Deep network traffic prediction method and system for multi-source data denoising verification

Abstract

The invention relates to the technical field of traffic prediction, and discloses a deep network traffic prediction method and system for multi-source data denoising verification, and the method comprises the steps: carrying out the multi-modal noise separation of original multi-source data, and obtaining the noise distribution characteristics and noise suppression data of a deep network; performing hierarchical instantiation on the denoising parameter configuration of the noise distribution characteristics to obtain a multi-level verification mechanism of the deep network; based on a multi-stage verification mechanism, performing iterative verification on the noise suppression data to obtain verification data of the deep network; tensor fusion is carried out on the time dependence relation and the space correlation characteristics of the verification data to obtain a space-time correlation graph of the deep network; mode track tracing is carried out on the spatial-temporal feature representation of the spatial-temporal correlation map to obtain an evolution law of the deep network; performing situation projection deduction on the deep network based on an evolution law to obtain a prediction result of the deep network; according to the invention, the efficiency of deep network traffic prediction of multi-source data denoising verification can be improved.

Description

Technical Field

[0001] This invention relates to the field of traffic prediction technology, and in particular to a deep network traffic prediction method and system for multi-source data denoising and verification. Background Technology

[0002] In deep network traffic prediction, multi-source data contains various modal attributes such as images, text, and numerical values. Existing technologies struggle to achieve accurate multimodal noise separation and cannot accurately extract noise distribution features, resulting in inconsistent quality of noise-suppressed data and posing potential risks to subsequent prediction work. At the same time, existing denoising parameter configurations lack a scientific hierarchical instantiation method, making it difficult to form an effective verification mechanism and comprehensively verify the processed data, further reducing data reliability.

[0003] Existing technologies fail to fully explore the intrinsic connection between temporal dependencies and spatial correlation features when processing the spatiotemporal characteristics of verification data. Tensor fusion results are poor, making it difficult to construct accurate spatiotemporal correlation maps. Furthermore, the tracing of the pattern trajectory represented by spatiotemporal features is not deep enough, making it impossible to accurately extract the evolution law of network traffic. This results in a lack of scientific basis for situation projection inference, ultimately leading to insufficient accuracy and reliability of deep network traffic prediction results, which are difficult to meet the needs of practical application scenarios. Summary of the Invention

[0004] This invention provides a deep network traffic prediction method and system for multi-source data denoising and verification, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a deep network traffic prediction method based on multi-source data denoising and verification, comprising:

[0006] S1. Perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network;

[0007] S2. The denoising parameter configuration of the noise distribution characteristics is instantiated in a hierarchical manner to obtain the multi-level verification mechanism of the deep network;

[0008] S3. Based on the multi-level verification mechanism, the noise suppression data is iteratively verified to obtain the verification data of the deep network;

[0009] S4. Perform tensor fusion on the temporal dependency and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network;

[0010] S5. Perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network;

[0011] S6. Based on the evolutionary pattern, perform situation projection deduction on the deep network to obtain the prediction results of the deep network.

[0012] In a preferred embodiment, the step of performing multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network includes:

[0013] The original multi-source data is classified and filtered according to the modal attributes of images, text, and numerical values ​​to obtain the modal dataset of the deep network.

[0014] Frequency domain filtering is performed on the image data of the modality dataset to obtain the image noise features of the deep network;

[0015] Semantic analysis is performed on the text data of the modal dataset to obtain the text noise features of the deep network;

[0016] Outlier points are labeled in the numerical data of the modality dataset to obtain the numerical noise features of the deep network;

[0017] The noise distribution features of the deep network are obtained by integrating the image noise features, the text noise features, and the numerical noise features.

[0018] Based on the noise distribution characteristics, the submodal dataset is reconstructed to obtain the noise suppression data of the deep network.

[0019] In a preferred embodiment, the hierarchical instantiation of the denoising parameter configuration for the noise distribution characteristics to obtain the multi-level verification mechanism of the deep network includes:

[0020] The parameter structure of the denoising parameter configuration is parsed to obtain the parameter parsing results of the deep network;

[0021] Based on the parameter parsing results, rules are configured for the primary verification rule base to obtain the primary verification mechanism of the deep network.

[0022] The logical relationships of the parameter parsing results are activated by a strategy to obtain the advanced verification mechanism of the deep network;

[0023] The primary verification mechanism and the advanced verification mechanism are logically assembled to obtain the multi-level verification mechanism of the deep network.

[0024] In a preferred embodiment, the step of iteratively verifying the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network includes:

[0025] The primary verification mechanism of the multi-level verification mechanism is invoked to perform consistency verification on the noise suppression data, thereby obtaining the primary verification data of the deep network.

[0026] Based on the conflict markers of the primary verification data, conflict repair is performed on the noise suppression data to obtain the conflict repair data of the deep network.

[0027] The advanced verification mechanism of the multi-level verification mechanism is invoked to perform an integrity assessment on the conflict repair data, thereby obtaining the data integrity report of the deep network;

[0028] Based on the missing features of the data integrity report, the conflict repair data is reconstructed to obtain the feature reconstruction data of the deep network.

[0029] The quality of the reconstructed feature data is scored to obtain the quality evaluation result of the deep network.

[0030] Based on the quality assessment results, the feature reconstruction data is finally filtered to obtain the verification data of the deep network.

[0031] In a preferred embodiment, the step of performing tensor fusion on the temporal dependencies and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network includes:

[0032] The temporal dependencies are expanded along the temporal dimension to obtain the temporal feature structure of the deep network.

[0033] The spatial correlation features are expanded in a spatial dimension to obtain the spatial dimension feature structure of the deep network.

[0034] The temporal dimension feature structure and the spatial dimension feature structure are aligned to obtain the feature combination of the deep network.

[0035] The feature combinations are subjected to correlation mining to obtain the feature association patterns of the deep network.

[0036] Based on the correlation patterns between the features, the temporal dimension feature structure and the spatial dimension feature structure are topologically integrated to obtain the spatiotemporal correlation map of the deep network.

[0037] In a preferred embodiment, the step of tracing the pattern trajectory of the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network includes:

[0038] State node identification is performed on the spatiotemporal feature representation to obtain the key state node set of the deep network;

[0039] Based on the set of key state nodes, the state transition path is reconstructed to obtain the state evolution path of the deep network;

[0040] The path evolution path is extracted to obtain the path evolution features of the deep network;

[0041] Based on the path evolution characteristics, pattern feature mining is performed to obtain the pattern feature set of the deep network;

[0042] The evolutionary rules of the deep network are obtained by refining the pattern feature set.

