Data acquisition instruction generation method based on uncertainty tracing and value evaluation

Abstract

The invention discloses a data acquisition instruction generation method based on uncertainty tracing and value evaluation, and relates to the technical field of artificial intelligence, and the method comprises the steps: constructing a dynamic time dependence graph, and quantifying temporal association and time migration between sensor data; tracing the propagation path of the time offset in the dependency network, and quantifying the cumulative influence of the propagation path on the decision; an acquisition action for time synchronization is defined, and the information value of the acquisition action for reducing the network uncertainty is evaluated; the information value and the acquisition cost are comprehensively considered, and the optimal action capable of maximizing the network-level net income is selected; and the optimal action is converted into an executable acquisition instruction, and time compensation processing logic is embedded. According to the method, networked tracing and modeling are carried out on the time migration uncertainty by constructing the dynamic time dependence graph, the cumulative effect of the time migration uncertainty is quantified based on the time-delay dependence attenuation propagation algorithm, a key propagation path can be automatically identified, the network-level information value can be evaluated, and a collaborative acquisition instruction with a time compensation mechanism can be generated.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, and in particular relates to a method for generating data collection instructions based on uncertainty tracing and value assessment. Background Technology

[0002] In existing technologies, intelligent monitoring systems for multi-source asynchronous data fusion typically employ unified time synchronization, fixed-period sampling, or global clock alignment to manage time synchronization issues. These methods primarily focus on reducing the timestamp errors of individual data sources, such as by broadcasting synchronization signals to all sensors or requesting high-precision time synchronization services. However, they fail to model and trace the propagation and cumulative effects of time offset uncertainty among networked data sources. In a monitoring network composed of multiple sensors with temporal dependencies, a time deviation in one data source not only affects the validity of its own data but may also propagate and amplify uncertainty to downstream related data sources through physical or logical dependency chains. Existing common technologies, lacking characterization of such dependency network structures and propagation mechanisms, can only perform isolated, point-to-point time corrections, failing to assess and address the networked impact of time offsets throughout the system. This results in "temporal mismatch" at the fusion point of multi-source data even with improved time accuracy of individual sensors due to untraceded and uncompensated transitive time delay errors, thus limiting the performance improvement potential of subsequent causal analysis or predictive decision models. Therefore, the following solutions are proposed to address these issues. Summary of the Invention

[0003] The purpose of this invention is to provide a data acquisition instruction generation method based on uncertainty tracing and value assessment. By constructing a dynamic time dependency graph, the method performs networked tracing and modeling of time offset uncertainty, and quantifies its cumulative effect based on a time-delay dependency decay propagation algorithm. This method can automatically identify key propagation paths, assess the value of network-level information, and generate collaborative acquisition instructions with a time compensation mechanism. This solves the problem that existing methods cannot trace and offset the transitive effect of time offset in networked dependencies, which leads to persistent temporal mismatch in multi-source asynchronous data at the fusion point.

[0004] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0005] This invention provides a data collection instruction generation method based on uncertainty tracing and value assessment, comprising the following steps:

[0006] Receive an uncertainty decomposition structure, the uncertainty decomposition structure containing at least one node representing a source of uncertainty, each node being associated with a data source and containing a quantitative assessment of the uncertainty of that data source;

[0007] Based on the aforementioned uncertainty decomposition structure, for each node, the information value that can be obtained by collecting additional data from the corresponding data source is evaluated, where the information value is the expected reduction in decision loss before and after the collection action is performed.

[0008] Based on the source type of each node, define at least one candidate data acquisition action and construct an action space;

[0009] Based on the calculated information value of each candidate data acquisition action and the estimated cost of performing the action, an optimization decision is made to select the optimal data acquisition action.

[0010] The optimal data acquisition action is converted into standardized data acquisition instructions that can be executed by the machine;

[0011] The data acquisition command is output to drive the external data acquisition system to perform the corresponding data acquisition operation.

