A sensor knowledge graph generation method, system and application

Abstract

This application proposes a sensor knowledge graph generation method, system, and application, including: determining the monitoring area and target sensor; constructing extreme entities and adjusting the graph focusing direction; analyzing the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions; analyzing the validity of historical detection data and the correlation between extreme entities and the target sensor's detection accuracy; expanding the detection data; and constructing a knowledge graph based on the expanded data. This application targets sensor detection scenarios in special environments, and through a technical approach of accurate modeling of extreme entities, data expansion and quality improvement, multi-dimensional knowledge integration, and dynamic reasoning adaptation, constructs a sensor knowledge graph that can deeply adapt to special environments, significantly improving the reliability, accuracy, and practicality of the knowledge graph in special environmental scenarios.

Description

Technical Field

[0001] This application relates to the field of sensor knowledge graphs, and more specifically, to a sensor knowledge graph generation method, system, and application. Background Technology

[0002] In fields such as industrial control and geological exploration, sensors need to accurately detect environmental parameters in special environments such as high temperature, high pressure, strong radiation, and high corrosion. The parameters in these special environments often exceed conventional ranges, exhibiting characteristics such as wide parameter ranges, complex interference factors, and drastic dynamic changes. These factors significantly affect the sensor's detection mechanism and pose a severe challenge to its detection performance. Sensor knowledge graphs are an important application of knowledge graphs in the sensor field. Through a semantic network that connects structured data with sensor entities, environmental parameters, and detection mechanisms, they effectively support the in-depth interpretation of sensor data.

[0003] However, existing sensor knowledge graph generation methods are mainly designed for conventional environments, constructing entities and relationships based on normal environmental parameter ranges. They do not design dedicated entities for extreme parameters in special environments, resulting in the graphs failing to accurately depict the relationship logic between special environments and sensors, thus limiting their application in special environments.

[0004] Therefore, there is an urgent need for a sensor knowledge graph generation scheme that can adapt to the characteristics of extreme environments. Summary of the Invention

[0005] This application provides a sensor knowledge graph generation method, system, and application to solve at least one of the above-mentioned problems. The specific solution is as follows:

[0006] In its first part, this application proposes a method for generating a sensor knowledge graph, including:

[0007] Determine the monitoring area and the target sensors located within the area for detecting target environmental parameters;

[0008] Based on the environmental characteristics of the special environment of the monitoring area, one or more extreme entities are constructed, and the focus direction of the map is adjusted based on whether the target environmental parameter is an environmental parameter represented by a certain extreme entity.

[0009] Acquire historical detection data and multi-source heterogeneous data of the target sensor, and analyze the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions;

[0010] Based on the aforementioned extreme entities, detection mechanisms, and environmental factors affecting detection accuracy, the validity of the historical detection data and the correlation between extreme entities and the target sensor detection accuracy are analyzed.

[0011] Predictive detection data is obtained by acquiring target environmental parameters of the monitoring area from a preset external data source, and expanded data is obtained by integrating the predicted detection data with valid historical detection data.

[0012] Based on the expanded data, entity attributes and relationship weights are set, and the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay are integrated into a knowledge graph that includes extreme entities, sensor entities, and detection performance entities.

[0013] In some specific embodiments, adjusting the focus direction of the spectrum specifically includes:

[0014] If the target environmental parameter is an environmental parameter represented by an extreme entity, then increase the weight between the target environmental parameter and the extreme entity to strengthen the correlation between the two.

[0015] If the target environmental parameters are different from the environmental parameters represented by any extreme entity, then the interference relationship between each extreme entity and the detection of the target environmental parameters is strengthened.

[0016] In some specific embodiments, constructing one or more extreme entities specifically includes:

[0017] Extract the threshold values ​​of differences between special environments and normal environments in physical parameters including temperature, humidity, pressure, and radiation dose; chemical parameters including concentration of corrosive media and redox potential; or spatial parameters including gradient change rate and instantaneous fluctuation amplitude.

[0018] The attributes of extreme entities are defined based on the difference threshold, including special environment type, parameter range, and duration.

[0019] In some specific embodiments, the acquisition of the predictive detection data specifically includes:

[0020] If the target environmental parameter has a quantitative relationship with other environmental parameters, historical co-occurrence data of the target environmental parameter and related environmental parameters are extracted from external data sources, the quantitative relationship is fitted by multiple linear regression, the measured value of the current related environmental parameter in the monitoring area is input, and the predicted value of the target environmental parameter is calculated by the quantitative relationship to obtain the predicted detection data.

[0021] And / or, select real-time detection data from other types of sensors deployed in the same area as the target sensor, perform spatiotemporal alignment and outlier filtering on the detection data, and derive predicted values ​​of the target environmental parameters through a data fusion algorithm to obtain predicted detection data.

[0022] In some specific embodiments, the method further includes:

[0023] Extract the parameter range of the extreme entity and the tolerance threshold of the detection mechanism, and determine whether the parameters of the extreme entity exceed the tolerance threshold of the detection mechanism;

[0024] If so, the historical detection data is deemed invalid because the extreme entities are obstructing the operation of the detection mechanism.

[0025] If not, analyze the parameter range of environmental factors affecting detection accuracy under the special environment and predict the detection error caused by the parameter range. Based on the detection error, determine whether each part of the historical detection data is valid.

[0026] In some specific embodiments, after determining the validity of the historical detection data, the method further includes:

[0027] Extreme entity parameter values, environmental factor values ​​affecting accuracy, and corresponding accuracy error values ​​are extracted from valid historical detection data. A multiple regression model of extreme entity parameters and environmental factor values ​​is constructed to quantify the correlation between extreme entities and the target sensor detection accuracy.

[0028] The relationships are stored in the knowledge graph as triples of extreme entities, environmental factors, and detection accuracy, with attributes including model parameters, applicable scope, and error contribution rate.

[0029] In some specific embodiments, the method further includes:

[0030] The impact path of extreme entities on detection accuracy is analyzed and path feature parameters are extracted. The impact path includes the direct impact path acting on the sensor's sensitive element and the indirect impact path affecting detection accuracy through the transmission link or auxiliary components that interfere with the sensor.

[0031] A multi-dimensional association dataset is constructed based on path feature parameters. The dataset includes the correspondence between extreme entities and detection accuracy in static data, dynamic data, and interactive data, respectively.

[0032] By combining the path feature parameters and associated datasets, a dynamic correlation model is established between extreme entity parameters and detection accuracy, involving both single-path and multi-path interaction effects.

[0033] Instantiate the association between extreme entities and detection accuracy in the knowledge graph, and add a confidence attribute based on the data sample size and model error.

[0034] In the second part, this application proposes a sensor knowledge graph generation system, including:

[0035] The target determination unit is used to determine the monitoring area and the target sensors located in the area for detecting target environmental parameters.

[0036] The difference analysis unit is used to construct one or more extreme entities based on the environmental characteristics of the special environment of the monitoring area, and to adjust the focus direction of the map based on whether the target environmental parameter is an environmental parameter represented by a certain extreme entity.

[0037] The first analysis unit is used to acquire historical detection data and multi-source heterogeneous data of the target sensor, and to analyze the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions.

[0038] The second analysis unit is used to analyze the validity of the historical detection data and the correlation between the extreme entities and the detection accuracy of the target sensor based on the extreme entities, the detection mechanism, and the environmental factors affecting the detection accuracy.

[0039] The data expansion unit is used to obtain target environmental parameters of the monitoring area from a preset external data source to obtain predictive detection data, and to integrate the predictive detection data with valid historical detection data to obtain expanded data.

