form data groups of the same object, and perform invalid association removal processing on the data groups of the same object to form object-level association clusters.

[0044] Specifically, based on the correspondence content in the candidate correspondence set, candidate data pointing to the same object are grouped together to form data combinations of the same object. Aggregation processing is then performed on these data combinations, including grouping candidate data within the data combinations according to source identifiers and uniformly organizing the candidate data according to spatial location identifiers and object feature information. After aggregation, source duplication and invalid association removal processing is performed on the data combinations of the same object. Source duplication removal processing includes retaining candidate data with the same source identifier and consistent correspondence content, while invalid association removal processing includes removing candidate data with significantly deviated spatial location identifiers or mismatched object feature information, forming object-level association clusters.

[0045] S3.3. For the associated data from different sources in the object-level association cluster, perform consistency judgment based on time state information, spatial state information and object feature information respectively to obtain consistency judgment information.

[0046] Specifically, temporal state information, spatial state information, and object feature information are extracted from each associated data set and organized according to the relationships within the same object-level association cluster. Temporal consistency is determined by comparing whether the temporal state information corresponding to each associated data set is at the same fusion time or within an adjacent range of change. Spatial consistency is determined by comparing whether the spatial state information corresponding to each associated data set maintains proximity in location and consistent direction of change. Feature consistency is determined by comparing whether the object feature information corresponding to each associated data set maintains matching feature content and similar change trends. After completing the temporal, spatial, and feature consistency determinations, the corresponding determination content is collected and organized to form consistency judgment information. This consistency judgment information characterizes the degree of consistency among the associated data sets within the object-level association cluster in terms of time, space, and feature dimensions, thus providing a basis for consistent data identification, conflict data differentiation, and subsequent fusion correction, improving the accuracy and reliability of multi-source data fusion.

[0047] S3.4. Based on the consistency determination information, assign consistency and conflict labels to each associated data in the object-level association cluster to form a label result set.

[0048] Specifically, when classifying and labeling consistency judgment information, the temporal consistency judgment results, spatial consistency judgment results, and feature consistency judgment results in the consistency judgment information are read accordingly and collected item by item according to the association relationship in the object-level association cluster. The consistency judgment information corresponding to each associated data is classified and processed. Associated data whose temporal consistency judgment results, spatial consistency judgment results, and feature consistency judgment results all meet the consistency conditions are marked as consistent associated data, and associated data in which any consistency judgment result has mismatched content is marked as conflicting associated data. After completing the classification and labeling, the consistent associated data and conflicting associated data in the object-level association cluster are organized and summarized according to the corresponding object relationship to form a labeling result set.

[0049] S4. Perform dual-channel splitting processing on the labeled result set to form the main fused state content and conflict explanation information.

[0050] S4.1. Based on consistent and conflict labels, the labeling result set is divided into consistent dataset and conflict dataset. The consistent dataset is input into the main fusion channel. The temporal state information, spatial state information and object feature information in the consistent dataset are weighted and fused to generate the main fusion state content.

[0051] Specifically, after dividing the labeling result set into consistent and conflict datasets based on consistent and conflict labels, the consistent dataset is input into the main fusion channel. The main fusion channel extracts and organizes the temporal state information, spatial state information, and object feature information in the consistent dataset according to the association relationship corresponding to the object-level association clusters, so that the temporal state information, spatial state information, and object feature information form a fusion information combination under the same object. According to the correspondence between the association data from different sources, the temporal state information, spatial state information, and object feature information are weighted. After the weight allocation is completed, the temporal state information, spatial state information, and object feature information are weighted and fused respectively. The weighted and fused temporal state information, weighted and fused spatial state information, and weighted and fused object feature information are then merged to generate the main fusion state content.

[0052] S4.2. Input the conflict dataset into the conflict processing channel, perform conflict analysis on the conflict dataset, extract conflict source information, conflict direction information and conflict degree information, and generate conflict explanation information.