[0012] Furthermore, in the step of receiving the uncertainty decomposition structure, the uncertainty decomposition structure is a tree structure, and each node further includes the influence weight of the uncertainty source on the final decision task and the source type identifier; the source type includes at least one of time offset uncertainty, parameter uncertainty, structural uncertainty and data missing.

[0013] Furthermore, the steps for assessing the value of the information specifically include:

[0014] For each candidate data collection action, multiple samples are taken from the posterior distribution of the decision model parameters;

[0015] For each sampling, calculate the decision loss when using the current dataset, and the decision loss of the dataset after simulating the sampling action and incorporating the new data.

[0016] The information value estimate of the sampling action is obtained by averaging the reduction in decision loss across all sampling attempts.

[0017] Furthermore, in the step of constructing the action space, candidate data acquisition actions are mapped to different operation categories according to the source type of the nodes; specifically, for time offset uncertainty, it is mapped to an operation requesting time synchronization; for parameter uncertainty, it is mapped to an operation requesting supplementary observation samples; for structural uncertainty, it is mapped to an operation triggering verification experiments; and for data missing, it is mapped to an operation requesting retrieval from external data sources.

[0018] Furthermore, the step of making optimization decisions is achieved by solving an optimization problem. The objective function of the optimization problem is the difference between the estimated information value and the acquisition cost after weight adjustment. The candidate action that maximizes this difference is selected as the optimal data acquisition action.

[0019] Furthermore, the step of converting the data into machine-executable instructions is implemented through a predefined instruction template. The instruction template is configured according to the action type, and the parameters in the optimal data acquisition action are filled into the corresponding positions of the template to generate an instruction string containing the target data source identifier, acquisition type, and necessary security parameters.

[0020] Furthermore, after outputting the data acquisition command, a feedback update step is also included:

[0021] Receive acquisition result data returned by an external data acquisition system according to the instructions;

[0022] The posterior probability distribution of the decision model is updated using the collected data.

[0023] Based on the updated posterior distribution, the uncertainty quantification assessment of each node in the uncertainty decomposition structure is re-evaluated to form a closed-loop learning process.

[0024] The present invention has the following beneficial effects:

[0025] 1. This invention, by constructing and maintaining a dynamic time dependency graph, can systematically trace the source of time offset uncertainty and its propagation path in sensor networks; based on a time delay dependency decay model, it quantifies the cumulative effect of offset on decision-making, thereby achieving network-level time alignment and compensation for multi-source asynchronous data; the generated data acquisition instructions not only include synchronization requests but also embed compensation processing instructions for dependencies, which helps to alleviate temporal mismatch problems from the source; this improves the overall consistency and timeliness of data fusion, provides a more reliable basic input for subsequent decision models, and enhances the effectiveness of data utilization in complex monitoring scenarios.

[0026] 2. This invention introduces network-level benefit analysis into value-oriented information assessment, taking into account the indirect reduction of uncertainty of dependent nodes caused by data collection actions, thereby more comprehensively measuring information gain. Combined with cost modeling and optimization decision-making, the system can select the data collection actions with the highest net value under resource constraints, such as prioritizing collaborative sampling for key dependency chains. This intelligent resource allocation strategy avoids the high overhead of uniform or frequent synchronization of all nodes, optimizing the utilization of communication bandwidth, equipment energy consumption, and computing resources. It supports balancing data collection benefits and operating costs in long-term operation, promoting the sustainability of the monitoring system.

[0027] 3. The dynamic time dependency graph of this invention has adaptive update capability, which can adjust the temporal dependency strength and offset variance estimation by continuously analyzing the data stream, reflecting changes in network state and environment. This allows the acquisition strategy to flexibly respond to dynamic factors such as sensor performance drift and network latency fluctuations. The instruction generation mechanism redirects the synchronization focus and compensation parameters based on the updated dependency graph, ensuring the applicability of time compensation measures under different operating conditions. This adaptability enhances the robustness of the method in dealing with complex real-world scenarios and maintains the continuity and effectiveness of data acquisition and processing.