[0040] The graph construction unit is used to set entity attributes and relationship weights based on the expanded data, and integrate the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay into a knowledge graph containing extreme entities, sensor entities, and detection performance entities.

[0041] Part Three: This application discloses a computer device, said computer device comprising:

[0042] One or more processors;

[0043] Memory, used to store one or more programs;

[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement a sensor knowledge graph generation method as described in any of the first parts.

[0045] In Part Four, this application proposes a computer program product including executable instructions for implementing, when executed by a processor, a sensor knowledge graph generation method as described in any of Part One.

[0046] Beneficial effects: This application proposes a sensor knowledge graph generation method, system, and application. Targeting sensor detection scenarios in special environments, it constructs a sensor knowledge graph that can deeply adapt to special environments through technical paths such as precise modeling of extreme entities, data expansion and quality improvement, multi-dimensional knowledge integration, and dynamic reasoning adaptation. This solves the problems of traditional graphs in extreme scenarios, such as blind spots, insufficient data support, and knowledge fragmentation, and significantly improves the reliability, accuracy, and practicality of knowledge graphs in special environmental scenarios.

[0047] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart illustrating the method described in this application;

[0050] Figure 2 This is a schematic diagram illustrating the principle of the method described in this application;

[0051] Figure 3 This is an example diagram illustrating the data validity assessment in this application;

[0052] Figure 4 This is a schematic diagram of the system modules of this application.

[0053] Figure labeling: 1-Target determination unit; 2-Difference analysis unit; 3-First analysis unit; 4-Second analysis unit; 5-Data augmentation unit; 6-Map construction unit. Detailed Implementation

[0054] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0055] This application proposes a sensor knowledge graph generation method. By systematically constructing a structured association system of special environment, sensor, and detection performance, it solves the problems of insufficient adaptability, ambiguous data validity, and one-sided association relationships in existing technologies for sensor knowledge modeling in special environments. It achieves end-to-end optimization from data acquisition to knowledge application, significantly improving the reliability, accuracy, and practicality of the knowledge graph in special environmental scenarios. A flowchart of the sensor knowledge graph generation method is attached. Figure 1 As shown in the attached diagram, the principle is as follows. Figure 2 As shown, the specific solution is as follows:

[0056] A method for generating a sensor knowledge graph, comprising:

[0057] 101. Determine the monitoring area and the target sensors located within the area for detecting target environmental parameters;

[0058] 102. Construct one or more extreme entities based on the environmental characteristics of the special environment of the monitoring area, and adjust the focus direction of the map based on whether the target environmental parameters are the environmental parameters represented by a certain extreme entity;

[0059] 103. Obtain historical detection data and multi-source heterogeneous data of the target sensor, and analyze the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions.

[0060] 104. Analyze the validity of historical detection data and the correlation between extreme entities and the detection accuracy of target sensors based on extreme entities, detection mechanisms, and environmental factors affecting detection accuracy;

[0061] 105. Obtain the target environmental parameters of the monitoring area from the preset external data source to obtain the predicted detection data, and integrate the predicted detection data with the effective historical detection data to obtain the expanded data;

[0062] 106. Based on the expanded data, set entity attributes and relationship weights, and integrate the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay into a knowledge graph that includes extreme entities, sensor entities, and detection performance entities.

[0063] This application clarifies the construction logic of entities in extreme environments, the criteria for judging data validity, the methods for depicting multi-dimensional relationships, and the strategies for fusing external data, thereby achieving accurate modeling of sensor detection patterns in extreme environments and providing support for data interpretation and performance optimization.

[0064] Step 101 is responsible for locating the research object, ensuring that subsequent steps focus on sensor data from a specific scenario. The monitoring area refers to the specific spatial range where environmental parameters need to be detected. This area can be defined by geographical boundaries, physical scenes, or functional attributes, and it possesses a unique environment distinct from the conventional environment. The target sensor is a device deployed within the monitoring area specifically designed to collect target environmental parameters; the environmental parameters it collects are the target environmental parameters. For example, in a high-temperature furnace, a temperature sensor detects the temperature; the temperature sensor is the target sensor, and the temperature is the target environmental parameter.

[0065] Step 102 involves constructing extreme entities and adjusting the focus direction of the knowledge graph, enabling the knowledge graph to accurately depict the essential characteristics of special environments while avoiding interference from irrelevant information and improving reasoning efficiency. It should be noted that the sensor application scenario addressed in this application is a special environment. A special environment refers to an environment in which one or more environmental parameters in the monitoring area significantly deviate from the normal range, and the degree of deviation is sufficient to substantially affect the operating state of the sensor. At least one environmental parameter (such as temperature, pressure, radiation dose, concentration of corrosive media, etc.) has a value significantly higher or lower than the conventional environmental threshold. For example, the normal industrial environment temperature is usually -20℃ to 50℃, while the 800℃ high temperature in a volcanic eruption zone and the 100MPa high pressure at a depth of 1000 meters in the deep sea are both parameter anomalies; the 1000Gy / h radiation dose around a nuclear reactor far exceeds the natural background radiation, which is also a typical parameter anomaly. Abnormal parameters can interfere with the normal operation of sensors through physical, chemical, or mechanical means. For example, extreme high temperatures can cause the insulation layer of internal wires to melt and short-circuit; strong electromagnetic environments can induce noise currents in circuits, disrupting signal transmission; strong radiation can cause abnormal changes in the carrier concentration of semiconductor sensors, leading to the failure of the photoelectric effect-based detection principle; high concentrations of corrosive gases can react chemically with the sensor's sensitive elements, altering their physical properties; and extreme high pressure can cause deformation of the sensor housing, compressing internal sensitive components and compromising its mechanical structural stability. In short, the essence of special environments is a combination of parameter anomalies and sensor impact, which is distinct from normal environments with stable parameters, and also different from ordinary extreme scenarios where parameters are abnormal but have no substantial impact on the sensor.

[0066] Environmental characteristics refer to environmental features in the monitoring area that exceed normal ranges, such as high temperature (>500℃), high pressure (>100MPa), strong radiation (>1000Gy), and high corrosion (Cl⁻ concentration >1000mg / L). Extreme entities are structured knowledge units built based on these special environmental characteristics to represent the core features of extreme environments. Their attributes include environmental type, parameter range, and duration. This transforms abstract special environments into quantifiable and associative entities, solving the problem of directly incorporating special environments into the knowledge graph. The graph's focus direction prioritizes the relationships between entities within the knowledge graph. By judging the correlation between target parameters and extreme entities, core relationships are strengthened, secondary relationships are weakened, and the core logical clarity of the graph is improved.

[0067] For example, a special environment might be a volcanic eruption zone, characterized by temperatures of 800℃~1200℃ and a continuous eruption lasting 3 hours, to construct an extreme high-temperature volcanic entity. If the target environmental parameter is temperature, a core parameter for representing the extreme entity, the correlation weight between temperature and the high-temperature volcanic entity is increased; if the target parameter is humidity, the correlation between the high-temperature volcanic entity and humidity detection interference is strengthened.

