Telemetering multi-source data framing method and system based on dynamic optimization
Patent Information
- Application Number
- CN202511059623.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-30
- Publication Date
- 2025-11-07
Smart Images
Figure CN120909786A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of data transmission, more particularly, it relates to a telemetry multi-source data framing method and system based on dynamic optimization. BACKGROUND
[0002] With the rapid development of smart city construction, a large number of distributed sensor networks are deployed for environmental monitoring, traffic management, public safety and other fields, which produce massive multi-source heterogeneous data every day. Under the conditions of limited bandwidth and energy, how to efficiently transmit these data becomes a key challenge.
[0003] The existing telemetry data framing technology mainly adopts a unified processing strategy, without distinguishing the semantic importance of data, resulting in that when a key event occurs, high-value data and low-value data are treated equally, causing the loss or delay of critical information, and at the same time, causing bandwidth waste under normal circumstances. In addition, although the existing semantic segmentation model can evaluate the value of data, it is difficult to adapt to the data distribution difference of the target domain after training in the source domain, and a large amount of manually annotated data is needed for retraining, which is costly and inefficient.
[0004] The domain adaptation method aiming at data distribution changes mainly designs for one-time domain migration, which cannot effectively handle the problem of continuous drift of data distribution in the long-term operation of the environmental monitoring system. With the passage of time, factors such as environmental changes and device aging cause the data distribution to gradually change, and the existing method often suffers from catastrophic forgetting while adapting to the new data distribution, losing the ability to recognize historical data patterns.
[0005] In addition, the existing telemetry data framing system lacks a dynamic resource scheduling mechanism, and cannot adjust the allocation of computing resources and transmission bandwidth in real time according to the semantic value of data and the system state, resulting in that the optimal performance cannot be maintained when the system load fluctuates or the network condition changes.
[0006] Therefore, there is an urgent need for a dynamic optimization telemetry multi-source data framing method that can be based on data semantic understanding, has environmental adaptability and continuous learning ability. SUMMARY
[0007] The present application provides a telemetry multi-source data framing method and system based on dynamic optimization, which solves the technical problems of low data transmission efficiency, poor environmental adaptability, short system effective period and low resource utilization in related technologies.
[0008] The present application provides a telemetry multi-source data framing method based on dynamic optimization, comprising: constructing a lightweight semantic segmentation network to generate a data semantic value distribution map and identify high-value segments in multi-source data streams; Based on the semantic segmentation results, dynamically generate adaptive quantization table, realize differential encoding; Apply adversarial training and feature alignment techniques to the encoding results to realize cross-domain knowledge transfer; On the basis of knowledge transfer, apply elastic weight integration and experience replay mechanism to realize continuous learning optimization; According to the data semantic value distribution diagram and the current system state information, dynamically schedule resources to optimize system performance.
[0009] Further, the step of constructing a lightweight semantic segmentation network to identify high-value sections in multi-source data streams includes: Construct a convolutional attention hybrid lightweight neural network model; Use multi-modal data fusion technology to improve recognition accuracy; Generate a data semantic value distribution diagram to divide data into multiple levels from high to low value.
[0010] Further, in the data semantic value distribution diagram, data value is divided into four levels: key data, high-value data, medium-value data, and low-value data.
[0011] Further, the step of dynamically generating adaptive quantization table based on semantic segmentation results to realize differential encoding includes: Construct an adaptive quantization table generation algorithm to dynamically adjust quantization parameters according to the semantic value level of data; Implement a hierarchical encoding strategy that uses different compression ratios and quality parameters for data of different value levels.
[0012] Further, the step of applying adversarial training and feature alignment techniques to the encoding results includes: Construct a dual-path structure of domain-invariant feature extractor and domain-specific encoder; Apply adversarial training method to minimize the difference in feature distribution between domains; Introduce semantic consistency constraints to maintain cross-domain semantic correspondence.
[0013] Further, the step of applying elastic weight integration and experience replay mechanism includes: Apply elastic weight integration mechanism to prevent catastrophic forgetting; Implement pseudo-label generation and screening mechanism to effectively use unlabeled data; Apply gradient projection strategy to balance new and old task learning; Maintain a historical data cache pool to realize experience replay.
