A Knowledge Graph-Based Method and System for Thermal Signal Analysis and Early Warning

Abstract

This invention relates to the field of thermal signal early warning technology, and discloses a method and system for thermal signal analysis and early warning based on knowledge graphs. The method includes the following steps: S101, collecting thermal signal data; S102, obtaining enthalpy and mass flow rate, and generating energy conservation residuals; S103, evaluating temperature change trends, generating thermal inertia prediction residuals, evaluating channel coupling to obtain cross-channel coefficients, and determining cross-channel coefficient residuals; S104, normalizing the energy conservation residuals, thermal inertia prediction residuals, and cross-channel coefficient residuals respectively, and fusing them to generate anomaly scores; S105, attaching the energy conservation residuals, thermal inertia prediction residuals, cross-channel coefficient residuals, and anomaly scores to a preset knowledge graph to generate fault early warning triples. This invention achieves high-precision signal processing, multi-dimensional fault detection, intelligent reasoning, and accurate early warning for thermal systems.

Description

Technical Field

[0001] This invention belongs to the field of thermal signal early warning technology, specifically relating to a method and system for thermal signal analysis and early warning based on knowledge graphs. Background Technology

[0002] In the operation of thermal systems (such as boilers, steam turbines, and heat exchangers) in energy, chemical, and power industries, real-time monitoring and fault early warning are core requirements for ensuring the safe, stable, and efficient operation of the system. Traditional thermal signal analysis methods mainly rely on single-parameter threshold alarms, which have the following significant drawbacks: 1. Lack of multi-parameter coupling relationships: Parameters such as temperature, pressure, flow rate, and power in thermal systems exhibit strong coupling characteristics. Traditional methods analyze only a single parameter in isolation, failing to capture abnormal correlations between parameters and easily missing early composite faults. 2. Insufficient dynamic characteristic modeling: The lack of quantitative modeling of dynamic behaviors such as system thermal inertia and energy conservation characteristics makes it impossible to effectively predict temperature change trends or assess the degree of energy imbalance, making it difficult to provide early warnings of progressive faults. 3. High subjectivity in threshold setting: Traditional fixed thresholds are not dynamically adjusted based on historical system operating data, failing to adapt to slow time-varying characteristics such as equipment aging and changes in operating conditions. False alarms are often caused by environmental interference or sensor drift, resulting in low reliability of early warnings. 4. Limited fault diagnosis dimensions: The lack of multi-dimensional integrated analysis of the overall system status makes it difficult to distinguish between apparent anomalies and actual faults. Maintenance personnel need to manually check multiple data sources, resulting in low response efficiency. Summary of the Invention

[0003] This invention provides a knowledge graph-based method and system for thermal signal analysis and early warning, which solves the technical problems in related technologies where thermal signal early warning relies only on a single physical quantity or empirical threshold, making it difficult to simultaneously capture energy imbalance, thermal inertia deviation, and channel coupling anomalies.

[0004] This invention provides a knowledge graph-based method for thermal signal analysis and early warning, comprising the following steps:

[0005] S101, Collect thermal signal data of the thermal system within a first preset time period and perform preprocessing. The thermal signal data includes temperature, pressure, flow rate and power. The preprocessing includes denoising and baseline drift correction of the thermal signal data to obtain preprocessed thermal signal data.

[0006] S102, enthalpy and mass flow rate are obtained based on the preprocessed thermal signal data, and then the inlet energy flow and outlet energy flow are determined. At the same time, the deviation of energy conservation is evaluated to form energy conservation residual.

[0007] S103, based on the preprocessed thermal signal data, evaluate the temperature change trend according to the thermal inertia model, generate thermal inertia prediction residuals, and obtain the cross-channel coefficient based on the preprocessed temperature and pressure to evaluate the channel coupling degree, thereby determining the cross-channel coefficient residuals.

[0008] S104, normalize the energy conservation residual, thermal inertia prediction residual and cross-channel coefficient residual respectively, and fuse them according to preset weights to generate anomaly scores;

[0009] S105, the energy conservation residual, thermal inertia prediction residual, cross-channel coefficient residual, and anomaly score are attached as attributes to the corresponding device nodes in the preset knowledge graph, and knowledge reasoning is performed on the attached attributes to generate fault warning triplet, wherein the fault warning triplet includes: device node, warning occurrence, and fault type; and based on the fault warning triplet, a warning signal is output according to the preset warning strategy.

[0010] Furthermore, the denoising employs a sliding window moving average filter;

[0011] In the baseline drift correction, the baseline mean of each parameter is obtained by calculating the mean values ​​of temperature, pressure, flow rate and power within a preset historical healthy period, and the difference between each parameter of the thermal signal data and the baseline mean value of the corresponding parameter is calculated to obtain the corrected parameter.