[0053] Specifically, the conflict dataset is input into the conflict processing channel. The channel extracts and organizes the temporal state information, spatial state information, and object feature information in the dataset according to the association relationships corresponding to object-level clusters, ensuring alignment of related data from different sources for the same object within the same processing scope. Conflict resolution processing is then performed on the temporal state information, spatial state information, and object feature information separately. This includes comparing the positions of differences between related data from different sources, identifying the directions of differences item by item, and statistically analyzing the magnitude of differences item by item. After comparing the differences in temporal state information, spatial state information, and object feature information, the sources of the differences are organized as conflict source information, the trends of difference changes between related data from different sources are organized as conflict direction information, and the magnitude of differences between related data from different sources is organized as conflict degree information. Finally, the conflict source information, conflict direction information, and conflict degree information are merged and output to generate conflict explanation information.

[0054] It should be noted that conflict source information is used to characterize which data sources the conflicting data in the object-level association cluster come from, and to identify the source subject that caused the difference; conflict direction information is used to characterize the deviation direction of the state expression of the data from different sources in the object-level association cluster, and to reflect the inconsistency of each source in terms of change trend, value direction or state judgment; conflict degree information is used to characterize the magnitude of the difference between the data from different sources in the object-level association cluster, and to reflect the strength of the impact of the conflict.

[0055] S4.3. Dual channels include a main fusion channel and a conflict resolution channel. The main fusion channel is the processing path that performs fusion processing on consistent data, while the conflict resolution channel is the processing path that performs conflict resolution processing on conflicting data.

[0056] Specifically, based on the consistent and conflicting tags corresponding to each associated data in the tagging result set, dual-channel splitting is performed on the associated data within the object-level associated cluster; associated data with consistent tags are assigned to the main fusion channel, and associated data with conflicting tags are assigned to the conflict processing channel, while maintaining the correspondence between the input data of the two channels under the same object-level associated cluster identifier. For the consistent data allocated to the main fusion channel, it is organized according to the receiving time identifier, spatial location identifier and object feature information to form a consistent data fusion combination. Based on the differences in timeliness, location proximity, feature matching and source stability of related data from different sources, the corresponding fusion weights are determined respectively. Then, weighted calculation, merged expression and state integration processing are performed on the state information of each dimension in the consistent data fusion combination to obtain the object-level main fusion state content. For conflict data assigned to the conflict processing channel, conflict analysis is performed on the difference distribution of each conflict-related data in the same object-level association cluster in terms of reception time identifier, spatial location identifier, and object feature information. The difference location, difference direction identification, difference magnitude quantification, and conflict source attribution processing are performed in sequence to extract conflict source information, conflict direction information, and conflict degree information to form conflict explanation information associated with the corresponding main fusion state content.

[0057] S5. Perform joint correction and conflict constraint modification on the main fusion state content and conflict interpretation information to obtain the target fusion set.

[0058] S5.1. Match the main fusion state content with the conflict interpretation information to form an object-level correction data set. Based on the conflict source, conflict direction and conflict degree in the object-level correction data set, perform deviation correction and expression adjustment on the corresponding state content in the main fusion state content to form corrected fusion content.

[0059] Specifically, the main fusion state content and conflict explanation information are matched item by item according to the object correspondence. The main fusion state content and conflict description information corresponding to the same object are collected and organized to form an object-level correction data group. Based on the conflict description information in the object-level correction data group, the current state expression content of the object in the main fusion state content is jointly corrected. The joint correction process includes comparing the conflict source content, conflict direction content, and conflict degree content recorded in the conflict description information, and applying the comparison content to the current state expression content of the object in the main fusion state content to complete the correction and organization of the current state expression content of the object, forming corrected fusion content. The corrected fusion content is the result obtained after correcting the deviation and adjusting the expression of the corresponding state content in the main fusion state content based on the conflict source information, conflict direction information, and conflict degree information in the conflict explanation information.

[0060] S5.2. Identify and correct content affected by conflicts in the fusion content, and perform scope restriction, direction constraint and source correction processing on the content affected by conflicts based on the conflict description information to form a target fusion set.