[0014] Further, the step of dynamically scheduling resources according to the data semantic value distribution diagram and the current system state information to optimize system performance includes: Construct a multi-objective resource allocation optimization model; Implement a priority scheduling strategy based on data value; An online reinforcement learning algorithm is used to continuously optimize the resource scheduling strategy.
[0015] Further, the multi-objective resource allocation optimization model comprehensively considers data value, processing delay, transmission delay and energy consumption to dynamically allocate computing and network resources for different tasks.
[0016] Further, in the priority scheduling strategy based on data value, the task priority is proportional to the data value.
[0017] The present application provides a telemetry multi-source data framing system based on dynamic optimization, which is used to execute the above-mentioned telemetry multi-source data framing method based on dynamic optimization, comprising: A semantic segmentation module is used to construct a lightweight semantic segmentation network to identify high-value sections in multi-source data streams. An adaptive encoding module is used to dynamically generate an adaptive quantization table based on the semantic segmentation result to realize differential encoding. A cross-domain migration module is used to realize cross-domain knowledge migration through adversarial training and feature alignment. A continuous learning module is used to realize continuous learning optimization through elastic weight integration and experience replay mechanism. A resource scheduling module is used to dynamically schedule resources based on data value and system state to optimize system performance.
[0018] The beneficial effects of the present application are: Through semantic value recognition and differential encoding, the key event recognition rate is improved, the transmission delay is reduced, and the overall storage demand is reduced. Through cross-domain knowledge migration technology, the system can quickly adapt to different monitoring environments, reduce cross-scene deployment cost, and improve model migration efficiency. Through the continuous learning mechanism, the system can maintain high performance for a long time, effectively cope with data distribution drift, and reduce model performance decay rate. Through semantic-driven differential processing and dynamic resource scheduling, the utilization rate of computing resources is improved, and the bandwidth utilization efficiency is improved. BRIEF DESCRIPTION OF DRAWINGS
[0019] Figure 1 is a flowchart of a telemetry multi-source data framing method based on dynamic optimization in the present application; Figure 2 is a column chart comparing the model migration efficiency of the traditional method and the method of the present application in three different cross-domain scenarios; Figure 3This is a line graph comparing the trend of model accuracy changes between the traditional method and the method of this invention during long-term operation. Detailed Implementation
[0020] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.
[0021] At least one embodiment of the present invention discloses a method for framing telemetry multi-source data based on dynamic optimization, such as... Figure 1 As shown, it includes: Step 1: Construct a lightweight semantic segmentation network to generate a data semantic value distribution map and identify high-value segments in multi-source data streams; This step aims to perform semantic understanding and value assessment on the input telemetry multi-source data, identifying data segments with high information value to provide a basis for subsequent differentiated processing. Specifically, it includes the following sub-steps: Step 1.1: Construct a lightweight neural network model that combines convolutional attention; To address the limited computing resources of edge devices, this application constructs a lightweight neural network model that reduces the number of parameters and computational cost while maintaining high recognition accuracy. The model employs a combination of depthwise separable convolution and channel attention mechanisms, keeping the total number of parameters below 500K to accommodate the computing capabilities of edge devices.
[0022] Specifically, this lightweight network architecture includes: an input layer, four depthwise separable convolutional layers, two channel attention modules, a global average pooling layer, and a fully connected output layer. The depthwise separable convolutions reduce computation by decomposing standard convolution into two independent steps: depthwise convolution and pointwise convolution. The channel attention modules enhance the model's ability to extract key features by adaptively learning the importance of different feature channels.
[0023] The mathematical expression of the model is as follows: Depthwise separable convolution computation: ; in This represents the output feature map of a depthwise separable convolution; Indicates a depthwise convolution operation; Indicates the input feature map; Represents the convolution kernel of a depthwise convolution; denotes a convolution operator; denotes a point convolution operation; denotes a convolution kernel for point convolution.
[0024] Channel attention calculation: ; wherein denotes a channel attention weight; denotes a Sigmoid activation function; denotes a multi-layer perception; denotes a global average pooling operation; denotes an input feature map; denotes a global max pooling operation; denotes an output feature map after channel attention weighting; denotes an element-wise multiplication operation.
[0025] Step 1.2, improve recognition accuracy by using multi-modal data fusion technology; In order to improve the recognition accuracy of high value data segments, the application adopts multi-modal data fusion technology, and comprehensively utilizes the data features of different types of sensors. For different types of sensor data such as temperature, humidity, air pressure and gas concentration commonly used in environmental monitoring systems, the feature level fusion and decision level fusion are combined for processing.