[0028] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0029] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0030] Figure 1 This is a flowchart illustrating the data collection instruction generation method based on uncertainty tracing and value assessment of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0032] Please see Figure 1 As shown, this invention is a data collection instruction generation method based on uncertainty tracing and value assessment, comprising the following steps:

[0033] Receive an uncertainty decomposition structure, which contains at least one node representing a source of uncertainty, each node being associated with a data source and containing a quantitative assessment of the uncertainty of that data source;

[0034] Based on the uncertainty decomposition structure, for each node, the information value that can be obtained by collecting additional data from the corresponding data source is evaluated. The information value is the expected reduction in decision loss before and after the collection action is performed.

[0035] Based on the source type of each node, define at least one candidate data collection action and construct the action space;

[0036] Based on the calculated information value of each candidate data acquisition action and the estimated cost of performing the action, an optimization decision is made to select the optimal data acquisition action.

[0037] Convert the optimal data acquisition actions into standardized data acquisition instructions that can be executed by the machine;

[0038] Output data acquisition commands to drive the external data acquisition system to perform the corresponding data acquisition operations.

[0039] In the step of receiving the uncertainty decomposition structure, the uncertainty decomposition structure is a tree structure, and each node further includes the influence weight of the uncertainty source on the final decision task and the source type identifier; the source type includes at least one of time offset uncertainty, parameter uncertainty, structural uncertainty and data missing.

[0040] The specific steps for assessing the value of information include:

[0041] For each candidate data collection action, multiple samples are taken from the posterior distribution of the decision model parameters;

[0042] For each sampling, calculate the decision loss when using the current dataset, and the decision loss of the dataset after simulating the sampling action and incorporating the new data.

[0043] The information value estimate of the sampling action is obtained by averaging the reduction in decision loss across all sampling attempts.

[0044] In the process of constructing the action space, candidate data acquisition actions are mapped to different operation categories based on the source type of the nodes. Specifically, for time offset uncertainty, the action is mapped to request time synchronization; for parameter uncertainty, the action is mapped to request supplementary observation samples; for structural uncertainty, the action is mapped to trigger a verification experiment; and for missing data, the action is mapped to request retrieval from an external data source.

[0045] The optimization decision-making process is achieved by solving an optimization problem. The objective function of the optimization problem is the difference between the estimated information value and the weighted collection cost. The candidate action that maximizes this difference is selected as the optimal data collection action.

[0046] The process of converting data into machine-executable instructions is achieved through predefined instruction templates. These templates are configured according to the action type, and the parameters from the optimal data acquisition action are filled into the corresponding positions in the templates to generate instruction strings that include the target data source identifier, acquisition type, and necessary security parameters.

[0047] After outputting the data acquisition command, a feedback update step is also included:

[0048] Receive acquisition result data returned by an external data acquisition system according to instructions;

[0049] Update the posterior probability distribution of the decision model using the collected data;

[0050] Based on the updated posterior distribution, the uncertainty quantification assessment of each node in the uncertainty decomposition structure is re-evaluated, forming a closed-loop learning process.

[0051] One specific application of this embodiment is:

[0052] Step S1: Construction of Uncertain Input and Dynamic Time Dependency Graph (DTDG)

[0053] The system receives output from the upstream uncertainty tracing module (USTT), which contains a set of nodes identified as "time offset uncertainty". Each node For the time uncertainty of a data source (such as a specific sensor), the attributes include uncertainty variance. Impact on decision-making losses ;

[0054] This method does not directly use these discrete nodes, but instead constructs a dynamic time-dependent graph. ,in, Let be the set of vertices of the graph, each vertex Corresponding to a data source (such as a sensor); Let be the set of directed edges of a graph, and each edge Indicates from vertex To the top Temporal dependencies; Let the set of edge weights be the weights of each edge. Corresponding to a weight vector Weight It is a dynamically updated vector. ,in, The time-delay cross-correlation coefficient is based on historical data and reflects the strength of dependency. For the estimation from arrive The reference time delay (such as water flow propagation time); The variance of the time offset is calculated by analyzing the timestamp difference sequence of historical data;

[0055] Construction and Initialization: The system is initialized based on physical topology (sensor geographic location) and domain knowledge (hydrological model). The edges; then, using historical data streams, the time-delay cross-correlation function within the sliding time window is dynamically updated. and For sensor pairs without prior physical association, edges are dynamically created or deleted by detecting Granger causality or transition entropy in their data sequences.