[0068] In some specific embodiments, constructing one or more extreme entities specifically includes: extracting the difference thresholds between special environments and normal environments in terms of physical parameters (including temperature, humidity, pressure, and radiation dose), chemical parameters (including corrosive medium concentration and redox potential), or spatial parameters (including gradient change rate and instantaneous fluctuation amplitude); defining the attributes of the extreme entities based on the difference thresholds, including special environment type, parameter range, and duration. The essence of constructing extreme entities is to transform abstract special environments into entity units recognizable by the knowledge graph through parameter difference quantification and attribute structure definition. The difference thresholds avoid subjective judgments about extreme environments, making entity construction repeatable. The parameter range, duration, and other attributes of the extreme entities provide key inputs for data validity judgment and accuracy correlation modeling in step 104. The special environment type attribute clarifies the essence of the extreme entities, enabling the knowledge graph to clearly distinguish between different influence paths such as material property changes caused by high temperatures and chemical corrosion caused by high salt.

[0069] Extract the difference threshold between special and normal environments. The difference threshold is a quantitative standard for judging whether an environment is extreme, and it needs to be calculated separately for three types of parameters: physical, chemical, and spatial. Physical parameters include temperature, humidity, pressure, radiation dose, etc., and the threshold is determined by comparing the parameter ranges of special and normal environments. Chemical parameters include the concentration of corrosive media, redox potential, etc., and the difference threshold is determined by analyzing the influence of the media on the sensor. Spatial parameters include the gradient rate of change (e.g., the rate of change of temperature with distance) and the instantaneous fluctuation amplitude (e.g., the change in pressure within 1 second), and the threshold is determined by the drasticness of the parameter changes.

[0070] The attributes of extreme entities are defined based on difference thresholds. These attributes must fully characterize the core features of the specific environment, including: Environment type: named according to the dominant parameter type, reflecting the essential characteristics of the environment. Parameter range: the specific range of the extreme parameters is defined based on the difference threshold, requiring refinement based on actual monitoring data. Duration: describes the stability or timeliness of the extreme environment, categorized as steady-state or transient.

[0071] In some specific embodiments, adjusting the focus direction of the knowledge graph includes: if the target environmental parameter is an environmental parameter represented by a certain extreme entity, then increasing the weight between the target environmental parameter and the extreme entity to strengthen the correlation between them; if the target environmental parameter is different from the environmental parameter represented by any extreme entity, then strengthening the interference relationship between each extreme entity and the detection of the target environmental parameter. Based on the relationship between the target parameter and the core features of the extreme environment, it is determined whether it is a direct correlation strengthening or an indirect interference strengthening. By dynamically adjusting the weights, the correlation strength of the knowledge graph matches the causal relationship of the actual scenario, ensuring that the core logic is prominent and the secondary logic is clear, ultimately improving the inference accuracy and efficiency of the knowledge graph. This adjustment method not only conforms to the actual laws of sensor detection under special environments, but also makes the focus direction operable through quantified weights, laying the foundation for the accurate application of knowledge graphs.

[0072] When the target environmental parameter is a core representation parameter of an extreme entity, the association is strengthened by increasing the weight of the relationship between the two. In such scenarios, there is a direct causal relationship between the target parameter and the extreme environment. For example, the temperature in a volcanic eruption zone is a core feature of an extreme high-temperature entity. Strengthening the association ensures that the knowledge graph prioritizes the core logic of the extreme environment-target parameter relationship, avoiding interference from secondary information. Assuming the association weight between temperature parameter and environmental temperature entity is 0.3 under normal conditions, in extreme high-temperature scenarios, if the target parameter is temperature, the weight of temperature parameter-extreme high-temperature entity is increased to 0.8~0.9 based on parameter correlation.

[0073] When the target environmental parameters are not core characterization parameters of extreme entities, the weighting of the interference relationship between extreme entities and target parameter detection is strengthened. Regarding the interference relationship between extreme entities and target parameter detection, the weighting is increased from 0.1~0.2 in normal environments to 0.5~0.6. Simultaneously, the type of interference is clearly defined, such as extreme high temperatures causing humidity sensor drift and strong radiation interference with vibration signal transmission. In such scenarios, although the extreme environment does not directly affect the physical properties of the target parameters, it indirectly affects accuracy by interfering with sensor hardware or the detection link. For example, high temperatures cause aging of the sensitive membrane of the humidity sensor. Strengthening the interference relationship ensures that the spectrum clearly depicts the transmission path of extreme environment - detection interference - target parameter error.

[0074] Step 103 involves acquiring historical detection data and multi-source heterogeneous data from the target sensor, and then analyzing the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions. This step is the data input and knowledge extraction stage for knowledge graph construction. By analyzing multiple types of data, it provides basic information for subsequent correlation analysis.

[0075] Historical detection data comprises environmental parameter data collected by the target sensor in the past, including timestamps, detection values, and anomaly markers. This type of data reflects the sensor's actual output in the real environment and serves as a direct basis for analyzing accuracy changes. Multi-source heterogeneous data is a collection of data from different sources and in different formats, including sensor hardware manuals, virtual data output from simulation platforms, and domain expert experience rules. The introduction of multi-source heterogeneous data can compensate for the limitations of historical detection data. The detection mechanism is the core principle by which the sensor converts environmental parameters into electrical signals, such as the Seebeck effect of thermocouples and the principle of dielectric constant change in capacitive sensors. Environmental factors affecting detection accuracy are environmental variables that lead to sensor detection errors, such as temperature drift, electromagnetic interference, and vibration. In practical applications, the detection mechanism and environmental factors affecting detection accuracy can be extracted from multi-source heterogeneous data. For example, the detection mechanism and environmental factors affecting accuracy can be extracted from a pressure sensor manual, and supplemented with relevant environmental factors based on domain expert experience.

[0076] Entities and relationships in a normal environment refer to knowledge units and associations that exist in a conventional environment, such as temperature sensor-detection-room temperature and humidity-impact-temperature sensor accuracy. This provides a normal environment benchmark for the knowledge graph. Subsequently, by comparing the differences between special environments and normal environments, the impact of extreme conditions can be highlighted. At the same time, the working principle and error sources of the sensor are clarified, laying the foundation for data validity analysis.

[0077] Multi-source heterogeneous data includes structured data, unstructured data, and semi-structured data. Specifically, it includes: Product list data (CSV format): providing a basic list of product names, reference codes, and subcategories. Product detail page data (CSV format): containing detailed product descriptions, manufacturer information, and highly unstructured detailed technical specifications stored as JSON strings. Category directory data (CSV format): defining the hierarchical structure of sensor categories. Filter label data (JSON format): defining the technical parameter filtering conditions (such as measurement principles) and their optional values ​​for specific categories, serving as the core basis for constructing the map classification and attribute skeleton. To address the data's non-standardization, this application employs a cleaning and normalization algorithm that integrates regular expressions and predefined rules to clean the multi-source heterogeneous data. The algorithm specifically includes:

[0078] Attribute key (field) standardization: Call the clean_attribute_key function to automatically remove extra spaces and special characters from attribute names, thus achieving field uniformity.

[0079] Semantic cleaning of attribute values: Call the `clean_filter_value` function, which has a built-in complex processing logic:

[0080] Irrelevant prefix / suffix removal: Using a predefined list of rules, accurately remove common, semantically meaningless leading words from attribute values ​​and extract the core semantic meaning.

[0081] Specific Format Preservation and Recognition: Utilizing regular expressions, this feature accurately identifies and fully preserves formats with domain-specific meanings, avoiding erroneous splitting of key technical parameters. Preserved characteristic value expression patterns include: Numerical range expressions: e.g., 0.1 to 10 bar; Composite numerical values ​​with units: e.g., 4–20mA, 24VDC, IP67; Industry standards and model codes: e.g., PG13, 5, G1 / 4, S7, VarioPin2.0; Complex certification numbers: e.g., II2GExiaIICT4...T6; Protocol fields: e.g., Modbus, IO-Link.

