CN-121980179-A - Heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation

CN121980179ACN 121980179 ACN121980179 ACN 121980179ACN-121980179-A

Abstract

The invention discloses a heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation, which relates to the technical field of heterogeneous information processing, and comprises the steps of receiving a collaborative task request, carrying out semantic analysis and encoding into a task state vector; the method comprises the steps of responding to a task request, receiving multi-domain streaming heterogeneous data blocks, completing time sequence deviation correction and alignment based on a task state vector dynamic modulation increment time sequence alignment strategy, outputting a preliminary alignment feature, evaluating dynamic feature trust modulus based on the task state vector and the preliminary alignment feature, generating task adaptation fusion feature by weighting fusion, generating a collaborative service result, generating a feedback signal according to the difference between a real-time effect index and an expected target, and updating task state vector codes. According to the invention, task semantic understanding and dynamic adaptation of stream data processing are realized through the double-circulation driving framework, the pertinence, reliability and instantaneity of cooperative processing are improved, and the method is suitable for heterogeneous information cross-domain cooperative processing requirements under multiple scenes.

Inventors

XU LUYAO
WANG XIAOXIN
LIU CHUNQIU

Assignees

连云港市规划展示中心
连云港市水利规划设计院有限公司

Dates

Publication Date: 20260505
Application Date: 20260120

Claims (10)

1. The heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation is characterized by comprising the following steps of: Step one, receiving a cooperative task request, carrying out semantic analysis on the cooperative task request, extracting task semantic elements, and encoding the task semantic elements into task state vectors, wherein the task state vectors are micro multidimensional vectors; Step two, responding to the collaborative task request, receiving stream type heterogeneous data blocks from at least two different domains in real time, dynamically modulating an increment time sequence alignment strategy for the stream type heterogeneous data blocks based on the task state vector, correcting and aligning time sequence deviation for the stream type heterogeneous data blocks according to the increment time sequence alignment strategy after modulation, and outputting a preliminary alignment characteristic; thirdly, based on the task state vector and the preliminary alignment feature, evaluating the dynamic feature trust modulus of the preliminary alignment feature, and carrying out weighted fusion on the preliminary alignment feature from different domains according to the dynamic feature trust modulus to generate a task adaptation fusion feature; and step four, generating a collaborative service result based on the task adaptation fusion characteristic, generating a task utility feedback signal based on a real-time effect index of the collaborative service result and an expected target of the collaborative task request, and updating the code of the task state vector by using the task utility feedback signal.
2. The heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation according to claim 1 is characterized in that in the first step, semantic analysis is performed on the collaborative task request, task semantic elements are extracted, and specifically the method comprises the steps of identifying and extracting task main bodies, associated domains, task targets and task constraints in the collaborative task request, wherein task state vectors at least comprise implicit expressions of data freshness requirements, feature reliability requirements and domain weight bias.
3. The heterogeneous information cross-domain collaborative processing method based on multi-dimensional feature adaptation according to claim 1, wherein in the step two, a task state vector based dynamic modulation increment time sequence alignment strategy is specifically included, a real-time bias coefficient and an accuracy bias coefficient are decoded from the task state vector, a target alignment strategy is selected from a plurality of preset alignment strategies according to the proportion relation between the real-time bias coefficient and the accuracy bias coefficient, and the plurality of preset alignment strategies at least include a positive interpolation strategy which ensures low delay preferentially and a buffer fine alignment strategy which ensures alignment accuracy preferentially.
4. The heterogeneous information cross-domain collaborative processing method based on multi-dimensional feature adaptation according to claim 1, wherein in the step three, the dynamic feature trust modulus of the preliminary alignment feature is evaluated, in particular calculated by the following formula: Wherein, the The dynamic feature trust modulus representing the ith feature at time t, Representing the sigmoid activation function, A task state vector representing a time t, A preliminary alignment feature vector representing the ith feature at time t, Is the task dependency adjustment coefficient, A quantized value of data uncertainty representing the ith feature at time t, A model uncertainty quantization value representing the ith feature of t, And The weight coefficients of the data uncertainty and the model uncertainty are respectively; The dynamic feature trust modulus is positively regulated by the association degree of the task state vector and the feature, and is negatively regulated by the data of the feature and the uncertainty of the model.
5. The heterogeneous information cross-domain collaborative processing method based on multi-dimensional feature adaptation according to claim 1 is characterized in that in the third step, the preliminary alignment features are subjected to weighted fusion according to the dynamic feature trust modulus, specifically, the dynamic feature trust modulus of each feature is subjected to normalization processing to obtain fusion weights, and the corresponding preliminary alignment features are subjected to weighted summation by using the fusion weights to generate the task adaptation fusion features.
6. The heterogeneous information cross-domain collaborative processing method based on multi-dimensional feature adaptation according to claim 1 is characterized in that in the fourth step, the task state vector is updated by using the task utility feedback signal, specifically by inputting the task utility feedback signal and the task state vector together into a task encoder neural network, wherein the task encoder neural network takes a negative value of the minimum task utility feedback signal as an optimization target, and adjusts network parameters thereof through a back propagation algorithm, so that the task encoder neural network outputs updated task state vectors for the same or similar collaborative task requests; The task encoder neural network adopts a multi-layer perceptron structure with an attention mechanism.
7. The heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation according to claim 1, wherein in the second step, modulation of the incremental time sequence alignment strategy, and evaluation and weighted fusion of the dynamic feature trust modulus in the third step form a data stream adaptive inner loop; Generating the task state vector in the first step, and updating the task state vector based on the task utility feedback signal in the fourth step to form a task strategy optimization outer loop; The data flow self-adaptive inner ring is dynamically regulated and controlled by the task strategy optimization outer ring, the task strategy optimization outer ring is optimized according to the operation effect of the data flow self-adaptive inner ring, and the two components form a double-circulation driving framework.
8. The multi-dimensional feature adaptation-based heterogeneous information cross-domain collaborative processing method according to claim 1, wherein in the second step, the timing deviation correction and alignment specifically comprises adding a high-precision timestamp to each streaming heterogeneous data block, and estimating missing or delayed data points by using a lightweight timing prediction model based on the high-precision timestamp and a current system timing, wherein the lightweight timing prediction model is a linear kalman filter.
9. The multi-dimensional feature adaptation-based heterogeneous information cross-domain collaborative processing method according to claim 1, wherein the different domains include data sources with different physical sources, different data formats, or different logical concepts; The streaming heterogeneous data block comprises at least two of text data, numerical sequence data and image data.
10. The multi-dimensional feature adaptation-based heterogeneous information cross-domain collaborative processing method according to claim 1, wherein the real-time effect index of the collaborative service result comprises at least one of prediction accuracy, response delay and resource utilization; The task utility feedback signal is a quantized function value of the difference between the real-time effect indicator and the intended target.