[0043] In a preferred embodiment, the formula for calculating the path comprehensive feature value of the path evolution feature is as follows: In the formula, For path comprehensive feature values, For the state nodes of the key state node set at the evolution time The flow amplitude function, For the state node at the evolution time The transfer cost density function, The dynamic weighting coefficient for the changes in the flow of the state node. The transfer cost adjustment parameter for the verification data. The preset time decay factor, The preset feature observation reference time, The length of the evolution observation time window for the state evolution path. For the preset gradient operator, This is the derivative of the flow rate amplitude function with respect to time.

[0044] In a preferred embodiment, the step of reconstructing the state transition path based on the set of key state nodes to obtain the state evolution path of the deep network includes:

[0045] Temporal correlation analysis is performed on the set of key state nodes to obtain the node correlation relationships of the deep network;

[0046] Paths are constructed based on the node relationships to obtain an initial path sketch of the deep network;

[0047] The path connectivity of the initial path sketch is verified to obtain the verification result of the deep network;

[0048] Based on the verification results, the initial path sketch is optimized to obtain the optimized path of the deep network;

[0049] The optimized path is fully integrated to obtain the state evolution path of the deep network.

[0050] In a preferred embodiment, the step of performing situation projection deduction on the deep network based on the evolutionary law to obtain the prediction result of the deep network includes:

[0051] Construct the situational scenario of the aforementioned evolutionary laws to obtain the evolutionary situational scenario of the deep network;

[0052] The trend of the evolutionary situation is extrapolated to obtain the situation extrapolation results of the deep network;

[0053] The consistency of the situation simulation results is verified to obtain the verification conclusion of the deep network.

[0054] Based on the verification conclusions, a reliability assessment is performed on the situation simulation results to obtain a reliability assessment report for the deep network.

[0055] Based on the reliability assessment report, path filtering is performed on the situation simulation results to obtain the prediction results of the deep network.

[0056] To address the aforementioned problems, this invention also provides a deep network traffic prediction system for multi-source data denoising and verification, the system comprising:

[0057] The data processing module is used to perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network.

[0058] The hierarchical verification module is used to perform hierarchical instantiation of the denoising parameter configuration of the noise distribution characteristics to obtain the multi-level verification mechanism of the deep network.

[0059] The data verification module is used to iteratively verify the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network.

[0060] The feature association module is used to perform tensor fusion on the temporal dependency and spatial association features of the verification data to obtain the spatiotemporal association map of the deep network.

[0061] The trajectory tracing module is used to perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network;

[0062] The result prediction module is used to perform situation projection deduction on the deep network based on the evolution law, and obtain the prediction result of the deep network.

[0063] Compared with the prior art, the present invention has the following beneficial effects:

[0064] 1. This invention accurately obtains the noise distribution characteristics and noise suppression data of deep networks through multimodal noise separation, and then iteratively verifies the noise suppression data through a hierarchical instantiation multi-level verification mechanism, which effectively improves the accuracy and completeness of the verification data and provides high-quality data support for subsequent network traffic prediction. At the same time, tensor fusion is performed on the temporal dependency and spatial correlation characteristics of the verification data to construct a comprehensive spatiotemporal correlation map, which further enhances the mining effect of correlation information between data.

[0065] 2. This invention accurately extracts the evolutionary laws of deep networks by tracing the pattern trajectories of spatiotemporal feature representations of spatiotemporal correlation graphs, and performs situation projection deduction based on these laws, which significantly improves the accuracy of network traffic prediction results. The entire method process is interconnected, which not only improves the quality of data processing, but also effectively improves the overall efficiency of deep network traffic prediction. Attached Figure Description

[0066] Figure 1 This is a flowchart illustrating a deep network traffic prediction method for multi-source data denoising and verification, provided in an embodiment of the present invention.

[0067] Figure 2 This is a functional block diagram of a deep network traffic prediction system for multi-source data denoising and verification provided in an embodiment of the present invention;

[0068] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0069] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0070] This application provides a deep network traffic prediction method with multi-source data denoising and verification. The execution entity of this method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application: a server, a terminal, etc. In other words, the deep network traffic prediction method with multi-source data denoising and verification can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0071] Reference Figure 1 The diagram shown is a flowchart illustrating a deep network traffic prediction method for multi-source data denoising and verification according to an embodiment of the present invention. In this embodiment, the deep network traffic prediction method for multi-source data denoising and verification includes:

[0072] S1. Perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network;

[0073] In this embodiment of the invention, the step of performing multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network includes:

[0074] The original multi-source data is classified and filtered according to the modal attributes of images, text, and numerical values ​​to obtain the modal dataset of the deep network.

[0075] Frequency domain filtering is performed on the image data of the modality dataset to obtain the image noise features of the deep network;

[0076] Semantic analysis is performed on the text data of the modal dataset to obtain the text noise features of the deep network;

[0077] Outlier points are labeled in the numerical data of the modality dataset to obtain the numerical noise features of the deep network;

[0078] The noise distribution features of the deep network are obtained by integrating the image noise features, the text noise features, and the numerical noise features.

[0079] Based on the noise distribution characteristics, the submodal dataset is reconstructed to obtain the noise suppression data of the deep network.

[0080] Specifically, after collecting the original multi-source data, the modal attributes of the data are identified one by one to determine whether the data belongs to image, text, or numerical modality. Image modal data is characterized by pixel matrix and color channel information, text modal data is characterized by character sequence and semantic expression, and numerical modal data is characterized by continuous or discrete numerical form. Based on the above modal attribute judgment results, the original multi-source data are classified into the corresponding modal categories, and all data under each modality are filtered out to finally obtain the submodal dataset of the deep network.

[0081] Furthermore, after acquiring the image data from the submodal dataset, the image data is first converted from the spatial domain to the frequency domain. Different frequency components are extracted by analyzing the variation law of image pixel brightness. Among them, high frequency components often correspond to noise in the image, and low frequency components correspond to the main structure of the image. Then, the noise frequency components identified in the frequency domain are removed, and the position and features of these noise components in the original image are determined by frequency domain inverse transformation. Finally, the image noise features of the deep network are obtained.

[0082] Furthermore, for the text data in the modal dataset, the text is first split into basic semantic units such as words, phrases or sentences. Then, combined with general language logic and professional semantic rules related to network traffic, it is determined whether each semantic unit conforms to normal expression logic. For example, redundant words and semantically contradictory sentences that are unrelated to network traffic are identified. These illogical semantic units are marked as text noise. The specific content and location of these noises are sorted out, and finally the text noise features of the deep network are obtained.

