Intelligent semantic data collection and experiment design system

Abstract

The invention discloses an intelligent semantic data collection and experiment design system. The system can automatically formulate an optimal data collection and experiment plan under the condition of giving a research or decision target (such as verifying a certain hypothesis or searching a certain equipment fault reason) of a user, and the optimal data collection and experiment plan comprises variables needing to be measured, equipment configuration, sampling frequency, experiment steps and the like. The system utilizes a knowledge base to infer key information gaps related to a target, evaluates the value of different potential data sources in the aspect of reducing uncertainty, and further generates an experiment or data collection scheme with the highest cost performance. By using the system, artificial intelligence is improved from passive analysis of existing data to active planning of data collection, the efficiency in the fields of scientific research, medical diagnosis, industrial troubleshooting and the like can be remarkably improved, and the system has a wide application prospect.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of artificial intelligence and experimental design, and particularly relates to an intelligent semantic data collection and experimental design system for automatically planning a data collection scheme according to a target task. BACKGROUND

[0002] In the process of scientific research and new product development, it is often dependent on the experience and intuition of experts to determine what data to collect and how to design experiments. Existing artificial intelligence systems lack the ability to actively obtain the required data, and most AI algorithms assume that the required data has been given, rarely involving the step of actively planning data collection from the research purpose. Therefore, when facing a new research problem or diagnosis task, researchers need to determine which data needs to be collected and which experimental steps to take, which is time-consuming and laborious and is easily affected by subjective judgment, which may lead to low experimental efficiency or missing key data. In particular, in the fields of scientific research, medical diagnosis, and industrial fault diagnosis, due to the lack of systematic intelligent experimental design tools, it is often necessary to repeat experiments multiple times to find the cause of the problem or verify the hypothesis, which is costly and time-consuming. SUMMARY

[0003] The present application aims to solve the limitations of passive data acceptance of artificial intelligence systems in the background art and provides an intelligent system that can automatically develop a data collection and experimental plan according to a user-proposed research or decision-making target. Through this system, artificial intelligence can actively determine the type of data to be obtained and the experimental method, obtaining the maximum useful information at the minimum cost, thereby greatly improving the efficiency of scientific research and diagnosis.

[0004] To achieve the above-mentioned purpose, the technical solution of the present application is as follows:

[0005] The intelligent semantic data collection and experimental design system includes the following modules:

[0006] (1) Target semantic analysis module: for obtaining and analyzing the user input research purpose or problem description to be solved, and locating the concepts and existing knowledge information related to the target in the knowledge base or knowledge graph, and determining the current knowledge boundary and existing uncertainty area. For example, when the user wants to diagnose a machine fault, the module discovers through semantic analysis that the common fault causes of the device in the knowledge base include overheating and component wear, but the current information is not sufficient to determine which one of them caused the fault. Therefore, the system determines "whether the fault cause is overheating or wear" as the uncertainty to be eliminated.

[0007] (2) Value assessment module: for generating a plurality of data collection or experiment candidate schemes for the uncertain places, and assessing the value of each candidate scheme. The value assessment comprehensively considers factors such as information gain and acquisition cost, and can calculate the expected information benefit of each candidate data based on information theory indicators (such as information gain or entropy reduction), statistical models or Bayesian optimization methods, that is, its role in reducing target uncertainty, and at the same time estimate the cost and time cost required to obtain the data. For example, in the above machine failure diagnosis situation, candidate schemes can include: "measuring temperature data of key parts of the equipment", "collecting equipment vibration signals", "extracting lubricating oil samples for component analysis", etc. For each candidate scheme, the system estimates its information value in distinguishing fault causes and acquisition cost: measuring temperature can provide direct evidence to support overheating failure, but requires temporary shutdown to install sensors, with high time and economic cost; vibration signals can be obtained without stopping, with low cost, but can only indirectly reflect the fault type, with limited diagnostic information; lubricating oil sample analysis can detect metal particles generated by wear, with high information value but requiring certain fees and a long time to wait for laboratory analysis results. The value assessment module will quantitatively evaluate such candidate schemes and give the expected information benefit and corresponding cost indicators of each candidate.

[0008] (3) Scheme optimization module: for performing combined optimization on the candidate scheme set to select a number of data collection actions to form a final scheme, so that the overall information value is maximized under the constraints of cost, time, etc. This module balances the goals of "maximizing uncertainty reduction" and "minimizing resource consumption" through multi-objective optimization algorithms, evaluates and compares different candidate combinations, searches for a Pareto optimal solution set, and selects the best one for output. Optimization algorithms can use heuristic search, genetic algorithm, particle swarm optimization or other problem solving methods suitable for high-dimensional combinatorial optimization, to iteratively optimize candidate scheme combinations and finally determine the experiment scheme combination with the maximum information benefit and acceptable resource cost.