[0082] Multi-value intelligent decomposition: For composite values ​​(such as ATEX / IECEx) that are not in the above specific format but are connected by delimiters (such as / or -), they are safely decomposed to generate multiple independent atomic attribute value entities, ensuring the semantic uniqueness and accuracy of each node.

[0083] Robust data parsing: For non-standard JSON strings in the details page, a safe_json_loads secure parsing module is used for preprocessing. This module can automatically correct common format errors (such as single quotes replacing double quotes, extra commas at the end), which significantly improves the success rate and stability of extracting specification parameters from semi-structured data.

[0084] Next, using the cleaned data, entities and relationships are constructed within the graph data structure. A directed graph data structure is created using the NetworkX library. The core modeling units of the graph include five node types and six edge relationships. For example, the node construction is as follows:

[0085]

[0086] The logic for constructing edges (relationships) is as follows:

[0087]

[0088] The construction process and nested parameter handling include the following: 1. Constructing the classification skeleton: Processing classification directories and filter tag data, generating three types of nodes: Subcategory, FilterCategory, and FilterValue, and establishing the HAS_FILTER_CATEGORY and HAS_POSSIBLE_VALUE relationships to form the taxonomic framework of the graph. 2. Populating entities and attributes: Traversing product details and product list data. Creating Product and Manufacturer nodes and establishing the MANUFACTURED_BY relationship. For product technical specifications, especially nested structures (such as {Power Supply:{Voltage:24VDC, Power Consumption:0.5W}}), this method first extracts the first-level parameter group (Power Supply), and then recursively or flattens it to decompose it into independent product->attribute value relationships, i.e., product->24VDC and product->0.5W, ensuring that 24VDC and 0.5W are associated with the correct FilterCategory (voltage and power consumption) respectively. 3. Integration and Association: Finally, using the information in the product list, the BELONGS_TO_SUBCATEGORY relationship is established between Product and Subcategory, attaching all product entities to the category skeleton, thus completing the comprehensive integration of knowledge. In real-world transaction data, product specifications often exist in complex, nested structures.

[0089] Traditional methods may struggle to directly and effectively map this information into a flattened knowledge graph structure. This application addresses this by decomposing nested structures into independent product-attribute value relationships and associating them with the correct FilterCategory. This not only extracts surface data but also delves deeper into the semantics behind it, giving attribute values ​​like 24VDC and 0.5W not merely strings, but explicit meanings of voltage and power consumption. This lays a solid foundation for subsequent knowledge graph-based search, recommendation, and analysis. This approach allows the knowledge graph to more accurately reflect detailed product specifications, facilitating precise user queries based on filtering criteria and providing high-quality data support for building smarter applications.

[0090] Step 104: Analyze the validity of historical detection data and the correlation between extreme entities and the target sensor's detection accuracy based on extreme entities, detection mechanisms, and environmental factors affecting detection accuracy. This step is the core of data filtering and relationship modeling in knowledge graph construction. By integrating the extreme entities, detection mechanisms, and environmental factors extracted in the previous stage, it completes the judgment of data validity and the quantification of the correlation between accuracy, laying the foundation for subsequent data expansion and graph integration.

[0091] Whether historical monitoring data can truly reflect the characteristics of target environmental parameters requires validity analysis, specifically whether the data has suffered physical failure or exceeded accuracy limits due to extreme environments. Invalid data includes two types: data that has lost its physical meaning due to mechanistic disruption; and data whose mechanism has not been disrupted but whose error exceeds the limit. A two-layer logic of mechanism threshold verification and error range verification is employed to ensure the rigor of data validity judgment.

[0092] First layer: Has the mechanism failed? Compare the parameter range of the extreme entity with the tolerance threshold of the detection mechanism: if the extreme parameters exceed the threshold, the core conversion law of the sensor is destroyed, and the historical data is completely invalid; if they do not exceed the threshold, proceed to the second layer of verification.

[0093] The second layer: whether the error exceeds the standard. For data where the mechanism has not failed, analyze the parameter range of environmental factors affecting accuracy under special conditions, calculate the total error through the error prediction model, and if the total error is ≤ the sensor's rated accuracy, the data is valid; otherwise, it is marked as partially invalid.

[0094] The principle behind constructing the correlation between extreme entities and detection accuracy: Based on valid data, statistical modeling is used to quantify the mapping relationship between extreme entity parameters, environmental factors, and detection errors. The core is to separate the independent influence and coupling effect of extreme entities and environmental factors. Single-factor analysis: With environmental factors fixed, the correspondence between extreme entity parameters and errors is extracted; multi-factor coupling analysis: The synergistic effect is quantified through a multiple regression model.

[0095] In some specific embodiments, example diagrams for judging the validity of historical detection data are attached. Figure 3 As shown, the specific steps include: extracting the parameter range of extreme entities and the tolerance threshold of the detection mechanism; determining whether the parameters of the extreme entities exceed the tolerance threshold of the detection mechanism; if so, determining that the historical detection data is invalid because the extreme entities hinder the operation of the detection mechanism; if not, analyzing the parameter range of environmental factors affecting detection accuracy under special conditions and predicting the detection error caused by this parameter range; and judging whether each part of the historical detection data is valid based on the detection error. Through two-layer judgment of mechanism threshold verification and environmental factor error prediction, accurate screening of the validity of historical detection data is achieved. First, it is determined whether the data is completely invalid due to mechanism failure; then, it is determined whether it is partially invalid due to excessive error when the mechanism has not failed. Completely invalid data is removed, and partially invalid data is marked and corrected, providing a reliable foundation for subsequent correlation modeling; the extreme environment-accuracy relationship built based on valid data is closer to reality, avoiding correlation bias caused by invalid data.

[0096] First, the extreme entity parameter range and the detection mechanism tolerance threshold are extracted. Both are quantifiable numerical boundaries and serve as the basis for judgment. The extreme entity parameter range is the boundary of specific environmental parameters defined in step 102, reflecting the extreme degree of the environment. The detection mechanism tolerance threshold is the limiting parameter that the core principle of the sensor can withstand, as analyzed in step 103. For example, the pressure tolerance threshold of a piezoelectric sensor is 150 MPa, based on the physical limit of the piezoelectric effect; the radiation tolerance threshold of a semiconductor sensor is 1000 Gy, based on the critical value of carrier ionization damage.

[0097] Next, it is determined whether the parameters of the extreme entity exceed the tolerance threshold. If the upper limit of the extreme entity's parameters is greater than the tolerance threshold of the detection mechanism, the core conversion principle of the sensor is destroyed. For example, if the piezoelectric crystal breaks at 200 MPa, it cannot convert pressure into an electric charge signal. In this case, the historical detection data is irrelevant to the actual environmental parameters and is deemed completely invalid. For example, if the extreme entity is a deep-sea high-pressure entity at 200 MPa, and the target sensor is a piezoelectric pressure sensor with a tolerance threshold of 150 MPa for its detection mechanism, then because 200 MPa > 150 MPa, the physical structure of the sensor's piezoelectric crystal is destroyed, and the detection value of 50 MPa at 200 MPa in the historical data is completely invalid and must be discarded.

[0098] When the mechanism is not in failure, the range of environmental factor parameters is analyzed and the detection error is predicted. When the extreme entity parameters are less than or equal to the tolerance threshold of the detection mechanism, the core mechanism of the sensor is operating normally, but may be affected by other factors in the special environment, leading to an increase in error. The specific parameter ranges of the environmental factors affecting detection accuracy identified in step 103 under special environments are determined. Based on the environmental factor-error correlation model extracted in step 103, the total detection error under special environments is calculated.