[0012] Furthermore, the enthalpy and mass flow rate are obtained based on the preprocessed thermal signal data, specifically including:

[0013] The enthalpy value is obtained by multiplying the difference between the pretreated temperature and the preset saturation temperature and the preset pressure specific heat capacity, and then adding the product to the saturation enthalpy corresponding to the preset saturation temperature.

[0014] The fluid density is obtained by using the pre-treated temperature and pressure as query conditions in the preset fluid state table, and then multiplying the fluid density by the pre-treated flow rate to obtain the mass flow rate.

[0015] Furthermore, in multiple inlet channels, the enthalpy value corresponding to each inlet channel is multiplied by the corresponding mass flow rate in sequence and then summed to obtain the inlet energy flow;

[0016] In multiple outlet channels, the enthalpy value corresponding to each outlet channel is multiplied by the corresponding mass flow rate in sequence and then summed to obtain the outlet energy flow.

[0017] Furthermore, the system energy storage change is determined based on the preset system heat capacity and the pre-processed temperature difference between adjacent sampling times. The difference between the inlet energy flow and the outlet energy flow is compared with the system energy storage change, and the absolute difference between the two is used as the energy conservation residual at the current sampling time.

[0018] Furthermore, within adjacent sampling times, based on the preprocessed temperature, inlet energy flow, outlet energy flow, and preset system heat capacity of the previous sampling time, the predicted temperature of the current sampling time is calculated using a thermal inertia model. Then, the predicted temperature is compared with the preprocessed temperature of the current sampling time, and the absolute difference between the two is used as the thermal inertia prediction residual of the current sampling time.

[0019] Furthermore, the determination of the cross-channel coefficient residual includes:

[0020] S201, within a preset sliding window time period, based on the preprocessed temperature and pressure, calculate the Pearson correlation coefficient of the two within the sliding window and use it as the cross-channel coefficient;

[0021] S202, within a preset historical health period, according to a preset sliding window time period, the Pearson correlation coefficients of multiple sliding windows are obtained using the same method as in S201, and the average value is calculated to obtain the reference cross-channel coefficient.

[0022] S203, compare the cross-channel coefficient with the reference cross-channel coefficient, and use the absolute difference between the two as the cross-channel coefficient residual of the sliding window corresponding to the current sampling time.

[0023] Furthermore, the specific steps of S104 include:

[0024] S301, within the preset historical health period, the energy conservation residuals, thermal inertia prediction residuals, and cross-channel coefficient residuals at multiple sampling times are arranged in chronological order to form the energy conservation residual sequence, thermal inertia prediction residual sequence, and cross-channel coefficient residual sequence, respectively.

[0025] S302, statistical processing is performed on the energy conservation residual sequence, thermal inertia prediction residual sequence and cross-channel coefficient residual sequence respectively to obtain the 99th percentile of each sequence, and use it as the health period threshold corresponding to the energy conservation residual, thermal inertia prediction residual and cross-channel coefficient residual;

[0026] S303, the energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual at the sampling time within the first preset time period are respectively compared with the corresponding health period threshold to obtain the normalized energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual;

[0027] S304: The normalized energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual are weighted and accumulated according to the pre-set weights to obtain the anomaly score.

[0028] Furthermore, knowledge reasoning is performed on the mounted attributes to generate fault warning triples, specifically including:

[0029] When the energy conservation residual of a device node is greater than or equal to the preset energy imbalance threshold, and the abnormal scores of the three most recent consecutive sampling times are all greater than or equal to 1, a fault warning triplet with the fault type of energy imbalance fault is generated.

[0030] When the thermal inertia prediction residual of the device node is greater than or equal to the preset thermal inertia threshold, and the anomaly score at the current sampling time is greater than or equal to 1, a fault warning triplet with the fault type of temperature trend deviation fault is generated.

[0031] When the cross-channel coefficient residual of the device node is greater than or equal to the preset coupling threshold, and the abnormal score at the current sampling time is greater than or equal to 1, a fault warning triplet with the fault type of coupling failure is generated.

[0032] When a fault warning triplet with the fault type of energy imbalance fault already exists in the preset knowledge graph, and the device node has a high-pressure state attribute, a fault warning triplet with the fault type of overheating high-pressure fault is generated. Here, high-pressure state means that the pre-processed pressure is greater than or equal to the preset safe pressure threshold.