[0083] Furthermore, when processing numerical data in the modal dataset, the overall distribution of numerical data is first statistically analyzed to determine the normal range of most numerical values. Then, each value is compared with the normal range to identify values ​​that exceed the normal range and have no reasonable business explanation. These abnormal values ​​are clearly marked, and finally, the numerical noise characteristics of the deep network are obtained.

[0084] Furthermore, the acquired image noise features, text noise features, and numerical noise features are summarized, and the specific types, distribution locations, and influence degrees of each type of noise feature are sorted out. This information is then integrated into a unified structured description, ultimately yielding the noise distribution features of the deep network.

[0085] Furthermore, based on the noise distribution characteristics, the data segments containing noise in the submodal dataset are located, and corresponding correction measures are taken for noise in different modalities: for noisy areas in image data, the average brightness of surrounding normal pixels is used to fill them; for noisy semantic units in text data, words or sentences that conform to network traffic semantics are used to replace them; for noisy values ​​in numerical data, the average value of adjacent normal values ​​in the same time period is used to supplement them. After the correction is completed, the submodal dataset is re-integrated to finally obtain the noise suppression data of the deep network.

[0086] In summary, classifying and filtering raw multi-source data by modal attributes can clearly distinguish different types of data, avoid noise processing deviations caused by modal mixing, and provide a clear data foundation for subsequent targeted denoising.

[0087] In summary, by employing dedicated noise extraction methods for different modalities, noise details of various data types can be accurately captured, ensuring that noise distribution characteristics comprehensively reflect the noise status of multi-source data and providing an accurate basis for data reconstruction.

[0088] In summary, reconstructing noise-suppressed data based on noise distribution characteristics can effectively remove invalid interference information from multi-source data, improve data quality, provide reliable data support for subsequent iterative verification and traffic prediction of deep networks, and ensure the accuracy of prediction results.

[0089] S2. The denoising parameter configuration of the noise distribution characteristics is instantiated in a hierarchical manner to obtain the multi-level verification mechanism of the deep network;

[0090] In this embodiment of the invention, the hierarchical instantiation of the denoising parameter configuration for the noise distribution characteristics to obtain the multi-level verification mechanism of the deep network includes:

[0091] The parameter structure of the denoising parameter configuration is parsed to obtain the parameter parsing results of the deep network;

[0092] Based on the parameter parsing results, rules are configured for the primary verification rule base to obtain the primary verification mechanism of the deep network.

[0093] The logical relationships of the parameter parsing results are activated by a strategy to obtain the advanced verification mechanism of the deep network;

[0094] The primary verification mechanism and the advanced verification mechanism are logically assembled to obtain the multi-level verification mechanism of the deep network.

[0095] Specifically, first, identify all parameter types included in the denoising parameter configuration, such as noise threshold parameters, verification frequency parameters, and data repair priority parameters. Then, analyze the definition, value range, and function of each parameter in noise processing. Next, sort out the hierarchical relationship between parameters (e.g., noise threshold parameters are basic parameters, and verification frequency parameters depend on the value of noise threshold parameters) and functional association (e.g., data repair priority parameters and noise threshold parameters jointly determine the order of data repair). Finally, organize the types, functions, hierarchies, and associations of these parameters into a structured document to obtain the parameter parsing results of the deep network.

[0096] Furthermore, parameters related to basic noise verification, such as noise threshold parameters and basic data format verification parameters, are extracted from the parameter parsing results. Based on these parameters, the specific content of the primary verification rules is determined: based on the noise threshold parameters, a rule is set that "data with a noise ratio exceeding the threshold is judged as data to be repaired"; based on the basic data format verification parameters, a rule is set that "data whose format does not conform to the preset network traffic data format is judged as abnormal data". These rules are entered into the primary verification rule library, and the triggering conditions of the rules are configured. After completing the rule configuration, the primary verification mechanism of the deep network is obtained.

[0097] Furthermore, the logical relationships between parameters in the parameter parsing results are analyzed, such as "when the noise threshold parameter is higher than the preset high threshold, the data integrity verification parameter needs to be enabled simultaneously" and "when the data repair priority parameter is at the highest level, the verification frequency parameter needs to be linked to increase the verification frequency." Based on these relationships, strategies are formulated: if a relationship of "noise threshold high threshold + data integrity verification parameter" is detected in the parameter parsing results, the "forced data integrity verification under high noise threshold" strategy is automatically activated; if a relationship of "highest repair priority + high frequency verification" is detected, the "real-time high frequency verification of high priority data" strategy is automatically activated. These activated strategies are integrated into a complete set of verification logic to obtain the advanced verification mechanism of the deep network.

[0098] Furthermore, the execution order of the primary verification mechanism and the advanced verification mechanism is first determined: all noise-suppressed data first undergoes basic format and noise ratio verification through the primary verification mechanism, and data that passes the primary verification then enters the advanced verification mechanism for in-depth verification under the correlation parameters; then the result coordination rules of the two mechanisms are clarified: if the primary verification determines that the data is abnormal, it directly outputs an abnormality mark without needing to enter the advanced verification; if the advanced verification finds correlation anomalies that were not identified by the primary verification, it supplements the abnormality type and feeds it back to the primary verification rule base for optimization. According to the above order and rules, the two mechanisms are integrated into a unified verification process to obtain the multi-level verification mechanism of the deep network.

[0099] In summary, structural analysis of the denoising parameter configuration can clearly clarify the function and relationship of the parameters, providing accurate parameter basis for the construction of subsequent hierarchical verification mechanisms and avoiding deviations in verification rules due to ambiguity in parameter understanding.

[0100] In summary, configuring a primary verification mechanism based on the parsing results can build a basic and comprehensive verification defense, quickly filter obvious abnormal data, reduce the processing pressure of subsequent advanced verification, and improve overall verification efficiency.

[0101] In summary, by activating advanced verification mechanisms through logical association, deep verification under parameter linkage can be achieved, making up for the limitations of primary verification and ensuring accurate identification of complex and abnormal data.

[0102] In summary, by assembling the logic of primary and advanced verification mechanisms, a layered verification system of "basic filtering + deep verification" can be formed, which ensures both the comprehensiveness and efficiency of verification, and provides a reliable mechanism to support the iterative verification of noise-suppressed data.