[0009] (4) Experiment plan generation module: for converting the selected optimized scheme into a specific executable experiment plan, including required measurement variables, instrument configuration, sampling method, experiment step sequence, etc. If it involves complex experiments, it can also provide arrangement schemes for control experiments and method suggestions for subsequent data analysis. This module presents the final experiment plan to the user in the form of natural language description and flowchart, making it easy for the user to review and understand. If the user has special preferences or constraint requirements, the system also supports adjusting the scheme according to user feedback, re-calling the scheme optimization module to calculate a new optimal combination, and generating an updated experiment plan by the experiment plan generation module.

[0010] The application realizes the conversion of artificial intelligence from "passive analysis of existing data" to "active planning of new data" through the cooperation of the above modules. Compared with the prior art, the system can automatically identify knowledge gaps and design targeted experimental schemes according to the high-level goals of the user, greatly reducing the number of manual trial and error and improving the efficiency of data collection and experimental design. In the fields of scientific research, medical diagnosis and industrial fault diagnosis that require decision support, the application can help users obtain effective conclusions more quickly and at a lower cost. In addition, the application fills the gap of "directly driven data acquisition" by intention in the data-information-knowledge-wisdom-intention (DIKWP) closed loop, realizes the data collection link directly guided by specific actions from target intention, and this function perfects the data-knowledge-intention closed loop process of the artificial intelligence decision system, which has important technical significance. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 is the overall architecture schematic diagram of the intelligent semantic data collection and experimental design system of the application. DETAILED DESCRIPTION

[0012] The working process of the application will be described below in combination with specific examples.

[0013] Firstly, the user inputs the target description of research or diagnosis through the man-machine interface, for example: "a device in a certain factory has a fault, and the cause needs to be found out". After receiving the description, the target semantic analysis module queries the knowledge base in the field of device fault diagnosis. Through semantic analysis, the system finds that the common fault causes in the existing knowledge for the device include overheating and component wear, but the current information is not sufficient to determine which cause leads to this fault. Therefore, the system determines "whether the fault is caused by overheating or wear" as the uncertainty to be eliminated at present.

[0014] Next, the value evaluation module generates several candidate data collection schemes for this uncertainty. The system retrieves the current available detection methods and obtains candidate schemes including: "measuring the temperature of the key parts of the device", "collecting the vibration signal of the device", "conducting lubricating oil sample composition analysis", etc. For each candidate scheme, the system estimates its effect and cost in distinguishing the fault causes: measuring the temperature may provide direct evidence to support the overheating fault, but it needs to install a temperature sensor during shutdown, which has a high cost; the vibration signal can be collected when the device is running without shutdown, which has a low cost, but the diagnostic information is limited; the lubricating oil sample analysis can detect the particulate matter generated by wear, which has high information value, but it consumes certain detection fees and needs to wait for the laboratory analysis results. The value evaluation module assigns an information benefit score and a corresponding cost index to each candidate.

[0015] Subsequently, the scheme optimization module performs combinatorial optimization on these candidate schemes. Assuming that the system has pre-set upper limit constraints on total downtime and budget, the optimization module will evaluate various scheme combinations, such as: "measure temperature and vibration signals simultaneously", "measure temperature only", "vibration monitoring plus oil sample analysis", etc. By multi-objective optimization algorithm, the system calculates the total information gain and total cost of each combination, and finds the scheme with the highest information gain within the resource constraints. For example, a possible optimization result is to recommend the combination of vibration monitoring and oil sample analysis, because this combination does not require long downtime and can comprehensively judge the fault type, with the highest comprehensive information gain. This combination is selected as the optimal experimental strategy.

[0016] Finally, the experimental plan generation module generates a specific execution plan according to the optimization results. The final output plan will include:

[0017] - Device vibration data collection: Use the specified vibration sensor to continuously monitor the device vibration for 30 minutes under normal operating conditions, with a sampling frequency of 1 kHz;

[0018] - Lubricating oil sample analysis: Take 50 mL of lubricating oil sample from the device oil tank and send it to the laboratory for metal particle composition detection, and the detection result is expected to be obtained within 48 hours.

[0019] The plan clearly specifies the required equipment (such as vibration sensor model), operation steps and time arrangement. The system presents this plan to the user in the form of a report, which contains natural language description and flowchart illustration, making it easy to understand and execute.