[0099] The validity of the data is determined based on the detection error. The predicted total detection error is compared with the sensor's rated accuracy threshold. If the total error is less than or equal to the rated accuracy threshold, the data reflects the actual parameters and is considered valid. If the total error is greater than the rated accuracy threshold, although the data retains physical meaning because the mechanism has not failed, the deviation is too large, and it is considered partially invalid. The error range needs to be marked for subsequent correction.

[0100] In some specific embodiments, after determining the validity of historical detection data, the method further includes: extracting extreme entity parameter values, environmental factor values ​​affecting accuracy, and corresponding accuracy error values ​​from the valid historical detection data; quantifying the correlation between extreme entities and target sensor detection accuracy by constructing a multiple regression model of extreme entity parameters and environmental factor values; and storing the correlation in a knowledge graph in the form of triples of extreme entities, environmental factors, and detection accuracy, with attributes including model parameters, applicable scope, and error contribution rate. After completing the validity determination of historical detection data, the correlation between extreme environment and sensor detection accuracy is transformed into structured knowledge through data extraction, model quantification, and graph storage. Key parameters are extracted from valid data, the correlation strength is quantified through mathematical models, and finally integrated into the knowledge graph in a standardized form.

[0101] First, based on the historical detection data determined to be valid or partially valid in step 104, three sets of core parameters are extracted: extreme entity parameter values ​​(specific quantified values ​​of extreme entities in special environments); environmental factor values ​​affecting accuracy (specific values ​​of interference factors identified in step 103 under special environments); and corresponding accuracy error values ​​(deviations between sensor detection values ​​and actual environmental parameters). In practical applications, only error-corrected data from valid or partially valid data is selected, and abnormal error values ​​caused by mechanism failure are eliminated to ensure the true and reliable correlation between parameters.

[0102] A multiple regression model is constructed to quantify the correlation between multiple independent variables (extreme entity parameters and environmental factors) and a dependent variable (accuracy error). The model analyzes the linear or nonlinear relationships between these variables and a single dependent variable (accuracy error), quantifying the contribution of each factor to accuracy through coefficients. In the multiple regression model, the accuracy error generated during sensor detection is the dependent variable, while extreme entity parameter values ​​and different environmental factor values ​​affecting detection accuracy are the independent variables. The model also includes the inherent error of the sensor itself due to manufacturing processes, assuming ideal, interference-free conditions for both extreme entity parameters and environmental factors. The sensor's accuracy error is jointly influenced by extreme entity parameters and environmental factors, weighted by their respective influence coefficients, and then added to the basic error. This transforms the complex extreme environment-accuracy error correlation into a calculable and quantifiable mathematical relationship, facilitating the analysis of the degree of influence of each factor on accuracy. This information can then be used for knowledge graph construction to accurately infer detection errors under different environments.

[0103] Through multiple regression analysis, the mathematical mapping relationship between extreme entity parameters, environmental factors, and accuracy error is determined, clarifying the independent influence and synergistic effect of each factor. By adjusting the combination of independent variables and testing the goodness of fit, the model is ensured to accurately reflect the actual correlation patterns. The quantified correlations are transformed into triples of extreme entities, environmental factors, and detection accuracy in the knowledge graph, where the subject is the extreme entity, the object is the detection accuracy, and the environmental factor is the correlation bridge. The multiple regression model eliminates the ambiguity of the correlations through mathematical quantification, making the relationships computable; the storage method of triples and attributes conforms to the standardized structure of entity-relationship-attribute in the knowledge graph, ensuring that the correlations can be directly called and reasoned about by the graph.

[0104] In some specific embodiments, the method further includes: parsing the impact path of extreme entities on detection accuracy and extracting path feature parameters, wherein the impact path includes direct impact paths acting on the sensor's sensitive elements and indirect impact paths affecting detection accuracy through transmission links or auxiliary components that interfere with the sensor; constructing a multi-dimensional association dataset based on the path feature parameters, wherein the dataset includes the correspondence between extreme entities and detection accuracy in static data, dynamic data, and interactive data, respectively; combining the path feature parameters and the association dataset, establishing a dynamic association model involving single-path impact and multi-path interactive impact between extreme entity parameters and detection accuracy; instantiating the association relationship between extreme entities and detection accuracy in a knowledge graph, and adding a confidence attribute based on data sample size and model error.

[0105] By distinguishing the types of influencing paths, we can accurately pinpoint the specific links in which extreme entities affect the sensor, providing a mechanistic basis for constructing subsequent correlations. Direct influencing paths: The parameters of the extreme entity directly affect the sensor's sensitive element (core detection component), influencing accuracy by altering its physical or chemical properties. For example, extreme high temperatures (1000℃) act on the metal electrodes of a thermocouple, causing the material resistivity to increase with temperature; strong radiation (1000Gy) acts on the PN junction of a semiconductor sensor, causing abnormal changes in carrier concentration and damaging photoelectric conversion efficiency. Indirect influencing paths: The extreme entity does not directly act on the sensitive element, but affects accuracy by interfering with the transmission link or auxiliary components. For example, a strong electromagnetic environment (100V / m) interferes with the sensor's signal lines, inducing noise voltage in the transmission link, leading to signal distortion; extreme low temperatures (-50℃) cause the operational amplifier offset voltage drift in the sensor's amplification circuit, indirectly amplifying the detection error. In practical applications, the key links in which extreme entities may act can be determined by disassembling the target sensor's structural components, such as the sensitive element, transmission link, signal processing module, and auxiliary components.

[0106] After dividing the path into direct and indirect impact paths, a unique identifier is assigned to each path and characteristic parameters are extracted. Direct paths correspond to the material property coefficients of sensitive components, while indirect paths correspond to the interference propagation coefficient. Characteristic parameters are key indicators for quantifying the intensity of path impact. The parameters of direct paths quantify the response characteristics of sensitive components to extreme environments and are the core link between materials, environment, and accuracy. The parameters of indirect paths characterize the transmission efficiency of interference from the source to the accuracy index, explaining the transmission law of environmental interference → link response → accuracy degradation. In practical applications, these parameters can be obtained through experiments or theoretical derivation, and through standardized tests such as material property testing and link testing, ensuring the objectivity and comparability of the data. Examples of direct path characteristic parameters include: the temperature coefficient of resistance of a thermocouple α = 0.004 / ℃, and the radiation damage coefficient of a semiconductor k = 0.02 / Gy; indirect path characteristic parameters include: the electromagnetic interference coupling coefficient, and the offset voltage drift rate of an operational amplifier at low temperatures.

[0107] The dataset serves as a bridge connecting path characteristics and accuracy impacts, covering complex scenarios in extreme environments through three dimensions: static, dynamic, and interactive. Static data reflects the correspondence between extreme entity parameters and sensor accuracy when they are in a stable state, demonstrating the influence of steady-state extreme environments. This includes the correspondence between static thresholds of extreme entity parameters and static sensor accuracy indicators, such as the correlation between the upper pressure limit of 500 atm for a high-pressure entity and the sensor's zero drift of ±0.1%FS. Dynamic data reflects the time-series correspondence between extreme entity parameters and sensor accuracy as they change over time, demonstrating the influence of transient extreme environments. This includes the time-series correspondence between the dynamic change rate of extreme entity parameters and the sensor's dynamic accuracy indicators, such as the time-series data of a 10 Gy increase in radiation dose per second and a 0.5 ms / Gy delay in sensor response time. Interactive data reflects the synergistic effects generated through path coupling when multiple extreme entities coexist, demonstrating the impact of complex extreme environments. This includes the coupling effects when multiple extreme entities coexist, such as measured data showing that the sensor accuracy attenuation rate under high temperature and high humidity conditions is 1.2 times that under a single extreme environment.