[0033] This invention provides a knowledge graph-based thermal signal analysis and early warning system, comprising:

[0034] The data acquisition and preprocessing module is used to acquire thermal signal data of the thermal system within a first preset time period and perform preprocessing. The thermal signal data includes temperature, pressure, flow rate and power. The preprocessing includes denoising and baseline drift correction of the thermal signal data to obtain preprocessed thermal signal data.

[0035] The energy conservation deviation assessment module is used to obtain enthalpy and mass flow rate based on preprocessed thermal signal data, thereby determining the inlet energy flow and outlet energy flow, and at the same time assessing the deviation of energy conservation to form energy conservation residual.

[0036] The multi-dimensional residual calculation module is used to evaluate the temperature change trend according to the thermal inertia model based on the preprocessed thermal signal data, generate thermal inertia prediction residuals, and obtain the cross-channel coefficients based on the preprocessed temperature and pressure to evaluate the channel coupling degree, thereby determining the cross-channel coefficient residuals.

[0037] The residual fusion module is used to normalize the energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual respectively, and then fuse them according to preset weights to generate anomaly scores.

[0038] The reasoning and early warning module is used to attach the energy conservation residual, thermal inertia prediction residual, cross-channel coefficient residual, and anomaly score as attributes to the corresponding device nodes in the preset knowledge graph, and perform knowledge reasoning based on the attached attributes to generate fault early warning triples. The fault early warning triples include: device node, early warning occurrence, and fault type. Based on the fault early warning triples, an early warning signal is output according to the preset early warning strategy.

[0039] The beneficial effects of this invention are as follows: It improves signal accuracy by acquiring thermal signals and performing noise reduction and baseline drift correction; it generates energy conservation residuals by calculating enthalpy, mass flow rate, and energy flow, combined with system heat capacity and temperature differences, to comprehensively evaluate energy balance; it generates multi-dimensional residuals using a thermal inertia model and Pearson correlation coefficient, normalizes them with historical data, and weights and fuses them into anomaly scores; it generates fault warning triplets by attaching residual attributes to a knowledge graph and performing multi-condition reasoning, and outputs responses based on a hierarchical warning strategy. This invention achieves high-precision preprocessing of thermal system signals, multi-dimensional fault feature extraction, knowledge-driven intelligent reasoning, and accurate hierarchical warnings, improving the accuracy, comprehensiveness, and timeliness of fault detection, enhancing system safety and operational efficiency, and reducing false alarm rates and potential risks. Attached Figure Description

[0040] Figure 1 This is a flowchart of the thermal signal analysis and early warning method based on knowledge graphs of the present invention. Detailed Implementation

[0041] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0042] It should be noted that, unless otherwise defined, the technical or scientific terms used in one or more embodiments of the present invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in one or more embodiments of the present invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed after the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0043] like Figure 1 As shown, the thermal signal analysis and early warning method based on knowledge graphs includes the following steps:

[0044] S101, Collect thermal signal data of the thermal system within a first preset time period and perform preprocessing. The thermal signal data includes temperature, pressure, flow rate and power. The preprocessing includes denoising and baseline drift correction of the thermal signal data to obtain preprocessed thermal signal data.

[0045] S102, enthalpy and mass flow rate are obtained based on the preprocessed thermal signal data, and then the inlet energy flow and outlet energy flow are determined. At the same time, the deviation of energy conservation is evaluated to form energy conservation residual.

[0046] S103, based on the preprocessed thermal signal data, evaluate the temperature change trend according to the thermal inertia model, generate thermal inertia prediction residuals, and obtain the cross-channel coefficient based on the preprocessed temperature and pressure to evaluate the channel coupling degree, thereby determining the cross-channel coefficient residuals.

[0047] S104, normalize the energy conservation residual, thermal inertia prediction residual and cross-channel coefficient residual respectively, and fuse them according to preset weights to generate anomaly scores;

[0048] S105, the energy conservation residual, thermal inertia prediction residual, cross-channel coefficient residual, and anomaly score are attached as attributes to the corresponding device nodes in the preset knowledge graph, and knowledge reasoning is performed on the attached attributes to generate fault warning triplet, wherein the fault warning triplet includes: device node, warning occurrence, and fault type; and based on the fault warning triplet, a warning signal is output according to the preset warning strategy.

[0049] In one embodiment of the present invention, a resistance temperature detector (RTD) sensor is used to obtain temperature, a diffused silicon pressure transmitter is used to obtain pressure, a vortex flow meter is used to obtain flow rate, and a power transmitter is used to collect voltage and current to calculate power.