[0056] Step S2: Time-off propagation tracing and uncertainty quantification based on DTDG

[0057] For each time uncertainty node of the USTT input (corresponding vertex) This method traces its origins back to... The scope of influence in;

[0058] Forward propagation tracing: From Starting from the outgoing edge, recursively evaluate how time offsets affect the uncertainty of downstream data sources through dependency chains; for the path spread to Aggregation time offset uncertainty It depends not only on the source variance It is also affected by the weights of each edge on the path;

[0059] This effect is quantified using a time-delay-dependent decay propagation model:

[0060]

[0061] In the formula, To start from the source Along the path propagate to target vertex The uncertainty of aggregation time offset; As the source apex Its own time offset uncertainty intensity (variance); From arrive A directed path is formed by connecting several edges in sequence; For path A directed edge in the middle, from vertex Pointing to the vertex For the edge The time-delay cross-correlation coefficient; The calculation based on the latest observation data from arrive The actual time delay; For the edge The reference time delay; For the edge Time offset variance; It is a natural exponential function used to achieve the decay effect; The multiplication symbol represents a path. Perform consecutive multiplication operations on the attenuation factors on all edges of the equation;

[0062] This formula means that as the time offset uncertainty at the source propagates along the dependent edge, its influence is multiplied by the dependence strength of that edge. It is also subject to exponential decay modulation based on the deviation between the observed time delay and the reference time delay; the greater the deviation, the more significant the time alignment anomaly of the link, and the greater the uncertainty of propagation.

[0063] Uncertainty affecting all nodes: Vertex Comprehensive time uncertainty Due to its own source uncertainty The sum of uncertainties propagating from all incoming paths:

[0064]

[0065] In the formula, As vertices The overall time uncertainty includes its own inherent uncertainty and the sum of its uncertainty due to the influence of other sources of propagation; As vertices Its own time offset uncertainty intensity (variance); For all pointing vertices The set of directed paths; For set One of the paths; For path The starting point (i.e., the source of uncertainty propagation); To spread from the source of this path to The uncertainty of aggregation time offset;

[0066] this This will be used to replace the original USTT node. This is because it more comprehensively reflects the cumulative time uncertainty caused by time-dependent networks.

[0067] Step S3: Value Oriented Information (VOI) Calculation and Action Space Definition

[0068] VOI calculation: uses the basic formula defined in the document, but the loss function... The evaluation will pay particular attention to the impact of time alignment errors; for candidate acquisition actions (For example, requesting a high-precision timestamp for a sensor, or requesting synchronized sampling of a pair of sensors), its The calculation is as follows:

[0069]

[0070] In the formula, To perform the data collection action The value-oriented information, namely the reduction in expected decision-making losses; For data that may be collected in the future The mathematical expectation; In the existing dataset Under these conditions, decision model parameters The corresponding decision loss function value; To assume that an action will be performed After acquiring new data, the dataset is updated to... The decision loss function value under the given conditions; For a candidate data collection action;

[0071] In practice, Bayesian approximation sampling and Monte Carlo estimation are used:

[0072]

[0073] In the formula, The estimated value of VOI is calculated approximately using Monte Carlo sampling; The total number of Monte Carlo samples; In the first In the next sampling, based on the current dataset The simulated decision loss value; In the first In this sampling, actions are performed based on assumptions. The dataset obtained later The simulated decision loss value; and In the In each sampling, the current data is used respectively. and assumed to perform actions Subsequent "time alignment improvement" data To assess decision-making losses (such as flood prediction errors); in simulation generation At that time, it not only simply aligns the timestamps, but also updates them according to step S2. The propagation model performs probabilistic compensation and reconstruction of the data time of other sensors that are dependent on the action target sensor, thereby more realistically reflecting the benefits of performing the synchronous action to the entire data network.