[0103] S3. Based on the multi-level verification mechanism, the noise suppression data is iteratively verified to obtain the verification data of the deep network;

[0104] In this embodiment of the invention, the step of iteratively verifying the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network includes:

[0105] The primary verification mechanism of the multi-level verification mechanism is invoked to perform consistency verification on the noise suppression data, thereby obtaining the primary verification data of the deep network.

[0106] Based on the conflict markers of the primary verification data, conflict repair is performed on the noise suppression data to obtain the conflict repair data of the deep network.

[0107] The advanced verification mechanism of the multi-level verification mechanism is invoked to perform an integrity assessment on the conflict repair data, thereby obtaining the data integrity report of the deep network;

[0108] Based on the missing features of the data integrity report, the conflict repair data is reconstructed to obtain the feature reconstruction data of the deep network.

[0109] The quality of the reconstructed feature data is scored to obtain the quality evaluation result of the deep network.

[0110] Based on the quality assessment results, the feature reconstruction data is finally filtered to obtain the verification data of the deep network.

[0111] Specifically, the primary verification mechanism is invoked from the multi-level verification mechanism. First, three key attributes—data format, numerical range, and semantic logic—are extracted from the noise-suppressed data. The data format must match the network traffic data format preset by the deep network, the numerical range must conform to the normal fluctuation range of this type of network traffic, and the semantic logic must ensure that the text data is consistent with the preset standards. Then, the extracted key attributes are compared with the preset standards one by one. If a part of the data has a format error, a value that is out of range, or a semantic contradiction, it is marked as conflicting data and the conflict location and type are recorded. Finally, the data that has passed the comparison and the data marked as conflicting are organized to form a dataset containing conflict markers, which is the primary verification data of the deep network.

[0112] Furthermore, based on the conflict markers in the initial verification data, specific data segments with conflicts in the noise suppression data are located. For example, data with a missing timestamp field marked as "format error", traffic values ​​exceeding the normal range for a certain period marked as "numerical anomaly", and text mentioning both "traffic interruption" and "traffic peak" marked as "semantic contradiction". Then, repair measures are taken for different conflict types: missing fields are added to data with format errors in a preset field order; abnormal values ​​are replaced with the average of adjacent normal data in the same time period; and semantically contradictory data is replaced with logically consistent expressions. After all conflict repairs are completed, the repaired data is integrated to obtain the conflict-repaired data of the deep network.

[0113] Furthermore, the advanced verification mechanism in the multi-level verification mechanism is invoked to first determine the integrity dimensions that the conflict repair data needs to cover, including core field integrity, time series integrity, and related data integrity; then, the compliance of the conflict repair data with each integrity dimension is checked one by one, and the missing core field names, missing time segments, and missing related data source information are recorded; finally, the check results and missing information are organized into a structured report to obtain the data integrity report of the deep network.

[0114] Furthermore, missing features are first extracted from the data integrity report to clarify the missing type and location; then, feature reconstruction is performed for different missing types: when core fields are missing, the missing field content is deduced and supplemented based on the field patterns of other data from the same data source; when time series data are missing, linear interpolation is used to supplement the traffic data for the missing time period; when related data are missing, a mapping relationship is established through existing data from similar data sources to deduce and supplement the missing data; after completing feature reconstruction, the data is integrated to obtain the feature reconstruction data of the deep network.

[0115] Furthermore, three dimensions are set for data quality scoring: accuracy, completeness, and consistency, with each dimension having a maximum score of 100 points. The performance of the feature reconstruction data in each dimension is then examined. For example, accuracy is scored by comparing it with historical real traffic data, completeness is scored based on the proportion of data without missing data, and consistency is scored by checking the format and logic. Finally, the average score of the three dimensions is calculated to form a document containing specific scores and evaluations of each dimension, thus obtaining the quality assessment results of the deep network.

[0116] Furthermore, a quality standard for the feature reconstruction data required by the deep network is pre-defined, such as an average score of no less than 80 points and scores of no less than 70 points for each dimension. The quality assessment results of the feature reconstruction data are then compared with the standard to select data that meets the standard. If some data are close to the standard but do not meet it, it is necessary to reconfirm whether they meet the core requirements of traffic prediction and retain only the data that meets the core requirements and meets the standard. Finally, all the filtered data are integrated to obtain the verification data of the deep network.

[0117] In summary, using a primary verification mechanism for consistency verification can quickly filter out noise and suppress obvious conflicts in the data, providing a clear direction for subsequent repairs and reducing the interference of invalid data on the verification process.

[0118] In summary, using conflict markers to repair noise-suppressed data can effectively address data inconsistency issues, improve data accuracy, and lay a high-quality data foundation for subsequent integrity assessments.

[0119] In summary, the integrity assessment of the advanced verification mechanism can comprehensively identify and resolve missing data issues related to conflict repair, and the resulting report provides precise guidance for feature reconstruction, avoiding the impact of missing data on subsequent predictions.

[0120] In summary, reconstructing data based on missing features and combining it with quality scoring for screening can ensure that the final validation data is both complete and of high quality, providing reliable data support for subsequent spatiotemporal feature fusion and traffic prediction in deep networks, and ensuring the accuracy of prediction results.

[0121] S4. Perform tensor fusion on the temporal dependency and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network;

[0122] In this embodiment of the invention, the step of performing tensor fusion on the temporal dependency and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network includes:

[0123] The temporal dependencies are expanded along the temporal dimension to obtain the temporal feature structure of the deep network.

[0124] The spatial correlation features are expanded in a spatial dimension to obtain the spatial dimension feature structure of the deep network.

[0125] The temporal dimension feature structure and the spatial dimension feature structure are aligned to obtain the feature combination of the deep network.

[0126] The feature combinations are subjected to correlation mining to obtain the feature association patterns of the deep network.

[0127] Based on the correlation patterns between the features, the temporal dimension feature structure and the spatial dimension feature structure are topologically integrated to obtain the spatiotemporal correlation map of the deep network.

[0128] Specifically, the time dependencies contained in the verification data are first sorted out to clarify the correlation logic of the data in the time dimension, such as the sequential relationship between network traffic data at different time points and the time order of traffic peaks and valleys. Then, these time dependencies are broken down and expanded according to a fixed time granularity. The information such as traffic value, traffic change amplitude, and traffic change direction in each time unit is organized into a structured sequence form to form an overall structure containing time unit identifiers, corresponding traffic characteristics, and time correlation logic, thus obtaining the time dimension feature structure of the deep network.