[0050] In one embodiment of the present invention, the denoising employs a sliding window moving average filter. Specifically, for each parameter among temperature, pressure, flow rate, and power, a sliding window of fixed length is set on the time series. The arithmetic mean of the signal values ​​within the window is calculated as the denoised value at the center point of the window. The average value within the window is recalculated every time the sliding window moves forward by one sampling point, and the entire time series is iteratively processed in this way. This method can effectively filter out high-frequency random noise and retain the true trend characteristics of the signal.

[0051] In the baseline drift correction, the baseline mean value of each parameter is obtained by calculating the mean values ​​of temperature, pressure, flow rate, and power within a preset historical health period. The difference between each parameter in the thermal signal data and its corresponding baseline mean value is then calculated to obtain the corrected parameter. Specifically, after the initial commissioning of the thermal system, a continuous period of known normal operation of the thermal system is selected as the historical health period, and sample data of each parameter are collected within this period. For each parameter, its arithmetic mean within the historical health period is calculated as the baseline mean value. During real-time monitoring, the parameter value at the current moment is subtracted from the corresponding baseline mean value to obtain the corrected parameter value. This method can effectively eliminate sensor zero-point drift and baseline offset caused by long-term system operation, improving signal accuracy.

[0052] In one embodiment of the present invention, obtaining the enthalpy value and mass flow rate based on preprocessed thermal signal data specifically includes:

[0053] The enthalpy value is obtained by multiplying the difference between the pretreated temperature and the preset saturation temperature by the preset pressure specific heat capacity, and then adding this product to the saturation enthalpy corresponding to the preset saturation temperature. The corresponding saturation temperature is obtained by looking up a preset pressure-saturation temperature correspondence table based on the pretreated pressure. The formula for calculating the enthalpy value is as follows: Where h represents the enthalpy value, Indicates the preset saturation enthalpy. This indicates the preset pressure specific heat capacity, and T represents the temperature after pretreatment. Indicates the preset saturation temperature;

[0054] The fluid density is obtained by using the pre-treated temperature and pressure as query conditions in the preset fluid state table, and the fluid density is multiplied by the pre-treated flow rate to obtain the mass flow rate. The preset fluid state table is pre-constructed based on the thermodynamic properties of the working fluid and can be determined by experimental measurement to ensure that the correspondence between density value and temperature and pressure is accurate and reliable.

[0055] The enthalpy and mass flow rate calculated using the above method accurately reflect the energy state and actual flow rate of the working fluid in the thermal system. The enthalpy calculation combines the saturation characteristics and superheat contribution of the working fluid, avoiding the limitations of single-parameter measurement; the mass flow rate, through density correction compensated for by temperature and pressure, eliminates the influence of medium state changes on flow rate measurement; this process provides crucial foundational data for subsequent calculations of inlet energy flow, outlet energy flow, and energy conservation residuals, ensuring the accuracy of system energy balance analysis and thus improving the reliability and timeliness of fault early warning.

[0056] In one embodiment of the present invention, in multiple inlet channels, the enthalpy value corresponding to each inlet channel is multiplied by the corresponding mass flow rate in sequence and then accumulated to obtain the inlet energy flow, wherein the inlet channel refers to the pipe in the thermal system that transports fluid into the equipment.

[0057] In multiple outlet channels, the enthalpy value corresponding to each outlet channel is multiplied by the corresponding mass flow rate in sequence and then summed to obtain the outlet energy flow. The outlet channel refers to the pipe in the thermal system used to transport fluid from inside the equipment to the outside.

[0058] The inlet energy flow and the outlet energy flow can accurately reflect the energy input-output balance of the thermal system; this process enables the thermal system to detect energy imbalance faults more sensitively, such as leakage and efficiency reduction, and significantly enhances the safety and reliability of the thermal system.

[0059] In one embodiment of the present invention, the change in system energy storage is determined based on the preset system heat capacity and the pre-processed temperature difference between adjacent sampling times. The difference between the inlet energy flow and the outlet energy flow is compared with the change in system energy storage, and the absolute difference between the two is used as the energy conservation residual at the current sampling time.

[0060] The formula for calculating the change in system energy storage is as follows: , This indicates the change in system energy storage, where C represents the preset system heat capacity. This represents the preprocessed temperature difference between adjacent sampling times. and These represent the k-th and (k-1)-th sampling times, respectively, where k represents the sampling time index. Indicates the sampling interval.