[0108] A dynamic association model is established by combining path feature parameters and associated datasets. The model integrates path features and multi-dimensional data to quantify the dynamic impact of extreme entities on accuracy. It is divided into a single-path impact model and a multi-path interaction impact model.

[0109] The single-path impact model establishes a quantitative relationship between extreme entity parameters and accuracy for a single impact path (direct or indirect). The expression uses a piecewise function to describe the impact of extreme entity parameters changing over time. Specifically, the construction logic involves multiplying the current parameter value of the extreme entity at a given moment and the rate of change of that parameter over time by their respective model coefficients, then directly adding these two products to obtain the sensor's detection accuracy error at that moment. The time-varying model can characterize dynamic disturbances, both changing and invariant, in extreme environments.

[0110] The multi-path interaction impact model addresses multi-path coexistence scenarios by introducing interaction terms to quantify the synergistic effects between paths, including individual impact terms and synergistic impact terms. Individual impact terms are obtained by multiplying the parameters of each of the two extreme entities by their own coefficients, yielding the independent contribution of a single extreme entity to detection accuracy. Synergistic impact terms are obtained by directly multiplying the parameters of the two extreme entities and then multiplying by the interaction coefficient, revealing the additional impact when the two extreme entities act together. The results of the individual impact terms and the synergistic impact terms are then directly added together to obtain the comprehensive impact effect under the combined action of multiple extreme entities. This interaction model can capture the complex synergistic effects where 1+1≠2 in multi-extreme environments.

[0111] Instantiating associations within the knowledge graph and adding a confidence attribute transforms the quantified association model into structured relationships recognizable by the knowledge graph, ensuring the callability and reliability of the associations. These relationships are stored in the knowledge graph as triples of extreme entities, influencing paths, and detection accuracy. The confidence score is used to evaluate the reliability of the associations and is dynamically adjusted based on the data sample size and model error: ≥0.9 when the sample size is ≥1000 sets and the model prediction error is ≤5%; ≤0.6 when the sample size is <300 sets or the error is >10%.

[0112] By employing a logic of path parsing, data construction, model quantization, and graph instantiation, the impact of extreme entities on detection accuracy is transformed from a mechanistic level into a computable and verifiable knowledge graph relationship. This approach retains the physical meaning of the influencing path while ensuring the accuracy and reliability of the association through quantification parameters and confidence levels, providing refined knowledge support for sensor accuracy optimization and fault diagnosis in extreme environments.

[0113] Step 105: Obtain the target environmental parameters of the monitoring area from preset external data sources to obtain predictive detection data. Integrate the predictive detection data with valid historical detection data to obtain expanded data. The external data sources are pre-selected external data sources related to the monitoring area, such as extreme environment simulation platforms, historical extreme event databases, domain expert rule bases, weather forecasts, and data from other sensors. The predictive detection data are estimates of the target environmental parameters derived from the external data sources. The expanded data is a dataset obtained by integrating valid historical data and predictive data. This dataset is processed through data cleaning, spatiotemporal alignment, and weight allocation, addressing the problem of scarce or insufficient sensor data coverage in special environments. It increases the amount and diversity of data, provides richer training data for the knowledge graph, reduces bias in associations caused by insufficient data, and improves the robustness of the graph.

[0114] In extreme environments, historical sensor detection data often has limited samples, focuses on a single scenario, and covers a short timeframe. However, correlation modeling and relation weight calculation require extensive data training. Expanding the data pool provides sufficient samples and improves model robustness. Predictive detection data consists of simulated values ​​for the same or similar monitoring scenarios, filling gaps in historical data. Matching the timestamps of the predicted data with the timelines of historical detection data and aligning the spatial coordinates with the monitoring area expands the sample size of the target sensor's detection data. Furthermore, historical and predicted detection data can be cross-validated to verify the target sensor's detection results and the accuracy of the predicted data.

[0115] In some specific embodiments, the acquisition of predictive detection data includes: if a quantitative relationship exists between the target environmental parameter and other environmental parameters, historical co-occurrence data of the target environmental parameter and related environmental parameters are extracted from external data sources. The quantitative relationship is fitted using multiple linear regression. The measured values ​​of the related environmental parameters in the monitoring area are input, and the predicted values ​​of the target environmental parameters are calculated using the quantitative relationship to obtain the predictive detection data. Specifically, through domain knowledge or data analysis, it is determined whether there is a quantifiable relationship between the target environmental parameter and other parameters. Historical data showing the simultaneous occurrence of the target parameter and related parameters are selected from external data sources to form a sample set. A multiple linear regression model is constructed with the related parameter as the independent variable and the target parameter as the dependent variable. The criteria for determining the quantitative relationship must conform to known scientific laws, such as proving a positive correlation between pressure and temperature, or verifying the statistical correlation between the target parameter and related parameters through the Pearson correlation coefficient (|r|≥0.7) or the Spearman rank correlation test (P<0.05). The expression of the multiple linear regression model is that the target parameter equals the sum of the products of the related parameter and the corresponding regression coefficient. The regression coefficients are used to minimize the sum of squared residuals between the predicted and actual values ​​using the least squares method. After fitting the regression equation, when inputting the measured values ​​of the correlation parameters, it is necessary to first determine whether they are within the parameter range of the historical co-occurrence data. If they are within the range (e.g., ΔP=0.5MPa, historical range 0.1~1MPa), they can be directly substituted into the model calculation; if they are outside the range (e.g., ΔP=1.2MPa), the predicted values ​​need to be adjusted through extrapolation error correction, and the relevant attributes need to be labeled.

[0116] Alternatively, real-time detection data from other types of sensors deployed in the same area as the target sensor can be selected. This data undergoes spatiotemporal alignment and outlier filtering. Predicted values ​​of the target environmental parameters are then derived using a data fusion algorithm to obtain the predicted detection data. This method is suitable for scenarios where other types of sensors are deployed in the same area as the target sensor. Its core principle is to compensate for data gaps in the target sensor through multi-source data fusion. First, sensors whose physical locations overlap with the target sensor are selected, and their detection parameters must be related to the target parameters. Next, data preprocessing is performed, converting the real-time data from other sensors into temporal granularity and spatial coordinates consistent with the target sensor to avoid data misalignment due to time differences or location deviations. Using the Laida criterion or box plot method, abrupt values ​​in the other sensor data are removed, retaining reliable data. Finally, a fusion algorithm suitable for multi-source heterogeneous data, such as weighted averaging, Kalman filtering, or neural networks, is used to map the preprocessed other sensor data to the target parameters.

[0117] These two methods enable reliable predictive detection data to be derived from correlation parameters or data from the same region, even if the target sensor data is missing or distorted. This provides sufficient and high-quality input for subsequent data expansion, ensuring the complete coverage of the knowledge graph in extreme environments.