[0074] Action space construction: Based on DTDG and node type, define more refined actions:

[0075] High-precision time synchronization request: to a single sensor Send time synchronization command;

[0076] Paired collaborative sampling request: To those with strong dependencies ( ) sensor Send a synchronous data acquisition command;

[0077] Dependency chain synchronization trigger request: For a critical propagation path, send synchronization instructions to the beginning and end nodes of the path, and request intermediate nodes to perform data time compensation marking according to the estimated time delay;

[0078] Historical data timestamp repair query: Send a request to the data center to re-correct or interpolate the timestamps of data from a specified sensor for a specific time period.

[0079] Step S4: Cost-Perceived Optimization Decisions

[0080] The decision objective is to select the action that maximizes the net information value within a limited communication, computation, and energy budget; the optimization problem is defined as:

[0081]

[0082] In the formula, The selected optimal data collection action; For from the action set Choose the action that maximizes the objective function value. The set of all candidate acquisition actions; For action The estimated value of VOI; This is a cost-weighting factor used to adjust the trade-off between information value and acquisition cost; To perform the action Total cost required This includes: communication costs (related to data packet size and transmission distance), energy consumption costs (energy consumption for sensor activation sampling and communication), and computational costs (computational overhead for time compensation reconstruction). For "paired collaborative sampling" or "dependency chain synchronization", the cost is not simply the sum of single-point costs, but takes into account the efficiency of joint scheduling, and may be eligible for a discount.

[0083] Decision-making considerations: In evaluating actions of In this process, not only is the direct reduction in uncertainty to the target node considered, but the uncertainty to downstream dependent nodes is also estimated using the DTDG model. This indirectly reduces the benefits, thereby enabling network-level VOI assessment.

[0084] Step S5: Generate executable acquisition instructions with time compensation.

[0085] Optimal action It is passed to the instruction generation module; this module generates machine-executable instructions based on the action type and relevant information in the DTDG.

[0086] Instruction generation mechanism: Instructions not only include "what to collect", but also "how to process according to time dependencies";

[0087] Example 1 (Paired Co-sampling Request):

[0088] SYNC_COLLECT--target_sensor_id=STATION_A--paired_sensor_id=STATION_B--trigger=ABSOLUTE_TIME--time=T2023-10-27T08:00:00Z--post_process=TIME_LAG_COMPENSATION--ref_sensor=STATION_A--estimated_lag=300s;

[0089] The instruction requires STATION_B and STATION_A to synchronize sampling at absolute time T, and informs the edge gateway or data center that after receiving the data from STATION_B, it needs to refer to the data from STATION_A, use an estimated time delay of 300 seconds for compensation and alignment processing before storing it in the database.

[0090] Example 2 (Dependency chain synchronization trigger request):

[0091] CHAIN_SYNC--chain=[STATION_U,STATION_M,STATION_D]--initiate_from=STATION_U--propagation_delays=[1800s,1200s]--action=STAMP_AND_COMPENSATE;

[0092] This instruction requires the initiation of a chain synchronization starting from the upstream STATION_U. STATION_M and STATION_D will trigger sampling sequentially after the estimated propagation delay, and the chain synchronization sequence number will be noted when the data is labeled, so that subsequent coherent time delay compensation analysis can be performed.

[0093] Step S6: Instruction Execution and Dynamic Time Dependency Graph Update

[0094] The generated instructions are sent to the corresponding sensor gateway or data acquisition device via standard APIs (such as RESTful interfaces and message queues); the system waits for the acquisition results to be returned.