[0061] Specifically, when correcting the content of the correction fusion based on the conflict description information in the object-level correction data set, the conflict description information in the object-level correction data set is read, and the conflict source, conflict direction, and conflict degree content in the conflict description information are matched item by item with the current state expression content of the object in the correction fusion content; the content affected by the conflict is constrained, including restricting the expression content corresponding to the conflict direction content, converging the expression range corresponding to the conflict degree content, and correcting and organizing the expression content corresponding to the conflict source content to form the constraint correction result; after completing the constraint processing of each constraint correction result, the constraint correction results are collected and summarized according to the object correspondence to form the target fusion set.

[0062] It should be noted that, Figure 5 This is used to illustrate the effect of controlling fusion error under different conflict occurrence ratios. Figure 5The horizontal axis represents the proportion of conflicts, and the vertical axis represents the average error. The three curves in the figure correspond to the single-channel fusion error, the main fusion state content error, and the target fusion set error, respectively. The single-channel fusion error reflects the overall deviation of the existing single processing mode when conflicting data is mixed in. The main fusion state content error reflects the initial output effect of the present invention after completing the construction of object-level association clusters, generation of labeled result sets, and splitting of the main fusion channel / conflict processing channel. The target fusion set error reflects the output effect of the present invention after completing joint correction and conflict constraint correction by combining conflict interpretation information. If the "target fusion set error" curve in the figure is always lower than the other two curves, it indicates that the present invention, through dual-channel parallel processing, first separates consistent data from conflicting data, and then uses the conflict source, conflict direction, and conflict degree to correct the main fusion state content, which can more effectively suppress the pollution of the fusion result by conflicting data, thus demonstrating the beneficial effect of more stable and reliable fusion results.

[0063] S6. Process the results of the target fusion set and output them in real time, while retaining the fusion credibility information to obtain the multi-source fusion output set.

[0064] S6.1. Collect the constraint correction results in the target fusion set according to the object-level association relationship to form an object-level output group.

[0065] Specifically, when processing the fusion results of each target in the target fusion set, each target fusion result is read item by item according to the object correspondence, and the target fusion results belonging to the same object are merged together to form an object-level output group.

[0066] S6.2. Organize the state expression content and conflict constraint correction content of the object-level output group to form object-level result data, and extract the corresponding fusion credibility information from the constraint correction results.

[0067] The target fusion results in each object-level output group are organized, including classifying the state expression content, conflict constraint correction content, and fusion credibility information corresponding to the same object, and arranging and integrating them according to a unified output order; after completing the collection processing and result organization, the organized content in each object-level output group is output as object-level result data.

[0068] S6.3. Associate the fusion credibility information with the object-level result data to form a multi-source fusion output set.

[0069] Specifically, the process involves reading the object-level result data, fusion credibility information, and the correspondence between the result receivers, verifying each correspondence item by item to clarify the location of the result receiver corresponding to each piece of object-level result data. After verifying the correspondence between the result receivers, the object-level result data is sent to the corresponding result receivers one by one according to the correspondence. During the sending process, the sending records of the object-level result data are registered. Then, the fusion credibility information is associated with the object-level result data. The association process includes matching and binding the fusion credibility information item by item according to the content range, output range, and sending range of the object-level result data. After completing the sending of object-level result data and the association of fusion credibility information, the sent object-level result data and the associated fusion credibility information are centrally organized to form a multi-source fusion output set.

[0070] It should be noted that the expression for calculating the fusion credibility information is as follows:

[0071] ; in, This indicates the integration of credibility information. This represents the average consistency level of the consistency determination information. This indicates the average degree of conflict in the conflict description information. This indicates the number of valid sources participating in the generation of content in the main fusion state. This indicates the total number of sources in the object-level association cluster. This represents the weighting coefficient corresponding to the average level of consistency. This represents the weighting coefficient corresponding to the average degree of conflict. This represents the weighting coefficient corresponding to the effective participation ratio of the source, where, , and The results were optimized and determined based on historical sample evaluations.

[0072] Furthermore, the weighting coefficients corresponding to the average consistency level, the average conflict level, and the effective participation ratio of sources can be jointly determined through historical sample evaluation results. Historical samples with fusion quality labels are selected, and the influence of the average consistency level, average conflict level, and effective participation ratio of sources on the final fusion quality are statistically analyzed. Initial weights are assigned based on the correlation, contribution, or sensitivity between the three factors and the fusion quality evaluation results. Then, the weights are iteratively adjusted through multiple rounds of sample validation so that factors with larger values ​​correspond to higher weights, resulting in a combination of weighting coefficients suitable for fusion credibility calculation.