[0129] Furthermore, the spatial correlation feature scope involved in the verification data is first determined, and the correlation objects of the data in the spatial dimension are clarified, such as the traffic transmission relationship between different network nodes and the traffic distribution relationship of multiple nodes in the same area. Then, these spatial correlation features are decomposed and expanded according to the hierarchy of spatial nodes, and the identifier of each spatial node, the traffic data corresponding to the node, the connection method of the node with other nodes and the amount of traffic interaction are recorded. The data is organized into an overall structure containing spatial node information, node traffic characteristics and spatial correlation logic, thus obtaining the spatial dimension feature structure of the deep network.

[0130] Furthermore, a unified time identifier is first extracted from the time dimension feature structure, and a unified spatial identifier is extracted from the spatial dimension feature structure. Then, based on the time identifier, the time dimension features under the same time identifier are matched one by one with the spatial dimension features under the corresponding spatial identifier to ensure that each time-space identifier combination corresponds to complete time and spatial features. Finally, these matched features are integrated in the format of "time identifier - spatial identifier - feature content" to obtain the feature combination of the deep network.

[0131] Furthermore, all features in the feature combination are first classified to distinguish between temporal and spatial features; then the correspondence between different categories of features is analyzed, for example, when a certain temporal feature appears, which spatial features will appear simultaneously, or when a certain spatial feature appears, which temporal features will appear subsequently; finally, these frequently occurring feature correspondences are organized into a regular description to obtain the feature association pattern of the deep network.

[0132] Furthermore, based on the correlation patterns between features, the integration rules for the time-dimensional feature structure and the spatial-dimensional feature structure are first determined. For example, if the correlation pattern shows that "the core node traffic interaction volume increases" is often accompanied by "the traffic peak after 10 minutes", then during integration, the spatial feature needs to be directly correlated with the corresponding subsequent time feature. Then, a framework of the topology structure is constructed, with the time axis and spatial node hierarchy as the basic framework. The time units in the time-dimensional feature structure and the spatial nodes in the spatial-dimensional feature structure are respectively used as nodes of the topology structure. Based on the correlation pattern, the time nodes and spatial nodes with correlation are connected by lines, and the correlation type is labeled. Finally, the information labeling of the topology structure is improved to ensure that each correlation corresponds to a specific feature description, thus obtaining the spatiotemporal correlation map of the deep network.

[0133] In summary, unfolding the time dependency relationship yields a time-dimensional feature structure, which can clearly present the changing patterns of the verification data over time, providing a clear time feature foundation for subsequent spatiotemporal fusion.

[0134] In summary, expanding spatial correlation features yields spatial dimensional feature structures, which can systematically analyze and verify the spatial distribution and interaction relationships of data, providing complete spatial feature support for spatiotemporal fusion.

[0135] In summary, aligning spatiotemporal feature structures to form feature combinations ensures that spatiotemporal features match under the same dimensional benchmark, avoiding fusion deviations caused by dimensional misalignment.

[0136] In summary, mining the correlation patterns between features and integrating them topologically to form a spatiotemporal correlation map can intuitively present the intrinsic connections between spatiotemporal features, providing a clear map basis for subsequent pattern trajectory tracing and evolutionary law analysis.

[0137] S5. Perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network;

[0138] In this embodiment of the invention, the step of tracing the pattern trajectory of the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network includes:

[0139] State node identification is performed on the spatiotemporal feature representation to obtain the key state node set of the deep network;

[0140] Based on the set of key state nodes, the state transition path is reconstructed to obtain the state evolution path of the deep network;

[0141] The path evolution path is extracted to obtain the path evolution features of the deep network;

[0142] Based on the path evolution characteristics, pattern feature mining is performed to obtain the pattern feature set of the deep network;

[0143] The evolutionary rules of the deep network are obtained by refining the pattern feature set.

[0144] In this embodiment of the invention, the formula for calculating the path comprehensive feature value of the path evolution feature is as follows: In the formula, For path comprehensive feature values, For the state nodes of the key state node set at the evolution time The flow amplitude function, For the state node at the evolution time The transfer cost density function, The dynamic weighting coefficient for the changes in the flow of the state node. The transfer cost adjustment parameter for the verification data. The preset time decay factor, The preset feature observation reference time, The length of the evolution observation time window for the state evolution path. For the preset gradient operator, This is the derivative of the flow rate amplitude function with respect to time.

[0145] In this embodiment of the invention, the step of reconstructing the state transition path based on the set of key state nodes to obtain the state evolution path of the deep network includes:

[0146] Temporal correlation analysis is performed on the set of key state nodes to obtain the node correlation relationships of the deep network;

[0147] Paths are constructed based on the node relationships to obtain an initial path sketch of the deep network;

[0148] The path connectivity of the initial path sketch is verified to obtain the verification result of the deep network;

[0149] Based on the verification results, the initial path sketch is optimized to obtain the optimized path of the deep network;

[0150] The optimized path is fully integrated to obtain the state evolution path of the deep network.

[0151] Specifically, the spatiotemporal feature representation in the spatiotemporal correlation graph is first analyzed to identify the network state-related features contained therein, such as the traffic peak of the core node in a specific time unit, the abnormal offline state of a node in a certain area, and the sudden increase / decrease in traffic transmission between different nodes. Then, based on the degree of influence of these features on the evolution of network traffic, the state features that have a direct driving effect on subsequent traffic trend changes are retained, while secondary state features with no significant impact are eliminated. Each key state feature selected is assigned to an independent state node, and finally integrated to form the set of key state nodes of the deep network.

[0152] Furthermore, each node in the set of key state nodes is first labeled with its corresponding occurrence time and spatial location. Then, the connection relationship between the nodes is sorted out in chronological order, for example, "peak traffic of core node A at time t1" → "sudden increase in traffic of surrounding nodes of node A at time t2" → "short offline occurrence of surrounding nodes at time t3". Next, the logical correlation between adjacent nodes is verified, nodes with logical breaks are removed, and missing nodes due to feature omissions are added. Finally, the sorted nodes are connected into a complete path according to the time sequence and logical relationship to obtain the state evolution path of the deep network.

[0153] Furthermore, key information reflecting the characteristics of the path is extracted from the state evolution path, including the time interval between adjacent state nodes in the path, the transition direction between state nodes, the duration of each state node in the path, and the frequency of occurrence of similar state nodes in the path. This extracted information is classified and organized into structured feature terms to obtain the path evolution features of the deep network.