[0026] The feature level fusion adopts an adaptive weighting mechanism, which dynamically adjusts the weight of different modal data in the fusion process according to the signal-to-noise ratio of the data: ; wherein, denotes the fused feature; denotes a summation symbol; denotes the fusion weight of the th modal data; denotes the feature of the th modal data. The weight is calculated by the following formula: ; wherein, denotes the normalized weight of the th modal; denotes an exponential function; and denote the quality scores of the th and th modal data, respectively; denotes a summation symbol.
[0027] Step 1.3, generate a data semantic value distribution map; Based on the output result of the lightweight neural network, a semantic value distribution map is generated for the input multi-source data stream, and the data is divided into multiple levels from high to low value. The semantic value score adopts a standardized score of 0-1, which is calculated by the following formula: ; wherein, denotes the semantic value score of the data point ; denotes the i-th data point; denotes the abnormality degree weight coefficient; denotes the abnormality degree score of the data point ; denotes the novelty weight coefficient; denotes the novelty score of the data point ; denotes the correlation with the key event weight coefficient; denotes the correlation score of the data point with the key event.
[0028] The data value is divided into 4 levels: key data (0.8 to 1.0), high value data (0.6 to 0.8), medium value data (0.3 to 0.6), and low value data (0 to 0.3). This classification will be used in the subsequent differential processing process.
[0029] Step 2, dynamically generate an adaptive quantization table based on the semantic segmentation result to realize differential encoding; According to the semantic value distribution map generated in step 1, different value level data is adopted for differential encoding strategy, and an adaptive quantization table is dynamically generated and updated to realize reasonable allocation of resources. Specifically, the following sub-steps are included: Step 2.1, constructing an adaptive quantization table generation algorithm; The application constructs an adaptive quantization table generation algorithm, which dynamically adjusts the quantization parameters according to the semantic value level of the data. This algorithm uses a non-uniform quantization strategy, which uses finer quantization steps for high-value data regions and coarser quantization steps for low-value data regions, to reduce the overall data volume while ensuring the accuracy of key information.
[0030] The calculation formula of the quantization step is as follows: ; wherein, denotes the quantization step of the data with semantic value ; denotes the semantic value score of the data; denotes the minimum quantization step; denotes the maximum quantization step; This represents the adjustment factor.
[0031] Step 2.2: Implement the layered coding strategy; Based on an adaptive quantization table, this application implements a hierarchical coding strategy, employing different compression ratios and quality parameters for data of different value levels: Key data (semantic value score 0.8 to 1.0): Near lossless encoding is used, with a compression ratio controlled between 1.5:1 and 2:1 to ensure data integrity and accuracy.
[0032] High-value data (semantic value score of 0.6 to 0.8): medium-to-high quality encoding is used with a compression ratio of 4:1 to 6:1 to ensure that important features are not lost.
[0033] Medium-value data (semantic value score of 0.3 to 0.6): Medium-quality coding is used with a compression ratio of 8:1 to 12:1 to preserve the main trends and features.
[0034] Low-value data (semantic value score of 0 to 0.3): High compression ratio encoding is used, with a compression ratio of 15:1 to 20:1, and only basic features are retained.
[0035] The formula for calculating the encoding parameters is as follows: ; ; in, The semantic value is The data compression ratio; This represents the semantic value score of the data; Indicates the minimum compression ratio; Indicates the maximum compression ratio; Indicates the reciprocal factor of value; The semantic value is Data quality parameters; Indicates the minimum mass parameter; This represents the maximum mass parameter.
[0036] Step 3: Apply adversarial training and feature alignment techniques to the encoding results to achieve cross-domain knowledge transfer; To address the variations in data characteristics across different cities and climates, this application constructs a domain-adaptive semantic segmentation framework. Through adversarial training and feature alignment techniques, it achieves cross-domain knowledge transfer, enabling the framing system to quickly adapt to different monitoring environments without requiring extensive retraining with labeled data. Specifically, it includes the following sub-steps: Step 3.1: Construct a dual-path structure consisting of a domain-invariant feature extractor and a domain-specific encoder; The present application constructs a dual-path network structure, including a domain-invariant feature extractor and two domain-specific encoders (for the source domain and the target domain respectively). The domain-invariant feature extractor is responsible for extracting common features shared between the source domain and the target domain, while the domain-specific encoders are responsible for processing the data distribution characteristics specific to each domain.