[0095] Data reception and processing: The returned data packet contains new high-precision timestamps or synchronous sampling data; the system calls time compensation algorithms (such as time delay filtering or data interpolation based on DTDG) to process the relevant data stream according to the "post_process" prompt in the instruction;

[0096] Dynamically updated:

[0097] Use the timestamp information from the new data to update the self-time offset uncertainty estimate involving the sensor. ;

[0098] Recalculate the observation time delay between related sensor pairs using the synchronized data. And update the weight parameters of the corresponding edges in the DTDG. ;

[0099] The updated and based on the new DTDG calculation Feedback is sent back to the upstream USTT module to update the posterior distribution of the decision model, thereby closing the active learning loop.

[0100] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0101] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The invention is limited only by the claims and their full scope and equivalents.

Claims

1. A method for generating data collection instructions based on uncertainty traceability and value assessment, characterized in that, The method includes the following steps: Receive an uncertainty decomposition structure, the uncertainty decomposition structure containing at least one node representing a source of uncertainty, each node being associated with a data source and containing a quantitative assessment of the uncertainty of that data source; Based on the aforementioned uncertainty decomposition structure, for each node, the information value that can be obtained by collecting additional data from the corresponding data source is evaluated. The information value is the expected reduction in decision loss before and after the collection action is performed. Based on the source type of each node, define at least one candidate data acquisition action and construct an action space; Based on the calculated information value of each candidate data acquisition action and the estimated cost of performing the action, an optimization decision is made to select the optimal data acquisition action. The optimal data acquisition action is converted into standardized data acquisition instructions that can be executed by the machine; The data acquisition command is output to drive the external data acquisition system to perform the corresponding data acquisition operation. 2.The data collection instruction generation method based on uncertainty trace and value evaluation according to claim 1, wherein, In the step of receiving the uncertainty decomposition structure, the uncertainty decomposition structure is a tree structure, and each node further includes the influence weight of the uncertainty source on the final decision task and the source type identifier; the source type includes at least one of time offset uncertainty, parameter uncertainty, structural uncertainty and data missing. 3.The data collection instruction generation method based on uncertainty trace and value evaluation according to claim 1, wherein, The steps for assessing the value of the information specifically include: For each candidate data collection action, multiple samples are taken from the posterior distribution of the decision model parameters; For each sampling, calculate the decision loss when using the current dataset, and the decision loss of the dataset after simulating the sampling action and incorporating the new data. The information value estimate of the sampling action is obtained by averaging the reduction in decision loss across all sampling attempts. 4.The data collection instruction generation method based on uncertainty trace and value evaluation according to claim 1, wherein, In the step of constructing the action space, candidate data acquisition actions are mapped to different operation categories according to the source type of the nodes; for time offset uncertainty, it is mapped to the operation of requesting time synchronization; for parameter uncertainty, it is mapped to the operation of requesting supplementary observation samples; for structural uncertainty, it is mapped to the operation of triggering verification experiments; and for data missing, it is mapped to the operation of requesting external data source retrieval. 5.The data collection instruction generation method based on uncertainty trace and value evaluation according to claim 1, wherein, The optimization decision-making step is achieved by solving an optimization problem. The objective function of the optimization problem is the difference between the estimated information value and the acquisition cost after weight adjustment. The candidate action that maximizes this difference is selected as the optimal data acquisition action. 6.The data collection instruction generation method based on uncertainty trace and value evaluation according to claim 1, wherein, The step of converting the data into machine-executable instructions is implemented through a predefined instruction template. The instruction template is configured according to the action type, and the parameters in the optimal data acquisition action are filled into the corresponding positions of the template to generate an instruction string containing the target data source identifier, acquisition type, and necessary security parameters.

7. The data collection instruction generation method based on uncertainty trace and value evaluation according to claims 1 to 6, characterized in that, After outputting the data acquisition command, a feedback update step is also included: Receive acquisition result data returned by an external data acquisition system according to the instructions; The posterior probability distribution of the decision model is updated using the collected data. Based on the updated posterior distribution, re-evaluate the uncertainty quantification assessment of each node in the uncertainty decomposition structure, forming a closed-loop learning process.

Images