[0061] The formula for calculating the energy conservation residual is: ,in, Represents the residual due to energy conservation. and These represent the inlet energy flow and the outlet energy flow, respectively. The energy conservation residual is used to measure the deviation between the actual energy balance of the thermal system and the theoretical value. When there is a leak in the system or the energy conversion efficiency decreases, the difference between the inlet energy flow and the outlet energy flow will deviate significantly from the change in energy storage, resulting in an increase in the residual value. This can effectively reflect the global energy balance state of the thermal system. Compared with the traditional monitoring method that only relies on a single parameter, this process comprehensively considers the dynamic relationship between energy input, output and energy storage, thereby improving the safety and operational stability of the thermal system.

[0062] In one embodiment of the present invention, within an adjacent sampling time, based on the preprocessed temperature of the previous sampling time, the inlet energy flow of the previous sampling time, the outlet energy flow of the previous sampling time, and the preset system heat capacity, the predicted temperature of the current sampling time is calculated by the thermal inertia model. Then, the predicted temperature is compared with the preprocessed temperature of the current sampling time, and the absolute difference between the two is used as the thermal inertia prediction residual of the current sampling time.

[0063] The formula for calculating the predicted temperature is as follows: , This represents the predicted temperature at the k-th sampling time. This represents the preprocessed temperature at the (k-1)th sampling time. and Let represent the inlet energy flow and outlet energy flow at the (k-1)th sampling time, respectively. The predicted temperature at the kth sampling time is compared with the preprocessed temperature, and the absolute value of the difference is calculated as the thermal inertia prediction residual at the kth sampling time. This residual reflects the degree of deviation between the actual temperature and the temperature predicted based on the thermal inertia model. When there are anomalies in the thermal system, such as changes in thermal resistance, leakage, or a decrease in energy conversion efficiency, the actual temperature change will deviate from the model prediction, leading to an increase in this residual.

[0064] By predicting temperature changes and calculating residuals using a thermal inertia model, abnormal changes in the system's thermal dynamic characteristics can be effectively detected. This method utilizes the physical relationship between the system's heat capacity and energy flow to dynamically predict temperature change trends, thereby improving the fault early warning capability and operational reliability of thermal systems.

[0065] In one embodiment of the present invention, the determination of the cross-channel coefficient residual includes:

[0066] S201, within a preset sliding window time period, based on the preprocessed temperature and pressure, calculate the Pearson correlation coefficient between the two within the sliding window and use it as the cross-channel coefficient. The Pearson correlation coefficient can measure the correlation between temperature and pressure. The closer the Pearson correlation coefficient is to 1, the stronger the correlation between temperature and pressure. Using a sliding window can adapt to the slow time-varying characteristics of thermal systems and avoid the interference of data fluctuations at a single moment.

[0067] S202, within a preset historical health period, according to a preset sliding window time period, the Pearson correlation coefficients of multiple sliding windows are obtained using the same method as in S201, and the average value is calculated to obtain the reference cross-channel coefficient. The reference cross-channel coefficient is used to represent the coupling relationship between temperature and pressure when the thermal system is operating normally, providing a benchmark for real-time monitoring.

[0068] S203, compare the cross-channel coefficient with the reference cross-channel coefficient, and use the absolute difference between the two as the cross-channel coefficient residual of the sliding window corresponding to the current sampling time.

[0069] By quantifying the cross-channel coupling between temperature and pressure using the Pearson correlation coefficient and establishing a reference benchmark by combining historical health period data, subtle changes in the operating status of thermal systems can be effectively captured, improving the accuracy of early warnings.

[0070] In one embodiment of the present invention, the specific steps of S104 include:

[0071] S301, within the preset historical health period, the energy conservation residuals, thermal inertia prediction residuals, and cross-channel coefficient residuals at multiple sampling times are arranged in chronological order to form the energy conservation residual sequence, thermal inertia prediction residual sequence, and cross-channel coefficient residual sequence, respectively.

[0072] S302, statistical processing is performed on the energy conservation residual sequence, thermal inertia prediction residual sequence, and cross-channel coefficient residual sequence respectively to obtain the 99th percentile of each sequence, which is used as the health period threshold corresponding to the energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual; specifically, the values ​​in each sequence are sorted from smallest to largest, and the value located at the 99th percentile is taken as the health period threshold of the corresponding residual;

[0073] S303, the energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual of the sampling time within the first preset time period are respectively compared with the corresponding health period threshold to obtain the normalized energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual; by uniformly mapping the residuals of different dimensions to the range of 0 to 1, the influence of dimensional differences on the fusion calculation is eliminated.

[0074] S304: The normalized energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual are weighted and accumulated according to the pre-set weights to obtain the anomaly score; among them, the energy conservation residual is given a higher weight.