[0154] Furthermore, statistical analysis is performed on the path evolution characteristics to identify recurring feature combinations, such as feature sequences that appear multiple times in different time periods or spatial regions, like "peak traffic at core node → sudden increase in traffic at adjacent nodes → offline of one peripheral node". The commonalities of these recurring patterns are then summarized, such as triggering conditions and evolution cycles. Each summarized commonality pattern is treated as an independent pattern feature and integrated to form the pattern feature set of the deep network.

[0155] Furthermore, each pattern feature in the pattern feature set is analyzed in depth to extract the stable logical patterns behind it. For example, from the pattern feature of "core node traffic peak → surrounding node traffic surge → surrounding node offline", the pattern "when the core node traffic reaches 90% of the bandwidth limit and lasts for more than 5 minutes, it will trigger a surge in surrounding node traffic within 2 minutes, which will lead to 1 to 2 surrounding nodes going offline for 1 to 2 minutes due to excessive load" is extracted. All the extracted patterns are then organized in the format of "trigger condition - evolution process - result state" to ensure that the patterns can clearly reflect the evolution logic of the network state, and finally the evolution law of the deep network is obtained.

[0156] Specifically, the occurrence timestamp and state description of each node are first extracted from the set of key state nodes. All nodes are sorted in order of timestamp from earliest to latest. Then, the logical relationship between adjacent nodes after sorting is analyzed one by one. It is determined whether the state of the previous node can trigger the state of the next node according to the principle of network traffic transmission. Nodes that have a sequential time and are logically valid are marked as "associated node pairs". At the same time, the association attributes are recorded. Finally, all associated node pairs and association attributes are sorted out to obtain the node association relationship of the deep network.

[0157] Furthermore, based on the "associated node pairs" in the node association relationship, the earliest key state node is taken as the starting point of the path, and the nodes associated with the starting point are connected in sequence. Then, the end node after the connection is taken as the new starting point, and subsequent associated nodes are connected to gradually form a linear path framework. If a node is associated with multiple subsequent nodes at the same time, all branch paths are temporarily retained to form a preliminary structure containing the main path and branch paths, thus obtaining the initial path sketch of the deep network.

[0158] Furthermore, connectivity checks are performed on each path in the initial path sketch: first, temporal connectivity is checked to confirm whether the timestamp interval between adjacent nodes is within the normal evolution range of the network state; then, logical connectivity is checked to confirm whether the association attributes of adjacent nodes conform to the network traffic pattern. The temporal break positions, logical break types, and isolated branches without subsequent nodes in the path are recorded and compiled into a document containing defect information to obtain the verification results of the deep network.

[0159] Furthermore, based on the time breakage issues in the verification results, the verification data is backtracked, and key state nodes that conform to logic within the breakage time interval are extracted and added to the path; for the logical breakage issues, the node connection order is adjusted or the association attributes are corrected; for isolated branches, branches without actual evolutionary significance are deleted, and after the adjustment is completed, a time-continuous and logically coherent path is formed, thus obtaining the optimized path of the deep network.

[0160] Furthermore, complete information is added to each node in the optimization path, including the spatial location of the node, the duration of the node state, and the level of the node's influence on subsequent evolution. At the same time, the total evolution time of the path is calculated. This information is integrated with the path structure into a complete document to obtain the state evolution path of the deep network.

[0161] Specifically, the flow amplitude function comes from the state nodes of the key state node set. After reconstructing the state transition path of the key state node set to obtain the state evolution path, the flow data of the state node at each evolution moment is extracted and the function is constructed.

[0162] Furthermore, the transition cost density function is derived from the state nodes of the key state node set. At each evolution moment of the state evolution path, the cost distribution during the state node transition process is statistically analyzed and the function is constructed.

[0163] Furthermore, the dynamic weighting coefficients determine the coefficient value at each evolution moment by analyzing the fluctuations in the flow changes of state nodes in the set of key state nodes, based on the amplitude and frequency of the flow changes.

[0164] Furthermore, the transition cost adjustment parameter is based on the verification data. The factors affecting the cost of the verification data during the state node transition process are analyzed, and the adjustment parameter value at each evolution time is calculated.

[0165] Furthermore, the time decay factor is a fixed value set before calculation based on the actual application scenario and prediction accuracy requirements.

[0166] Furthermore, the feature observation reference time is a fixed time determined before calculation based on the time range of the state evolution path.

[0167] Furthermore, the length of the evolution observation time window is a fixed value set before calculation based on the total duration of the state evolution path and the required observation accuracy.

[0168] Furthermore, the gradient operator is a pre-defined computational tool that calculates the change in the value of the flow amplitude function at each evolution moment under a small time increment, thereby obtaining the derivative of the flow amplitude function with respect to time at that moment.

[0169] Furthermore, the calculation yields a comprehensive path feature value, which is used to extract path evolution features from the state evolution path, integrating information such as the flow, transition cost, and time decay of state nodes to comprehensively reflect the comprehensive characteristics of the state evolution path.

[0170] Furthermore, the calculated path comprehensive feature value provides core data for subsequent path feature extraction, the path evolution feature is used for pattern feature mining, and the pattern feature set is refined to obtain the evolution law of the deep network.

[0171] Furthermore, as the difference between the evolution time and the characteristic observation reference time increases, the corresponding attenuation effect strengthens, and the contribution of this part to the path comprehensive characteristic value weakens.

[0172] Furthermore, when the derivative of the flow amplitude function with respect to time increases while other parameters remain constant, the path comprehensive characteristic value increases accordingly; when the derivative decreases, the path comprehensive characteristic value decreases accordingly.

[0173] Furthermore, when the value of the transfer cost density function increases while other parameters remain unchanged, the path comprehensive characteristic value increases accordingly, and vice versa.

[0174] Furthermore, when the dynamic weighting coefficient increases and the value of the flow amplitude function is positive, the path comprehensive characteristic value increases accordingly; when the value of the flow amplitude function is negative, the path comprehensive characteristic value decreases accordingly.

[0175] In summary, identifying the set of key state nodes can accurately focus on states that have a core impact on network evolution, laying a precise node foundation for subsequent path reconstruction and avoiding secondary nodes from interfering with the accuracy of the evolution path.

[0176] In summary, reconstructing the state evolution path can transform discrete key nodes into a continuous evolution process, clearly presenting the temporal and logical relationships of network states, and providing a complete analytical object for path feature extraction.

[0177] In summary, extracting path evolution features and mining pattern feature sets can screen out representative features and recurring patterns from the evolution path, providing specific analytical basis for pattern condensation and avoiding the subjectivity of pattern extraction.

[0178] In summary, condensing evolutionary patterns can transform scattered pattern characteristics into universal evolutionary logic, providing clear pattern support for subsequent situation projection and deduction of deep networks, and ensuring the rationality and reliability of prediction results.