[0073] It should be noted that, Figure 6Used to characterize the overall performance of the invention in terms of conflict interpretation capability; Figure 6 The quality of the structured conflict interpretation information output by the conflict processing channel can be displayed in the form of a radar chart, based on evaluation dimensions such as the accuracy of conflict source identification, the accuracy of conflict direction identification, the accuracy of conflict degree estimation, and the completeness of structured conflict interpretation information. The larger the value of each evaluation dimension, the stronger the ability of the present invention to identify, analyze, and structure conflict information. Figure 6 The closer to the outer ring, the stronger the ability of this invention to extract conflict source information, conflict direction information, and conflict degree information from object-level association clusters with conflict markers, and to form structured conflict explanation information, reflecting better transparency and interpretability of the fusion output.

[0074] This embodiment also provides a multi-source data fusion intelligent terminal, including: a data access module, used to collect multi-source raw data and perform unified access processing and standardization before writing it into a real-time cache to form a standardized access dataset; a time-series alignment processing module, used to perform time-series difference identification and hierarchical compensation processing on the standardized access dataset and unify it to the same fusion time, outputting a time-series alignment candidate dataset; a consistency judgment module, used to identify cross-source correspondences based on the time-series alignment candidate dataset, form object-level association clusters, and perform consistency judgment on the object-level association clusters to generate a labeled result set; a dual-channel module, used to perform dual-channel split processing on the labeled result set to form the main fusion state content and conflict explanation information; a constraint correction module, used to perform joint correction and conflict constraint correction on the main fusion state content and conflict explanation information to obtain a target fusion set; and an output management module, used to process the results of the target fusion set and perform real-time output processing, while retaining fusion credibility information to obtain a multi-source fusion output set.

[0075] In summary, this invention achieves refined processing of consistent and conflicting data by establishing a parallel processing architecture for the main fusion channel and the conflict handling channel. It extracts the specific source, manifestation direction, and degree of contradiction of the conflict and generates structured conflict explanation information, thereby enhancing the transparency and interpretability of the fusion results and providing a direct and clear basis for subsequent decision correction and source data quality assessment and optimization.

[0076] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A multi-source data fusion method, characterized in that, include: After collecting raw data from multiple sources and performing unified access processing and standardized organization, the data is written into a real-time cache to form a standardized access dataset. The standardized access dataset is subjected to temporal difference identification and hierarchical compensation processing and unified to the same fusion time, outputting a temporal alignment candidate dataset; Cross-source correspondence is identified based on the time-series aligned candidate dataset, forming object-level association clusters. Consistency judgment is then performed on the object-level association clusters to generate a labeled result set. The labeled result set is processed in a dual-channel split to form the main fused state content and conflict explanation information; The target fusion set is obtained by jointly correcting the main fusion state content and conflict interpretation information and refining the conflict constraints. The target fusion set is processed and output in real time, and the fusion credibility information is retained to obtain a multi-source fusion output set.

2. The multi-source data fusion method as described in claim 1, characterized in that, The process involves collecting raw data from multiple sources, performing unified access processing and standardization, and then writing it into a real-time cache to form a standardized access dataset. The specific steps are as follows: The system receives raw data from different data sources through the data access interface and attaches corresponding source identifiers, receiving time identifiers, and data type identifiers to form access data records. The access data records are standardized and organized to form standard access records. The standard access records are written into the corresponding real-time cache according to the source identifier and then summarized to form a standardized access dataset.

3. The multi-source data fusion method as described in claim 2, characterized in that, The steps for identifying temporal differences in the standardized access dataset, performing hierarchical compensation processing, and unifying it to the same fusion time, to output a temporally aligned candidate dataset, are as follows: Standard access records are extracted from the standardized access dataset, and temporal differences are identified for each standard access record to form temporal classification information. Based on the temporal classification information, the standard access records are divided into direct use records and compensation candidate records. Hierarchical compensation processing is performed on the compensation candidate records to generate compensation aligned records, which are then merged with the directly used records to form a time-series aligned candidate dataset.