Description

Heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation Technical Field The invention relates to the technical field of heterogeneous information processing, in particular to a heterogeneous information cross-domain collaborative processing method based on multidimensional feature adaptation. Background Along with the deep fusion of big data and artificial intelligence technology, data generated by various information systems show typical characteristics of mass scale, heterogeneous sources, morphological flow and sparse value. Under the background, the knowledge migration and the value gain are realized by mining and utilizing heterogeneous information dispersed in different fields and among different systems through cross-domain collaborative processing, and the method becomes a key path for improving the service efficiency of the big data resource. For example, in smart city management, real-time traffic flow data, historical meteorological data and social media event information are effectively coordinated, so that the smart city management system has important significance for realizing accurate abnormal early warning and decision support. However, inherent differences in heterogeneous information at the data schema, timing characteristics, and semantic levels make cross-domain co-processing challenging. The prior art proposes several solutions for different sides of heterogeneous information cross-domain collaboration. The idea is focused on improving the accuracy and robustness of cross-domain features or knowledge migration. For example, the invention patent publication number CN115757529B discloses a cross-domain commonality migration recommendation method based on multi-element auxiliary information fusion, which extracts the commonality characteristics among the user domains through a variation self-encoder, and generates embedded characteristics by using a self-attention mechanism to recommend. The method has the core contribution of enhancing the common information migration by minimizing the mutual information between the common characteristics and the individual characteristics, thereby relieving the negative migration problem. However, such methods are typically offline trained and modeled based on historical static data sets, with feature adaptation and migration strategies solidifying upon model deployment, lacking the ability to perceive and respond to dynamically evolving real-time collaborative task intents. The other thought can solve the alignment problem of the multi-source heterogeneous data at the structure level. For example, the invention patent publication number CN116050374a discloses a data alignment method of cross-domain and cross-source, which innovatively fuses text representation and visual position information of table data, and determines an alignment result by calculating semantic distance of multi-modal vector expression. The method improves alignment accuracy, but the processing paradigm is essentially static and batch-oriented form data, the alignment process is independent of an upper-layer specific cooperative task target, and a streaming scene where data continuously arrives at a high speed is not considered, so that low-delay real-time cooperative requirements are difficult to support. In summary, one technical bottleneck faced in the field of heterogeneous information cross-domain collaborative processing in the current technology is that the existing scheme fails to carry out organic through and closed-loop linkage on semantic understanding of high-level and changeable collaborative tasks and bottom-layer and dynamic streaming data processing. Specifically, the static migration and alignment method cannot adapt to the real-time change of the task targets in the open environment, but the lack of the flow processing guided by the task semantics makes it difficult to ensure that the collaborative process is always focused on the core targets, and resource waste and effect deviation are easily caused. Therefore, a new technical solution is needed, which can enable the system to dynamically understand and adapt to the continuously evolving collaborative task targets while continuously processing the streaming heterogeneous data. In view of this, the present invention aims to solve the above-mentioned technical problem of task awareness and data stream processing split. Disclosure of Invention The invention aims to make up the defects of the prior art, provides a heterogeneous information cross-domain collaborative processing method based on multi-dimensional feature adaptation, generates a task state vector by analyzing collaborative task semantics, constructs a double-circulation driving framework, dynamically modulates a time sequence alignment strategy, evaluates feature dynamic trust modulus and carries out weighted fusion, can realize closed-loop linkage of task semantic understanding and streaming he