[0075] By dynamically determining the residual threshold using historical health period data, the subjectivity of traditional fixed thresholds is avoided, making the threshold more closely match the actual operating characteristics of the system. Normalization processing eliminates dimensional differences and ensures the comparability of multi-dimensional residuals. The weighted fusion algorithm can highlight the role of key monitoring indicators, such as energy conservation residuals, so that the anomaly scores can comprehensively reflect the overall status.

[0076] In one embodiment of the present invention, the device nodes in the preset knowledge graph correspond to various devices or components in the actual thermal system and are the main objects of fault occurrence. Each device node is equipped with multiple attributes, such as operating parameters like pressure and temperature, as well as device status information. These attributes are important bases for determining whether the device is operating normally.

[0077] Fault types are classifications of possible equipment faults in the knowledge graph, such as energy imbalance faults, temperature trend deviation faults, coupling failure faults, and overheating / high-pressure faults. The knowledge graph clarifies the relationships between different fault types and equipment nodes and their associated attributes.

[0078] Based on the device node attachment attributes in a pre-defined knowledge graph, different types of fault warning triples are generated through multi-condition logical judgments. Each triple consists of a device node, a warning occurrence, and a fault type, used to accurately describe the fault subject, whether a warning has been issued, and the nature of the fault. The specific reasoning rules are as follows:

[0079] When the energy conservation residual of a device node is greater than or equal to the preset energy imbalance threshold, and the abnormal scores of the three most recent consecutive sampling times are all greater than or equal to 1, a fault warning triplet with the fault type of energy imbalance fault is generated.

[0080] When the thermal inertia prediction residual of the device node is greater than or equal to the preset thermal inertia threshold, and the anomaly score at the current sampling time is greater than or equal to 1, a fault warning triplet with the fault type of temperature trend deviation fault is generated.

[0081] When the cross-channel coefficient residual of the device node is greater than or equal to the preset coupling threshold, and the abnormal score at the current sampling time is greater than or equal to 1, a fault warning triplet with the fault type of coupling failure is generated.

[0082] When a fault warning triplet with the fault type of energy imbalance fault already exists in the preset knowledge graph, and the device node has a high-pressure state attribute, a fault warning triplet with the fault type of overheating high-pressure fault is generated. Here, high-pressure state means that the pre-processed pressure is greater than or equal to the preset safe pressure threshold.

[0083] In one embodiment of the present invention, the preset early warning strategy includes:

[0084] When a single fault warning triplet is generated, a yellow indicator will flash at the corresponding location on the local monitoring interface.

[0085] If the same fault warning triplet is generated repeatedly within three consecutive sampling times, the maintenance personnel will be notified via SMS or email.

[0086] When a fault warning ternary group for the overheating and high-pressure fault type is generated, a red emergency shutdown alarm is triggered and the on-site personnel are alerted via an audible and visual alarm.

[0087] The aforementioned early warning strategy, through a tiered response mechanism, enables accurate delivery and efficient processing of alarm information, thereby improving operational efficiency and system reliability while reducing labor costs and potential economic losses.

[0088] Embodiments of the present invention also provide a knowledge graph-based thermal signal analysis and early warning system, comprising:

[0089] The data acquisition and preprocessing module is used to acquire thermal signal data of the thermal system within a first preset time period and perform preprocessing. The thermal signal data includes temperature, pressure, flow rate and power. The preprocessing includes denoising and baseline drift correction of the thermal signal data to obtain preprocessed thermal signal data.

[0090] The energy conservation deviation assessment module is used to obtain enthalpy and mass flow rate based on preprocessed thermal signal data, thereby determining the inlet energy flow and outlet energy flow, and at the same time assessing the deviation of energy conservation to form energy conservation residual.

[0091] The multi-dimensional residual calculation module is used to evaluate the temperature change trend according to the thermal inertia model based on the preprocessed thermal signal data, generate thermal inertia prediction residuals, and obtain the cross-channel coefficients based on the preprocessed temperature and pressure to evaluate the channel coupling degree, thereby determining the cross-channel coefficient residuals.

[0092] The residual fusion module is used to normalize the energy conservation residual, thermal inertia prediction residual, and cross-channel coefficient residual respectively, and then fuse them according to preset weights to generate anomaly scores.

[0093] The reasoning and early warning module is used to attach the energy conservation residual, thermal inertia prediction residual, cross-channel coefficient residual, and anomaly score as attributes to the corresponding device nodes in the preset knowledge graph, and perform knowledge reasoning based on the attached attributes to generate fault early warning triples. The fault early warning triples include: device node, early warning occurrence, and fault type. Based on the fault early warning triples, an early warning signal is output according to the preset early warning strategy.