[0037] The domain-invariant feature extractor adopts a convolutional network structure with shared weights, and through adversarial training, the features extracted by it have similar distributions between domains; the domain-specific encoders adopt an autoencoder structure trained independently, and perform feature reconstruction and decoding according to the data characteristics of each domain.
[0038] Step 3.2, apply the adversarial training method to minimize the difference in feature distribution between domains; Through the adversarial training mechanism, the domain-invariant feature extractor generates domain-independent feature representations. Specifically, a domain discriminator network is introduced, which attempts to distinguish whether the features are from the source domain or the target domain, while the feature extractor attempts to generate features that the discriminator cannot distinguish the source of.
[0039] The adversarial loss function is defined as follows: ; Where, represents the adversarial loss function; represents the expected operation on the source domain data ; represents the data sample of the source domain; represents the natural logarithm function; represents the domain discriminator network; represents the feature extractor applied to the source domain; represents the expected operation on the target domain data ; represents the data sample of the target domain; represents the feature extractor applied to the target domain.
[0040] Step 3.3, introduce semantic consistency constraints to maintain cross-domain semantic correspondence; To ensure that the semantic information remains consistent during feature alignment, the present application introduces a semantic consistency loss, which forces the features of the source domain and the target domain to maintain a corresponding relationship in semantics.
[0041] The semantic consistency loss is defined as follows: ; Where, represents the semantic consistency loss function; represents the square operation of the L2 norm; represents the specific encoder of the source domain; represents the feature extractor applied to the source domain. a data sample representing the source domain; a specific encoder representing the target domain; a feature extractor applied to the target domain; a data sample representing the target domain.
[0042] In addition, the present application introduces an adaptive instance normalization (AdaIN, Adaptive Instance Normalization) technique by adjusting the statistical properties of the target domain features to match the source domain features: ; wherein, denotes an adaptive instance normalization operation; denotes the target domain feature; denotes the source domain feature; denotes an operation for calculating the standard deviation of the source domain feature ; denotes an operation for calculating the mean of the target domain feature ; denotes an operation for calculating the standard deviation of the target domain feature ; denotes an operation for calculating the mean of the source domain feature .
[0043] As shown in Figure 2 , the model migration efficiency of the present application method and the traditional method in three different cross-domain scenarios is shown, and the results show that in all scenarios, the migration efficiency of the present application method is higher than that of the traditional method, and the average is increased by about 60 percentage points, proving that the adversarial training and feature alignment technology of the present application can effectively promote cross-domain knowledge migration and reduce the cost of deploying the system in different environments.
[0044] Step 4, on the basis of knowledge migration, apply the elastic weight integration and experience replay mechanism to realize continuous learning optimization; In view of the problem of continuous drift of data distribution in the long-term operation of the environmental monitoring system, the present application constructs an incremental continuous learning framework, so that the system can continuously adapt to new data distribution while retaining historical knowledge to avoid catastrophic forgetting phenomenon. Specifically, the following sub-steps are included: Step 4.1, apply the elastic weight integration mechanism to prevent catastrophic forgetting; The present application adopts the elastic weight integration (EWC, Elastic Weight Consolidation) algorithm, which prevents the model from forgetting the old task knowledge when learning new tasks by imposing constraints on important parameters. EWC assesses the importance of parameters through the Fisher information matrix and imposes greater penalties on the changes of important parameters.
[0045] The elastic weighted integration loss function is defined as follows: ; in, This represents the elastic weighted integration loss function; This represents the cross-entropy loss of the current task; This represents the hyperparameter that balances learning for the current task with retaining information from previous tasks; Indicates the summation symbol; Represents the first Fisher information matrix One diagonal element; Indicates the current parameter value; This represents the parameter values after completing the training for the old task.
[0046] Step 4.2: Implement a pseudo-label generation and filtering mechanism to effectively utilize unlabeled data; This application constructs a pseudo-label generation and screening mechanism, which generates high-quality pseudo-labels for unlabeled data in new environments through model ensemble and uncertainty estimation, thereby expanding the training dataset.