[0179] In summary, performing temporal correlation analysis on key state nodes can clarify the temporal and logical relationships between nodes, providing a precise basis for path construction and avoiding path chaos caused by disordered node connections.

[0180] In summary, building an initial path sketch based on relationships allows for the rapid construction of a path framework while preserving potential branches, reserving room for adjustment in subsequent optimizations, and ensuring coverage of all possible evolutionary directions.

[0181] In summary, by verifying connectivity to identify path defects, we can accurately pinpoint time and logic breaks, provide clear directions for path optimization, and ensure path continuity.

[0182] In summary, optimizing the path based on the verification results can fix defects, remove redundant branches, and form a high-quality optimized path, laying the foundation for complete integration.

[0183] In summary, the complete integration of optimized paths can supplement key node information, improve path attributes, and ultimately form a comprehensive state evolution path, providing a complete object for subsequent path feature extraction and evolution law analysis.

[0184] S6. Based on the evolutionary pattern, perform situation projection deduction on the deep network to obtain the prediction results of the deep network.

[0185] In this embodiment of the invention, the step of performing situation projection deduction on the deep network based on the evolutionary law to obtain the prediction result of the deep network includes:

[0186] Construct the situational scenario of the aforementioned evolutionary laws to obtain the evolutionary situational scenario of the deep network;

[0187] The trend of the evolutionary situation is extrapolated to obtain the situation extrapolation results of the deep network;

[0188] The consistency of the situation simulation results is verified to obtain the verification conclusion of the deep network.

[0189] Based on the verification conclusions, a reliability assessment is performed on the situation simulation results to obtain a reliability assessment report for the deep network.

[0190] Based on the reliability assessment report, path filtering is performed on the situation simulation results to obtain the prediction results of the deep network.

[0191] Specifically, the core elements of the evolutionary pattern are first broken down, including the triggering conditions of each pattern, such as "core node traffic reaches 90% of the bandwidth limit", the evolutionary stages such as "traffic surge stage", "node load increase stage", "node offline stage", and the result state. Then, combined with the actual operating environment parameters of the deep network, the corresponding actual operating scenario is matched for each evolutionary pattern. For example, the evolutionary pattern of "core node traffic peak → surrounding node traffic surge → node offline" is mapped to "core node high load scenario from 8 am to 9 am on a weekday", clarifying the time range, spatial nodes involved and initial state in the scenario. Finally, all the matched scenarios are integrated to obtain the evolutionary state scenario of the deep network.

[0192] Furthermore, for each evolutionary scenario, the state transition sequence and time interval are determined according to the corresponding evolutionary rules. For example, in the "high load scenario of core nodes during the morning peak", the deduction steps are set according to the evolutionary rules: "core node traffic reaches 90% of bandwidth at time t0 → traffic of surrounding nodes increases by 30% at time t2 → one surrounding node goes offline at time t5". At the same time, the deduction details are adjusted in combination with the real-time traffic data of the current network, the network status at each time node is recorded, and a document containing a complete time series and state description is formed to obtain the situation deduction result of the deep network.

[0193] Furthermore, the situational simulation results are compared with the historical evolution patterns and verification data of the deep network: on the one hand, it is checked whether the state transition logic in the simulation results is consistent with the historical evolution patterns; on the other hand, it is checked whether the data range in the simulation results is within the normal fluctuation range of the verification data. If both are consistent, it is judged as "consistent"; if there is one discrepancy, it is marked as "local deviation"; if there are multiple discrepancies, it is judged as "serious deviation". These judgment results are sorted out to obtain the verification conclusion of the deep network.

[0194] Furthermore, based on the statistical consistency ratio, deviation type, and deviation degree of the verification conclusions, reliability level standards are set. The reliability level of each situational projection result is evaluated against the standards, and the evaluation basis is explained. Finally, a structured evaluation document is formed, resulting in the reliability evaluation report of the deep network.

[0195] Furthermore, based on the level in the reliability assessment report, the situational simulation results of the "high reliability" level are selected and the "low reliability" results are eliminated; for the "medium reliability" results, it is further verified whether their deviation affects the core prediction target. The final simulation results are integrated in chronological order to clarify the network state of each future time period and obtain the prediction results of the deep network.

[0196] In summary, constructing evolutionary situational scenarios can ground abstract evolutionary laws in the actual network environment, providing specific analytical objects for trend projection and preventing projections from deviating from actual operating conditions.

[0197] In summary, scenario-based trend extrapolation can accurately predict network state changes based on patterns, generating structured extrapolation results that provide a clear data foundation for subsequent verification and evaluation.

[0198] In summary, consistency verification can check the matching degree between the simulation results and historical patterns and the verification data, detect deviations in a timely manner, and ensure the rationality of the simulation results.

[0199] In summary, reliability assessment clarifies the credibility of the projection results by quantifying the levels, providing an objective standard for path selection and avoiding selection bias caused by subjective judgment.

[0200] In summary, selecting highly reliable projection paths based on the evaluation report can ensure that the final prediction results are accurate and consistent with the actual evolution logic of the network, providing a reliable output for deep network traffic prediction.

[0201] like Figure 2 The diagram shown is a functional block diagram of a deep network traffic prediction system for multi-source data denoising and verification provided in an embodiment of the present invention.

[0202] The deep network traffic prediction system 100 for multi-source data denoising and verification described in this invention can be installed in an electronic device. Depending on the functions implemented, the deep network traffic prediction system 100 for multi-source data denoising and verification may include a data processing module 101, a hierarchical verification module 102, a data verification module 103, a feature association module 104, a trajectory tracing module 105, and a result prediction module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0203] In this embodiment, the functions of each module / unit are as follows:

[0204] The data processing module 101 is used to perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network.

[0205] The hierarchical verification module 102 is used to perform hierarchical instantiation of the denoising parameter configuration of the noise distribution characteristics to obtain the multi-level verification mechanism of the deep network.

[0206] The data verification module 103 is used to iteratively verify the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network.

[0207] The feature association module 104 is used to perform tensor fusion on the temporal dependency and spatial association features of the verification data to obtain the spatiotemporal association map of the deep network.

[0208] The trajectory tracing module 105 is used to perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network.

[0209] The result prediction module 106 is used to perform situation projection deduction on the deep network based on the evolution law, and obtain the prediction result of the deep network.

[0210] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0211] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0212] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0213] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0214] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0215] Finally, 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.