[0094] It should be noted that the interval and threshold sizes are set for ease of comparison. The size of the threshold depends on the amount of sample data and the base number set by those skilled in the art for each set of sample data, as long as it does not affect the proportional relationship between the parameter and the quantized value. Furthermore, the above formulas are all dimensionless calculations, and the formulas are derived from software simulations using a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0095] The embodiments of the present invention have been described above, but the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of the present embodiments, all of which are within the protection scope of the present embodiments.

Claims

1. A method for thermal signal analysis and early warning based on a knowledge graph, characterized in that, The method comprises the following steps: S101, collecting thermal signal data in a first preset time period in a thermal system, and performing preprocessing, wherein the thermal signal data comprises temperature, pressure, flow and power, and the preprocessing comprises denoising and baseline drift correction of the thermal signal data to obtain preprocessed thermal signal data; S102, obtaining enthalpy and mass flow according to the preprocessed thermal signal data, and then determining inlet energy flow and outlet energy flow, and evaluating the deviation of energy conservation to form an energy conservation residual; S103, evaluating the temperature change trend according to the preprocessed thermal signal data according to a thermal inertia model, and generating a thermal inertia prediction residual, and evaluating the channel coupling degree according to the preprocessed temperature and pressure to obtain a cross-channel coefficient, and then determining a cross-channel coefficient residual; wherein, in adjacent sampling time, based on the preprocessed temperature of the previous sampling time, the inlet energy flow of the previous sampling time, the outlet energy flow of the previous sampling time and the preset system heat capacity, the predicted temperature of the current sampling time is calculated through the thermal inertia model, then the predicted temperature is compared with the preprocessed temperature of the current sampling time, and the absolute difference between the two is taken as the thermal inertia prediction residual of the current sampling time; The calculation formula of the predicted temperature is: , represents the predicted temperature at the kth sampling moment, represents the preprocessed temperature at the k-1th sampling moment, C represents a preset system heat capacity, represents a sampling interval, and respectively represent the inlet energy flow and the outlet energy flow at the k-1th sampling moment. S104, normalizing the energy conservation residual, the thermal inertia prediction residual and the cross-channel coefficient residual respectively, and fusing to generate an abnormal score according to a preset weight; S105, mounting the energy conservation residual, the thermal inertia prediction residual, the cross-channel coefficient residual and the abnormal score to the corresponding device node in the preset knowledge graph as attributes, and performing knowledge reasoning on the mounted attributes to generate a fault warning triple, wherein the fault warning triple comprises a device node, a warning occurrence and a fault type; and outputting a warning signal according to a preset warning strategy based on the fault warning triple. 2.The knowledge graph-based thermal signal analysis and early warning method according to claim 1, characterized in that, The denoising adopts a sliding window moving average filter; In the baseline drift correction, the baseline mean of each parameter is obtained by calculating the mean of the temperature, pressure, flow and power in a preset historical healthy period, and the difference between the thermal signal data parameters and the baseline mean of the corresponding parameters is calculated to obtain the corrected parameters. 3.The knowledge graph-based thermal signal analysis and early warning method according to claim 2, characterized in that, Obtaining enthalpy and mass flow according to the preprocessed thermal signal data specifically comprises: Multiplying the difference between the preprocessed temperature and the preset saturation temperature by the preset specific heat capacity, and then adding the product to the saturated enthalpy corresponding to the preset saturation temperature to obtain the enthalpy value; In the preset fluid state table, the fluid density is obtained by taking the preprocessed temperature and pressure as the query condition, and the mass flow is obtained by multiplying the fluid density by the preprocessed flow. 4.The knowledge graph-based thermal signal analysis and early warning method according to claim 1, characterized in that, In a plurality of inlet channels, the enthalpy value corresponding to each inlet channel is multiplied by the corresponding mass flow in sequence and accumulated to obtain the inlet energy flow; In a plurality of outlet channels, the enthalpy value corresponding to each outlet channel is multiplied by the corresponding mass flow in sequence and accumulated to obtain the outlet energy flow.