[0047] Specifically, multiple model variants (obtained through different initializations, data augmentations, or architectural changes) are used to predict the same data, and high-quality pseudo-labels are selected based on prediction consistency and confidence scores. For each unlabeled sample, predictions are made using multiple model variants; Calculate the consistency score and average confidence level of the predictions; If both the consistency score and the average confidence score exceed the preset threshold, the prediction result is adopted as a pseudo-label. The selected high-confidence pseudo-label data are added to the training set for subsequent model updates.
[0048] Step 4.3: Apply gradient projection strategy to balance the learning of new and old tasks; This application employs a gradient projection strategy to project and correct gradient components that may cause performance degradation in older tasks, thereby balancing the learning process of new and old tasks.
[0049] Gradient projection is calculated as follows: ; in, This represents the gradient after projection; Represents the original gradient; This represents gradient components that are detrimental to the performance of historical tasks. Represents gradient components The square of the L2 norm; This represents the L2 norm operator.
[0050] Step 4.4, maintain the historical data cache pool, implement experience replay; The application constructs an experience replay mechanism, maintains a cache pool containing historical data samples, and mixes historical samples when training new data to strengthen the model's memory of historical knowledge. The samples in the cache pool are selected using a sampling strategy based on diversity and representativeness, ensuring that the most informative historical samples are stored in a limited space.
[0051] The cache pool update strategy includes: Diversity maintenance: ensure that the samples in the cache pool cover different categories and distribution areas; Difficult example priority: preferentially retain samples that the model is difficult to correctly classify; Periodic review: regularly use samples in the cache pool for model training to strengthen historical knowledge.
[0052] As Figure 3 shown, the model accuracy trend of the traditional method and the method of the application in the long-term running process is shown. As can be seen from the figure, the model accuracy of the traditional method gradually decreases over time, and decreases by about 30% after 6 months; while the method of the application can maintain relatively stable high performance through continuous learning and elastic weight integration mechanism, only decreases by about 1% after 6 months, and the model performance decay rate is reduced by 70%.
[0053] Step 5, dynamically schedule resources according to data semantic value distribution diagram and current system state information, optimize system performance; According to the data semantic value and the system running state (including computing load, network condition, etc.), the application realizes dynamic adjustment of computing resource and transmission bandwidth allocation, ensures that the system maintains the optimal performance under various conditions. Specifically, the following sub-steps are included: Step 5.1, construct a multi-objective resource allocation optimization model; The application constructs a multi-objective resource allocation optimization model, which considers multiple indicators such as data value, processing delay, transmission delay, energy consumption, etc., and dynamically allocates computer network resources for different tasks.
[0054] The resource allocation optimization problem is formalized as: ; Where, represents the minimization operation; represents the summation symbol; represents the total number of tasks; represents the weight coefficient of processing delay; represents the processing delay function when task uses resource ; represents the resource vector allocated to task ; weight coefficient representing transmission delay; representing task using resource transmission delay function at time; weight coefficient representing energy consumption; representing task using resource computing energy consumption function at time; representing task using resource communication energy consumption function at time; representing constraint condition; representing the number of resource types allocated to task ; representing the total available amount of resource type ; representing all; representing belonging to operator ; representing the total number of resource types.
[0055] Step 5.2, implement priority scheduling strategy based on data value; Based on the semantic value of data, the application implements a priority scheduling strategy to ensure that high-value data gets more resources and higher processing priority. Specifically, the priority of a task is proportional to the data value : ; wherein, representing the priority of task ; representing the base priority; representing the amplification coefficient; representing the average semantic value of data processed by task .
[0056] Step 5.3, adopt online reinforcement learning algorithm to continuously optimize resource scheduling strategy; The application adopts online reinforcement learning algorithm (such as DeepQ-Network or PPO) to continuously optimize resource scheduling strategy according to system feedback, adapting to dynamically changing environment and task requirements.
[0057] The state space of the reinforcement learning model includes: current system load, network condition, queue length of various tasks, data value distribution, etc.; the action space includes resource allocation ratio, task scheduling priority, etc.; the reward function considers processing efficiency, energy consumption and data quality: ; wherein, a reward function value is represented; a weight coefficient representing data transmission quality is represented; a data transmission quality score is represented; a weight coefficient representing resource consumption is represented; a resource consumption score is represented; a weight coefficient representing processing delay is represented; a processing delay score is represented.
[0058] Through continuous interaction and learning, the reinforcement learning model gradually optimizes the resource scheduling strategy and improves the overall performance of the system.