4. The multi-source data fusion method as described in claim 1, characterized in that, The specific steps for identifying cross-source correspondences based on the temporally aligned candidate dataset to form object-level association clusters are as follows: Temporal alignment candidate data from different sources are extracted from the temporal alignment candidate dataset, and a candidate correspondence set is generated by cross-source correspondence identification. The candidate corresponding relationship sets are combined and merged to form data groups of the same object, and invalid associations of the data groups of the same object are removed to form object-level association clusters.

5. The multi-source data fusion method as described in claim 1, characterized in that, The specific steps for performing consistency judgment on object-level association clusters and generating a labeled result set are as follows: Based on time status information, spatial status information, and object feature information, consistency judgment is performed on the association data from different sources in the object-level association cluster to obtain consistency judgment information. Based on the consistency determination information, each associated data in the object-level association cluster is assigned a consistency label and a conflict label, forming a label result set.

6. The multi-source data fusion method as described in claim 1, characterized in that, The specific steps for performing dual-channel splitting processing on the labeled result set to form the main fused state content and conflict explanation information are as follows: Based on consistent and conflicting labels, the labeling result set is divided into consistent datasets and conflicting datasets. The consistent dataset is input into the main fusion channel, and the temporal state information, spatial state information, and object feature information in the consistent dataset are weighted and fused to generate the main fusion state content. Input the conflict dataset into the conflict processing channel, perform conflict analysis on the conflict dataset, extract conflict source information, conflict direction information and conflict degree information, and generate conflict explanation information.

7. The multi-source data fusion method as described in claim 1, characterized in that, The dual channels include a main fusion channel and a conflict resolution channel. The main fusion channel is the processing path that performs fusion processing on consistent data, and the conflict resolution channel is the processing path that performs conflict resolution processing on conflicting data.

8. The multi-source data fusion method as described in claim 7, characterized in that, The joint correction and conflict constraint adjustment of the main fusion state content and conflict interpretation information to obtain the target fusion set are as follows: The main fusion state content is matched with the conflict interpretation information to form an object-level correction data set. Based on the conflict source, conflict direction and conflict degree in the object-level correction data set, the corresponding state content in the main fusion state content is corrected for deviation and the expression is adjusted to form the corrected fusion content. Identify conflict-affected content in the fusion content and perform scope restriction, direction constraint and source correction processing on the conflict-affected content based on the conflict description information to form a target fusion set.

9. The multi-source data fusion method as described in claim 8, characterized in that, The process of organizing and outputting the target fusion set in real time, while retaining fusion credibility information, to obtain a multi-source fusion output set, is as follows: The constraint correction results in the target fusion set are aggregated according to the object-level association relationship to form an object-level output group; The state expression content and conflict constraint correction content of the object-level output group are organized to form object-level result data, and the corresponding fusion credibility information is extracted from the constraint correction results. The fusion credibility information is correlated with the object-level result data to form a multi-source fusion output set.

10. A multi-source data fusion intelligent terminal, based on the multi-source data fusion method according to any one of claims 1 to 9, characterized in that, include: The data access module is used to collect raw data from multiple sources, perform unified access processing and standardization, and then write the data into a real-time cache to form a standardized access dataset. The temporal alignment processing module is used to identify temporal differences and perform hierarchical compensation processing on the standardized access dataset and unify it to the same fusion time, outputting a temporal alignment candidate dataset; The consistency decision module is used to identify cross-source correspondences based on the time-series aligned candidate dataset, form object-level association clusters, and perform consistency judgment on the object-level association clusters to generate a labeled result set. The dual-channel module is used to perform dual-channel split processing on the labeled result set to form the main fused state content and conflict explanation information; The constraint correction module is used to jointly correct and correct the conflict constraint of the main fusion state content and conflict interpretation information to obtain the target fusion set. The output management module is used to organize and process the results of the target fusion set in real time, and retain the fusion credibility information to obtain the multi-source fusion output set.

Images