[0118] Step 106: Based on the expanded data, set entity attributes and relationship weights, and integrate the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association between environmental entities and accuracy decay into a knowledge graph containing extreme entities, sensor entities, and detection performance entities. Entity attributes are parameters describing entity characteristics; relationship weights are numerical values ​​representing the strength of association between entities. Integrated relationship types include: focus direction associations from Step 102, normal environment associations from Step 103, and environment and accuracy decay associations from Step 104. Core entities of the knowledge graph: extreme entities, sensor entities, and detection performance entities. The multi-dimensional knowledge analyzed in the previous steps is fused through a unified graph structure to form a complete knowledge network. The generated knowledge graph can comprehensively depict the working rules of sensors under special environments, supports multi-scenario applications, and ensures that core associations are prioritized for inference through weight settings.

[0119] Sensor entities: such as specific models of temperature sensors, humidity sensors, etc., whose attributes need to be extracted or calculated based on extended data, including: basic attributes: deployment location, installation height, factory parameters, detection range, accuracy class; dynamic attributes: average detection error based on extended data statistics, response delay in extreme environments, and compatibility with other sensors.

[0120] Extreme Entities: The extreme environment entities constructed in step 102 have attributes that need to be refined in conjunction with the extended data, including: quantification threshold (determined based on the critical value in the extended data that causes a sharp drop in sensor accuracy); frequency of occurrence (number of times it has occurred in the monitoring area in the past year); and scope of influence (number of sensors affected by a certain extreme environment, analyzed through the performance of multiple sensors in the extended data).

[0121] Detection performance entity: An entity that characterizes the detection capability of a sensor. Its attributes need to be quantified based on augmented data, including: accuracy retention: the rate of accuracy decay after 30 days of continuous operation under normal conditions; interference resistance: the error increase under extreme conditions.

[0122] The relationships between entities in a knowledge graph need to be quantified by weights, which are calculated based on augmented data. Specifically, the relationship between sensor entities and extreme entities regarding accuracy degradation can be calculated using the error amplification of sensors under extreme environments from the augmented data; the interference relationship between environmental entities and sensor entities can be determined by weights based on the correlation coefficient between dust concentration and sensor error from the augmented data; and the determining relationship between sensor entities and detection performance entities can be set by weights based on the regression coefficients between sensor parameters and detection accuracy from the augmented data.

[0123] The scattered relationship and intensity information from the previous steps is systematically integrated, specifically including the adjusted focus direction of the map in step 102, entities and relationships under normal conditions, and the correlation between environmental entities and accuracy attenuation. The correlation intensity corresponding to the adjusted focus direction of the map in step 102 needs to be quantified and integrated. For example, if the focus direction is the impact of extreme low temperatures on the accuracy of the humidity sensor, the correlation intensity between extreme low temperature entities and humidity sensor entities in the expanded data needs to be converted into weights and labeled as core correlations. Entities and relationships under normal conditions parsed in step 103 need to be matched and integrated with normal scene data in the expanded data. For example, the non-interference relationship between the normal environmental entity 25℃ and the sensor entity PM2.5 sensor, based on the statistical result in the expanded data that the sensor error is always <3% at 25℃, is set to a relationship weight of 0.9 to represent strong stability. The correlation between environmental entities and accuracy attenuation analyzed in step 104 needs to be integrated with the quantification results in the expanded data. For example, the relationship between high-dust environments and the accuracy decay of laser particulate matter sensors is based on the statistics in the expanded data that when the dust concentration is >1000μg / m³, the sensor error increases from 2% to 15%. The relationship weight is set to 0.13, and the nonlinear decay characteristics are marked.

[0124] Through the above integration, the final knowledge graph is presented in a structured form. Extreme entities: These are structured representations of special environments, with attributes directly related to the difference threshold in step 102 and the expanded data in step 105, including basic and derived attributes. Sensor entities: These are the detection subjects in special environments, with attributes integrating the detection mechanism in step 103 and the effectiveness analysis results in step 104, including hardware attributes and environmental adaptation attributes. Detection performance entities: These are the carriers of sensor performance in special environments, with attributes dependent on the correlation in step 104 and the weight settings in step 106, including accuracy and reliability attributes. Accuracy attributes include normal environment error, extreme environment error, and error contribution rate; reliability attributes include mean time between failures (MTBF) and fault warning threshold in extreme environments based on expanded data.

[0125] Furthermore, the structure of the knowledge graph is not a simple accumulation of entities, but rather a dynamic network adapted to specific environments by adjusting multi-dimensional relationships through a focus direction. Based on the graph's focus direction in step 102, core associations are given higher weights, while non-core associations have lower weights. By integrating the normal environment relationships in step 103, the precision decay relationships in step 104, and the dynamic associations in step 106, a three-layer relationship network—the basic layer, the extreme layer, and the interaction layer—is formed. Relying on the expanded data in step 105 and the entity attribute update mechanism in step 106, the graph can dynamically adjust with new data; simultaneously, all relationships are associated with specific detection mechanisms, ensuring that the reasoning results can be traced back to physical principles.

[0126] Compared to knowledge graphs for general environments, the knowledge graph constructed in this application, specifically for special environments, integrates extreme entity parameter ranges, detection mechanism tolerance thresholds, and error contribution rates. This allows for direct inference of sensor states under specific conditions, and the graph can cross-validate information from different sources, enhancing reliability. It retains the extreme parameter characteristics of special environments while ensuring knowledge reliability through mechanism correlation and data verification. Ultimately, it achieves interpretable, predictable, and optimizable sensor behavior under extreme scenarios, providing structured knowledge support for monitoring special environments.

[0127] This application also proposes a sensor knowledge graph generation system, as shown in the attached figure. Figure 4 As shown, the system includes:

[0128] Target determination unit 1 is used to determine the monitoring area and the target sensor located in the area for detecting target environmental parameters;

[0129] Difference analysis unit 2 is used to construct one or more extreme entities based on the environmental characteristics of the special environment of the monitoring area, and adjust the focus direction of the map based on whether the target environmental parameter is an environmental parameter represented by a certain extreme entity;

[0130] The first analysis unit 3 is used to acquire historical detection data and multi-source heterogeneous data of the target sensor, and to analyze the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions.

[0131] The second analysis unit 4 is used to analyze the validity of historical detection data and the correlation between extreme entities and the detection accuracy of target sensors based on extreme entities, detection mechanisms and environmental factors affecting detection accuracy.

[0132] Data expansion unit 5 is used to obtain target environmental parameters of the monitoring area from a preset external data source to obtain predictive detection data, and to integrate the predictive detection data with valid historical detection data to obtain expanded data;

[0133] The graph construction unit 6 is used to set entity attributes and relationship weights based on the expanded data, and integrate the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay into a knowledge graph containing extreme entities, sensor entities and detection performance entities.

[0134] This application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform a sensor knowledge graph generation method. Applying a sensor knowledge graph generation method to a computer program product facilitates execution.

[0135] This application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a sensor knowledge graph generation method as described above.

[0136] The computer storage medium of this application can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. For example, a computer-readable storage medium can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. This application applies a sensor knowledge graph generation method to a computer-readable storage medium storing a computer program. When executed by a processor, this program implements the method steps provided in this application, which is simple, fast, easy to store, and not easily lost.

[0137] Those skilled in the art will understand that the modules described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using computer-executable program code, allowing them to be stored in a storage system for execution by the computing system. Alternatively, they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, this application is not limited to any particular combination of hardware and software.

[0138] Note that the above description is merely a preferred embodiment and the technical principles employed in this application. Those skilled in the art will understand that this application is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of this application. Therefore, although this application has been described in detail through the above embodiments, this application is not limited to the above embodiments. Many other equivalent embodiments may be included without departing from the concept of this application, and the scope of this application is determined by the scope of the appended claims.