5. The knowledge graph-based thermal signal analysis and early warning method according to claim 4, characterized in that, The system energy storage change is determined according to the difference between the preprocessed temperatures of adjacent sampling times and the preset system heat capacity, the difference between the inlet energy flow and the outlet energy flow is compared with the system energy storage change, and the absolute difference between the two is taken as the energy conservation residual of the current sampling time. 6.The knowledge graph-based thermal signal analysis and early warning method according to claim 2, characterized in that, The determination of the cross-channel coefficient residual includes: S201, in a preset sliding window time period, based on the pre-processed temperature and pressure, calculating the Pearson correlation coefficient of the two in the sliding window as the cross-channel coefficient; S202, in a preset historical health period, according to the preset sliding window time period, the same method as S201 is used to obtain the Pearson correlation coefficients of multiple sliding windows, and the mean value is obtained as the reference cross-channel coefficient; S203, compare the cross-channel coefficient with the reference cross-channel coefficient, and take the absolute difference value as the cross-channel coefficient residual of the sliding window corresponding to the current sampling time. 7.The knowledge graph-based thermal signal analysis and early warning method according to claim 2, characterized in that, The specific steps of S104 include: S301, in a preset historical health period, the energy conservation residual, the thermal inertia prediction residual and the cross-channel coefficient residual of multiple sampling times are respectively composed into an energy conservation residual sequence, a thermal inertia prediction residual sequence and a cross-channel coefficient residual sequence in time sequence; S302, respectively, the energy conservation residual sequence, the thermal inertia prediction residual sequence and the cross-channel coefficient residual sequence are statistically processed to obtain the 99th percentile of each sequence, and the 99th percentile is taken as the health period threshold value corresponding to the energy conservation residual, the thermal inertia prediction residual and the cross-channel coefficient residual; wherein the 99th percentile refers to the value at the 99th position after ascending sorting the values in the sequence; S303, the energy conservation residual, the thermal inertia prediction residual and the cross-channel coefficient residual of the sampling time in the first preset time period are respectively subjected to ratio operation with the corresponding health period threshold value to obtain the normalized energy conservation residual, thermal inertia prediction residual and cross-channel coefficient residual; S304, the normalized energy conservation residual, thermal inertia prediction residual and cross-channel coefficient residual are weighted and accumulated according to the pre-set weight to obtain the abnormal score. 8.The knowledge graph-based thermal signal analysis and early warning method according to claim 1, characterized in that, The knowledge reasoning is performed on the mounting attribute to generate a fault warning triple, specifically including: When the energy conservation residual of the device node is greater than or equal to the preset energy imbalance threshold, and the abnormal scores of the last three consecutive sampling times are all greater than or equal to 1, a fault warning triple with the fault type of energy imbalance fault is generated; When the thermal inertia prediction residual of the device node is greater than or equal to the preset thermal inertia threshold, and the abnormal score of the current sampling time is greater than or equal to 1, a fault warning triple with the fault type of temperature trend deviation fault is generated; When the cross-channel coefficient residual of the device node is greater than or equal to the preset coupling threshold, and the abnormal score of the current sampling time is greater than or equal to 1, a fault warning triple with the fault type of coupling failure fault is generated; When there is already a fault warning triple with the fault type of energy imbalance fault in the preset knowledge graph, and the device node has a high pressure state attribute, a fault warning triple with the fault type of overheating high pressure fault is generated, wherein the high pressure state refers to that the pre-processed pressure is greater than or equal to the preset safety pressure threshold.

9. A thermal signal analysis and early warning system based on a knowledge graph, characterized in that, The knowledge graph-based thermal signal analysis and warning method according to any one of claims 1-8, comprising: The data acquisition and preprocessing module is configured to acquire thermal signal data in the thermal system within a first preset time period and perform preprocessing, wherein the thermal signal data includes temperature, pressure, flow rate and power, and the preprocessing includes denoising and baseline drift correction on the thermal signal data to obtain preprocessed thermal signal data. The energy conservation deviation evaluation module is configured to obtain enthalpy and mass flow rate based on the preprocessed thermal signal data, determine inlet energy flow and outlet energy flow, and evaluate deviation of energy conservation to form an energy conservation residual. The multi-dimensional residual calculation module is configured to evaluate temperature variation trend according to a thermal inertia model based on the preprocessed thermal signal data, generate a thermal inertia prediction residual, obtain a cross-channel coefficient based on a channel coupling degree evaluated by preprocessed temperature and pressure, and determine a cross-channel coefficient residual. The residual fusion module is configured to normalize the energy conservation residual, the thermal inertia prediction residual and the cross-channel coefficient residual respectively, and fuse them according to a preset weight to generate an anomaly score. The reasoning and early warning module is configured to mount the energy conservation residual, the thermal inertia prediction residual, the cross-channel coefficient residual and the anomaly score as attributes to a corresponding device node in a preset knowledge graph, perform knowledge reasoning on the mounted attributes to generate a fault early warning triple, wherein the fault early warning triple includes a device node, an early warning occurrence and a fault type, and output an early warning signal according to a preset early warning strategy based on the fault early warning triple.

Images