[0020] After the user reviews the experimental plan, if he is not satisfied with some aspects, for example, he wants to avoid downtime as much as possible or wants to add another detection means (such as acoustic signal analysis), he can feedback his preferences. After receiving the new constraints or preferences, the system will recalculate the optimal scheme by the scheme optimization module, and output the updated plan by the experimental plan generation module. The whole process can be iterated according to user feedback until the user is satisfied with the scheme.

[0021] Through the above embodiment, it can be seen how the intelligent semantic data collection and experimental design system of the present application works in practical scenarios. This system can be widely used in experimental and data acquisition decisions in different fields, such as: in medical diagnosis, the system can suggest the most information valuable combination of examination items to assist in the diagnosis of difficult cases; in scientific research, the system can plan a series of experiments to verify scientific hypotheses; in the engineering field, the system can guide sensor placement and testing to quickly locate faults. As can be seen, the present application has significant advantages in improving data acquisition efficiency and decision accuracy in complex decision-making scenarios.

Claims

1. A method of intelligent semantic data collection and experimental design, characterized by, The method comprises the following steps: (1) receiving a semantic description of a research or decision target input by a user, analyzing the description, identifying known information and unknown information related to the target in a knowledge base or knowledge graph, and determining existing knowledge gaps or uncertainties; (2) based on the determined uncertainties, generating a plurality of data collection or experiment candidate schemes, including different types of available data sources or executable experiment operations, and estimating the expected information return value of each candidate scheme and the cost of obtaining the data; (3) taking the plurality of candidate schemes as a candidate set, selecting a scheme combination with the maximum information return under the premise of meeting the preset resource constraints by using a multi-objective optimization algorithm, and determining an optimal data collection and experiment scheme for reducing the uncertainties; and (4) converting the selected optimal scheme into a specific experiment plan, including required measurement variables, instrument device configurations, sampling methods and step sequences, and providing the experiment plan to the user.

2. The method of claim 1, wherein, In step (1), the key entities and associated relationships involved in the target description are located by semantic analysis combined with knowledge graph technology, so as to identify the knowledge gaps or uncertainties.

3. The method of claim 1, wherein, The expected information return in step (2) is calculated by information theory indicators or Bayesian inference methods to quantify the value of each candidate data in distinguishing target situations or reducing uncertainties, and the resource consumption required to obtain each candidate data is evaluated by combining a preset cost model.

4. The method of claim 1, wherein, In step (3), the multi-objective optimization algorithm uses heuristic search, genetic algorithm or particle swarm optimization algorithm to evaluate the candidate scheme combinations in terms of information return and cost, and selects solutions meeting the Pareto optimality condition as the optimal data collection and experiment scheme.

5. The method of claim 1, wherein, In step (4), the generated experiment plan includes control test arrangements for complex experiments and methods for subsequent data analysis, and the experiment plan is output in the form of natural language description and flowchart for the user to understand and execute.

6. An intelligent semantic data collection and experimental design system, characterized by, The method comprises: (1) a target semantic analysis module for obtaining and analyzing the target description input by the user, and identifying related known information and knowledge gaps; (2) a value evaluation module for generating a plurality of data collection or experiment candidate schemes for the knowledge gaps, and estimating the expected information return and acquisition cost of each candidate scheme; (3) a scheme optimization module for selecting an optimal scheme combination from the candidate schemes to maximize the information return under the condition of meeting the preset cost or time constraints; and (4) an experiment plan generation module for converting the optimal scheme into a specific experiment execution plan and outputting the experiment plan to the user.

7. The system of claim 6, wherein, The target semantic analysis module uses a knowledge graph or domain knowledge base for semantic reasoning to determine the key elements involved in the target description and the corresponding unknown problems.

8. The system of claim 6, wherein, The value evaluation module comprises an information evaluation unit and a cost evaluation unit, wherein the information evaluation unit is used to calculate the expected value of the candidate schemes based on information gain, entropy reduction or Bayesian update method, the cost evaluation unit is used to estimate the resource consumption required to obtain each candidate data, and the comprehensive value score of each candidate scheme is output.

9. The system of claim 6, wherein, The scheme optimization module performs a heuristic optimization process such as a genetic algorithm, a particle swarm optimization or a simulated annealing algorithm, and performs multi-objective combination optimization on the candidate scheme set to obtain a Pareto optimal solution satisfying the preset conditions.

10. The system of claim 6, wherein, The experiment plan generation module is further configured to adjust the experiment scheme according to user feedback, and when a user provides a new preference or constraint condition, the scheme optimization module is re-invoked to generate an updated optimal scheme and output a corresponding experiment plan.

Images