[0059] A telemetry multi-source data framing system based on dynamic optimization is used to perform the above-mentioned telemetry multi-source data framing method based on dynamic optimization, comprising: A semantic segmentation module is used to construct a lightweight semantic segmentation network to identify high-value sections in multi-source data streams; An adaptive encoding module is used to dynamically generate an adaptive quantization table based on the semantic segmentation results to achieve differentiated encoding; A cross-domain migration module is used to achieve cross-domain knowledge migration through adversarial training and feature alignment; A continuous learning module is used to achieve continuous learning optimization through elastic weight integration and experience replay mechanism; A resource scheduling module is used to dynamically schedule resources based on data value and system state to optimize system performance.
[0060] The embodiments of the present application are described above, but the embodiments are not limited to the specific implementation described above, and the specific implementation described above is only illustrative and not limiting. Those skilled in the art can make more forms of equivalent embodiments under the inspiration of the embodiments, which are all within the protection of the embodiments.
Claims
1. A method for dynamic optimization based telemetry multi-source data framing, comprising: The method comprises the following steps: constructing a lightweight semantic segmentation network to generate a data semantic value distribution map and identify high-value segments in multi-source data streams; dynamically generating an adaptive quantization table based on the semantic segmentation results to achieve differentiated coding; applying adversarial training and feature alignment techniques to the coding results to achieve cross-domain knowledge transfer; applying elastic weight integration and experience replay mechanisms on the basis of knowledge transfer to achieve continuous learning optimization; dynamically scheduling resources based on the data semantic value distribution map and current system state information to optimize system performance.
2. The method of claim 1, wherein, The step of constructing a lightweight semantic segmentation network to identify high-value segments in multi-source data streams comprises: constructing a convolutional attention hybrid lightweight neural network model; using multi-modal data fusion technology to improve recognition accuracy; generating a data semantic value distribution map to divide data into multiple levels from high to low value.
3. The method of claim 2, wherein, In the data semantic value distribution map, data value is divided into four levels: key data, high-value data, medium-value data, and low-value data.
4. The method of claim 1, wherein, The step of dynamically generating an adaptive quantization table based on the semantic segmentation results to achieve differentiated coding comprises: constructing an adaptive quantization table generation algorithm to dynamically adjust quantization parameters based on the semantic value level of the data; implementing a hierarchical coding strategy that uses different compression ratios and quality parameters for data of different value levels.
5. The method of claim 1, wherein, The step of applying adversarial training and feature alignment techniques to the coding results comprises: constructing a dual-path structure of domain-invariant feature extractor and domain-specific encoder; applying an adversarial training method to minimize the difference in feature distribution between domains; introducing semantic consistency constraints to maintain cross-domain semantic correspondence.
6. The method of claim 1, wherein, The step of applying elastic weight integration and experience replay mechanisms comprises: applying an elastic weight integration mechanism to prevent catastrophic forgetting; implementing a pseudo-label generation and screening mechanism to effectively utilize unlabeled data; applying a gradient projection strategy to balance new and old task learning; maintaining a historical data cache pool to implement experience replay.
7. The method of claim 1, wherein, The step of dynamically scheduling resources based on the data semantic value distribution map and current system state information to optimize system performance comprises: constructing a multi-objective resource allocation optimization model; implementing a priority scheduling strategy based on data value; adopting an online reinforcement learning algorithm to continuously optimize resource scheduling strategies.
8. The method of claim 7, wherein, The multi-objective resource allocation optimization model takes into account data value, processing delay, transmission delay, and energy consumption to dynamically allocate computer network resources for different tasks.
9. The method of claim 7, wherein, In the priority scheduling strategy based on data value, task priority is proportional to data value.
10. A dynamic optimization based telemetry multi-source data framing system, comprising: A telemetry multi-source data framing method based on dynamic optimization for executing any one of claims 1-9, comprising: a semantic segmentation module for constructing a lightweight semantic segmentation network to identify high-value segments in multi-source data streams; an adaptive coding module for dynamically generating an adaptive quantization table based on the semantic segmentation results to achieve differentiated coding; a cross-domain migration module for achieving cross-domain knowledge transfer through adversarial training and feature alignment; a continuous learning module for achieving continuous learning optimization through elastic weight integration and experience replay mechanisms; a resource scheduling module for dynamically scheduling resources based on data value and system state to optimize system performance.