[0139] The above disclosures are only a few specific implementation scenarios of this application. However, this application is not limited to these. Any variations that can be conceived by those skilled in the art should fall within the protection scope of this application.

Claims

1. A sensor knowledge graph generation method, characterized in that, include: Determine the monitoring area and the target sensors located within the area for detecting target environmental parameters; Based on the environmental characteristics of the special environment of the monitoring area, one or more extreme entities are constructed, and the focus direction of the spectrum is adjusted based on whether the target environmental parameter is an environmental parameter represented by a certain extreme entity; the special environment refers to an environmental scenario in the monitoring area where one or more environmental parameters deviate significantly from the normal range, and the degree of deviation is sufficient to have a substantial impact on the operating status of the sensor; the extreme entity is a structured knowledge unit constructed based on the characteristics of the special environment, used to represent the core features of the extreme environment; The system acquires historical detection data and multi-source heterogeneous data of the target sensor, and analyzes the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions. The multi-source heterogeneous data is a collection of data from different sources and in different formats related to the target sensor. Based on the aforementioned extreme entities, detection mechanisms, and environmental factors affecting detection accuracy, the validity of the historical detection data and the correlation between extreme entities and the target sensor detection accuracy are analyzed. Predictive detection data is obtained by acquiring target environmental parameters of the monitoring area from a preset external data source, and expanded data is obtained by integrating the predicted detection data with valid historical detection data. Based on the expanded data, entity attributes and relationship weights are set, and the association strength corresponding to the focus direction of the graph, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay are integrated into a knowledge graph containing extreme entities, sensor entities, and detection performance entities; the focus direction of the graph is the priority of the association relationship between entities in the knowledge graph; Specifically, the acquisition of the predicted detection data includes: if the target environmental parameter has a quantitative relationship with other environmental parameters, extracting historical co-occurrence data of the target environmental parameter and related environmental parameters from external data sources, fitting the quantitative relationship through multiple linear regression, inputting the current measured value of the related environmental parameter in the monitoring area, and calculating the predicted value of the target environmental parameter through the quantitative relationship to obtain the predicted detection data; or, selecting real-time detection data from other types of sensors deployed in the same area as the target sensor, performing spatiotemporal alignment and outlier filtering on the detection data, and deriving the predicted value of the target environmental parameter through a data fusion algorithm to obtain the predicted detection data. 2.The method of claim 1, wherein, Adjusting the focus direction of the spectrum specifically includes: If the target environmental parameter is an environmental parameter represented by an extreme entity, then increase the weight between the target environmental parameter and the extreme entity to strengthen the correlation between the two. If the target environmental parameters are different from the environmental parameters represented by any extreme entity, then the interference relationship between each extreme entity and the detection of the target environmental parameters is strengthened. 3.The sensor knowledge graph generation method of claim 1, wherein, Constructing one or more extreme entities specifically includes: Extract the threshold values ​​of differences between special environments and normal environments in physical parameters including temperature, humidity, pressure, and radiation dose; chemical parameters including concentration of corrosive media and redox potential; or spatial parameters including gradient change rate and instantaneous fluctuation amplitude. The attributes of extreme entities are defined based on the difference threshold, including special environment type, parameter range, and duration. 4.The method of claim 1, wherein, The method further includes: Extract the parameter range of the extreme entity and the tolerance threshold of the detection mechanism, and determine whether the parameters of the extreme entity exceed the tolerance threshold of the detection mechanism; If so, the historical detection data is deemed invalid because the extreme entity is obstructing the operation of the detection mechanism. If not, analyze the parameter range of environmental factors affecting detection accuracy under the special environment and predict the detection error caused by the parameter range. Based on the detection error, determine whether each part of the historical detection data is valid. 5.The method of claim 4, wherein, After determining the validity of the historical detection data, the method further includes: Extreme entity parameter values, environmental factor values ​​affecting accuracy, and corresponding accuracy error values ​​are extracted from valid historical detection data. A multiple regression model of extreme entity parameters and environmental factor values ​​is constructed to quantify the correlation between extreme entities and the target sensor detection accuracy. The relationships are stored in the knowledge graph as triples of extreme entities, environmental factors, and detection accuracy, with attributes including model parameters, applicable scope, and error contribution rate. 6.The method of claim 1, wherein, The method further includes: The impact path of extreme entities on detection accuracy is analyzed and path feature parameters are extracted. The impact path includes the direct impact path acting on the sensor's sensitive element and the indirect impact path affecting detection accuracy through the transmission link or auxiliary components that interfere with the sensor. A multi-dimensional association dataset is constructed based on path feature parameters. The dataset includes the correspondence between extreme entities and detection accuracy in static data, dynamic data, and interactive data, respectively. By combining the path feature parameters and associated datasets, a dynamic correlation model is established between extreme entity parameters and detection accuracy, involving both single-path and multi-path interaction effects. Instantiate the association between extreme entities and detection accuracy in the knowledge graph, and add a confidence attribute based on the data sample size and model error.

7. A sensor knowledge graph generation system, characterized by, include: The target determination unit is used to determine the monitoring area and the target sensors located in the area for detecting target environmental parameters. The difference analysis unit is used to construct one or more extreme entities based on the environmental characteristics of the special environment of the monitoring area, and to adjust the focus direction of the spectrum based on whether the target environmental parameter is an environmental parameter represented by a certain extreme entity; the special environment refers to an environmental scenario in the monitoring area where one or more environmental parameters deviate significantly from the normal range, and the degree of deviation is sufficient to have a substantial impact on the operating status of the sensor; the extreme entity is a structured knowledge unit constructed based on the characteristics of the special environment, used to represent the core features of the extreme environment; The first analysis unit is used to acquire historical detection data and multi-source heterogeneous data of the target sensor, and to deduce the sensor's detection mechanism, environmental factors affecting detection accuracy, and their entities and relationships under normal conditions; the multi-source heterogeneous data is a collection of data about the target sensor from different sources and in different formats; The second analysis unit is used to analyze the validity of the historical detection data and the correlation between the extreme entities and the detection accuracy of the target sensor based on the extreme entities, the detection mechanism, and the environmental factors affecting the detection accuracy. The data expansion unit is used to obtain target environmental parameters of the monitoring area from a preset external data source to obtain predictive detection data, and to integrate the predictive detection data with valid historical detection data to obtain expanded data. The graph construction unit is used to set entity attributes and relationship weights based on expanded data, and integrate the association strength corresponding to the graph focus direction, entities and relationships under normal conditions, and the association relationship between environmental entities and accuracy decay into a knowledge graph containing extreme entities, sensor entities, and detection performance entities; the graph focus direction is the priority of the association relationship between entities in the knowledge graph; Specifically, the acquisition of the predicted detection data includes: if the target environmental parameter has a quantitative relationship with other environmental parameters, extracting historical co-occurrence data of the target environmental parameter and related environmental parameters from external data sources, fitting the quantitative relationship through multiple linear regression, inputting the current measured value of the related environmental parameter in the monitoring area, and calculating the predicted value of the target environmental parameter through the quantitative relationship to obtain the predicted detection data; or, selecting real-time detection data from other types of sensors deployed in the same area as the target sensor, performing spatiotemporal alignment and outlier filtering on the detection data, and deriving the predicted value of the target environmental parameter through a data fusion algorithm to obtain the predicted detection data.

8. A computer device, comprising: The computer device includes: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a sensor knowledge graph generation method as described in any one of claims 1-6.

9. A computer program product, characterised in that, It includes executable instructions that, when executed by a processor, implement a sensor knowledge graph generation method as described in any one of claims 1-6.

Images