Claims

1. A deep network traffic prediction method based on multi-source data denoising and verification, characterized in that, The method includes: S1. Perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network; S2. The denoising parameter configuration of the noise distribution characteristics is instantiated in a hierarchical manner to obtain the multi-level verification mechanism of the deep network; S3. Based on the multi-level verification mechanism, the noise suppression data is iteratively verified to obtain the verification data of the deep network; S4. Perform tensor fusion on the temporal dependency and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network; S5. Perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network; S6. Based on the evolutionary pattern, perform situation projection deduction on the deep network to obtain the prediction results of the deep network.

2. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The process of performing multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network includes: The original multi-source data is classified and filtered according to the modal attributes of images, text, and numerical values ​​to obtain the modal dataset of the deep network. Frequency domain filtering is performed on the image data of the modality dataset to obtain the image noise features of the deep network; Semantic analysis is performed on the text data of the modal dataset to obtain the text noise features of the deep network; Outlier points are labeled in the numerical data of the modality dataset to obtain the numerical noise features of the deep network; The noise distribution features of the deep network are obtained by integrating the image noise features, the text noise features, and the numerical noise features. Based on the noise distribution characteristics, the submodal dataset is reconstructed to obtain the noise suppression data of the deep network.

3. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The denoising parameter configuration for the noise distribution characteristics is hierarchically instantiated to obtain the multi-level verification mechanism of the deep network, including: The parameter structure of the denoising parameter configuration is parsed to obtain the parameter parsing results of the deep network; Based on the parameter parsing results, rules are configured for the primary verification rule base to obtain the primary verification mechanism of the deep network. The logical relationships of the parameter parsing results are activated by a strategy to obtain the advanced verification mechanism of the deep network; The primary verification mechanism and the advanced verification mechanism are logically assembled to obtain the multi-level verification mechanism of the deep network.

4. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The step of iteratively verifying the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network includes: The primary verification mechanism of the multi-level verification mechanism is invoked to perform consistency verification on the noise suppression data, thereby obtaining the primary verification data of the deep network. Based on the conflict markers of the primary verification data, conflict repair is performed on the noise suppression data to obtain the conflict repair data of the deep network. The advanced verification mechanism of the multi-level verification mechanism is invoked to perform an integrity assessment on the conflict repair data, thereby obtaining the data integrity report of the deep network; Based on the missing features of the data integrity report, the conflict repair data is reconstructed to obtain the feature reconstruction data of the deep network. The quality of the reconstructed feature data is scored to obtain the quality evaluation result of the deep network. Based on the quality assessment results, the feature reconstruction data is finally filtered to obtain the verification data of the deep network.

5. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The step of performing tensor fusion on the temporal dependencies and spatial correlation features of the verification data to obtain the spatiotemporal correlation map of the deep network includes: The temporal dependencies are expanded along the temporal dimension to obtain the temporal feature structure of the deep network. The spatial correlation features are expanded in a spatial dimension to obtain the spatial dimension feature structure of the deep network. The temporal dimension feature structure and the spatial dimension feature structure are aligned to obtain the feature combination of the deep network. The feature combinations are subjected to correlation mining to obtain the feature association patterns of the deep network. Based on the correlation patterns between the features, the temporal dimension feature structure and the spatial dimension feature structure are topologically integrated to obtain the spatiotemporal correlation map of the deep network.

6. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The process of tracing the pattern trajectory of the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network includes: State node identification is performed on the spatiotemporal feature representation to obtain the key state node set of the deep network; Based on the set of key state nodes, the state transition path is reconstructed to obtain the state evolution path of the deep network; The path evolution path is extracted to obtain the path evolution features of the deep network; Based on the path evolution characteristics, pattern feature mining is performed to obtain the pattern feature set of the deep network; The evolutionary rules of the deep network are obtained by refining the pattern feature set.

7. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 6, characterized in that, The formula for calculating the path comprehensive feature value of the path evolution feature is as follows: In the formula, For path comprehensive feature values, For the state nodes of the key state node set at the evolution time The flow amplitude function, For the state node at the evolution time The transfer cost density function, The dynamic weighting coefficient for the changes in the flow of the state node. The transfer cost adjustment parameter for the verification data. The preset time decay factor, The preset feature observation reference time, The length of the evolution observation time window for the state evolution path. For the preset gradient operator, This is the derivative of the flow rate amplitude function with respect to time.

8. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 6, characterized in that, The process of reconstructing the state transition path based on the set of key state nodes to obtain the state evolution path of the deep network includes: Temporal correlation analysis is performed on the set of key state nodes to obtain the node correlation relationships of the deep network; Paths are constructed based on the node relationships to obtain an initial path sketch of the deep network; The path connectivity of the initial path sketch is verified to obtain the verification result of the deep network; Based on the verification results, the initial path sketch is optimized to obtain the optimized path of the deep network; The optimized path is fully integrated to obtain the state evolution path of the deep network.

9. The deep network traffic prediction method for multi-source data denoising and verification as described in claim 1, characterized in that, The process of performing situation projection deduction on the deep network based on the evolutionary pattern to obtain the prediction results of the deep network includes: Construct the situational scenario of the aforementioned evolutionary laws to obtain the evolutionary situational scenario of the deep network; The trend of the evolutionary situation is extrapolated to obtain the situation extrapolation results of the deep network; The consistency of the situation simulation results is verified to obtain the verification conclusion of the deep network. Based on the verification conclusions, a reliability assessment is performed on the situation simulation results to obtain a reliability assessment report for the deep network. Based on the reliability assessment report, path filtering is performed on the situation simulation results to obtain the prediction results of the deep network.

10. A deep network traffic prediction system for multi-source data denoising and verification, characterized in that, The system includes: The data processing module is used to perform multimodal noise separation on the original multi-source data to obtain the noise distribution characteristics and noise suppression data of the deep network. The hierarchical verification module is used to perform hierarchical instantiation of the denoising parameter configuration of the noise distribution characteristics to obtain the multi-level verification mechanism of the deep network. The data verification module is used to iteratively verify the noise suppression data based on the multi-level verification mechanism to obtain the verification data of the deep network. The feature association module is used to perform tensor fusion on the temporal dependency and spatial association features of the verification data to obtain the spatiotemporal association map of the deep network. The trajectory tracing module is used to perform pattern trajectory tracing on the spatiotemporal feature representation of the spatiotemporal correlation graph to obtain the evolution law of the deep network; The result prediction module is used to perform situation projection deduction on the deep network based on the evolution law, and obtain the prediction result of the deep network.

Images