Intelligent heat meter remote data acquisition method based on edge calculation

Abstract

The invention relates to the technical field of heat supply system remote monitoring, and discloses an intelligent heat meter remote data acquisition method based on edge calculation. The method comprises the following four steps: constructing a user-device double-causal network model, realizing an anomaly detection and attribution mechanism based on causal reasoning, developing an interpretable anomaly report generation system and realizing a predictive maintenance suggestion system. By constructing the double-causal network reflecting the causal relationship between the user behavior and the system physical state, accurate attribution of abnormal data is realized, user behavior changes and equipment faults can be accurately distinguished, the diagnosis accuracy is improved, the false alarm rate is reduced, and the user experience is improved. The system is particularly suitable for remote monitoring and maintenance management of heat supply systems of large residential districts, commercial buildings and industrial districts and remote or difficult-to-reach heat supply stations.

Description

Technical Field

[0001] This invention relates to the field of remote monitoring technology for heating systems, and more specifically, to a remote data acquisition method for smart heat meters based on edge computing. Background Technology

[0002] With the widespread application of smart heat meters in centralized heating systems, remote data acquisition and anomaly monitoring technology for heat meters has become an important component of smart heating network management. Existing remote monitoring technologies for heat meters mainly include simple threshold detection and anomaly identification based on statistical correlation.

[0003] These technologies have the following problems in practical applications: First, traditional heat meter monitoring systems struggle to accurately distinguish between data anomalies caused by changes in user behavior (such as a sudden decrease in heat consumption due to long-term business trips) and equipment malfunctions (such as sensor damage). They often misjudge normal user behavior changes as equipment malfunctions, or vice versa. Second, existing heat meter fault diagnosis relies primarily on simple threshold detection or statistical correlation-based methods, failing to accurately trace the causes of faults in complex scenarios. This is especially true when multiple fault factors coexist or interact, significantly reducing diagnostic accuracy. Third, existing anomaly detection systems are mostly "black box" models, unable to provide understandable interpretations of detection results. This makes it difficult for maintenance personnel to assess the credibility and urgency of anomaly alarms, affecting the accuracy of maintenance decisions. Furthermore, the inability to accurately distinguish anomaly types often leads to excessive alarms, requiring frequent deployments of maintenance personnel and wasting significant human and time resources on handling false alarms. Finally, traditional methods typically require transmitting all raw data collected by the heat meter to the cloud for processing, increasing network transmission burden and causing system response delays, failing to meet the needs of real-time anomaly detection. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a remote data acquisition method for smart heat meters based on edge computing, comprising the following steps: Construct a user-device dual-causal network model, which includes a set of user behavior variables, a set of system state variables, and the causal relationships between them; An anomaly attribution algorithm based on counterfactual reasoning is used to analyze real-time data from heat meters, and anomaly detection results and explanations of their causes are output. Transform attribution results into human-understandable explanatory reports, outputting anomaly reports that include diagnostic results, explanations, and recommendations; Utilize predictive maintenance algorithms to analyze historical anomaly data and equipment status, and output predictive maintenance recommendations; The anomaly attribution algorithm calculates: Determine the most likely cause of the abnormal event. Indicates the observed anomalous events, Represents the set of possible causal variables. This indicates a causal intervention operation. Indicating intervention Observed under the conditions The probability, Indicates the conciseness of the explanation.

[0005] Preferably, the step of constructing the user-device dual causal network model includes: Define a dual-causal network structure; Extract user behavior variables and build a user behavior model; Extract the system's physical state variables and establish a system physical model; We use causal discovery algorithms to learn causal relationships between variables from historical data and construct a complete bicausal network.

[0006] Preferably, the dual-causal network structure is as follows: ,in Represents a set of user behavior variables. Represents the set of system state variables. This indicates the internal causal relationship between user behavior variables. This represents the internal causal relationship between system state variables. This represents the interactive causal relationship between user behavior variables and system state variables.

[0007] Preferably, the step of analyzing real-time data from the calorie meter using an anomaly attribution algorithm based on counterfactual reasoning includes: Build an abnormal pattern feature library; Anomaly detection is performed on the real-time data collected by the heat meter; Generate causal attributions and explanations for detected abnormal events; Counterfactual reasoning was used to verify the reliability of the attribution results.

[0008] Preferably, the counterfactual verification step includes: Constructing a counterfactual scenario: "If the cause" Not present, abnormal Will it still happen? Calculating counterfactual conditional probabilities using a bicausal network model ; Calculate the causal effect strength; when the causal effect strength exceeds a preset threshold, confirm. for The real reason.

[0009] Preferably, the step of converting the attribution results into a human-understandable explanatory report includes: Classify and assess the severity of attribution results; Transform the results of technology attribution into natural language descriptions; Generate targeted maintenance recommendations based on attribution results; Choose an appropriate report delivery strategy based on the type and severity of the anomaly.

[0010] Preferably, the step of analyzing historical anomaly data and equipment status using predictive maintenance algorithms includes: Assess the current health status of the device based on a dual-causal network model; Predict future failure risks by using historical failure data and current equipment status; Based on the failure risk prediction results, an optimized maintenance plan is generated.

[0011] Preferably, the step of assessing the health status of the equipment employs a health status assessment function. ,in express The state vector of the device at any given time. Indicates the first The state variables are in The value at time, This represents the normal values ​​or distribution of the variable. The function calculates the degree of deviation between the current value and the normal value. These are the weighting coefficients of each variable.

[0012] Preferably, the method is executed on the edge computing unit of the heat meter, and only the processed anomaly detection results, interpretation reports and maintenance suggestions are transmitted to the cloud server, reducing the data transmission burden.

[0013] A remote data acquisition system for smart heat meters based on edge computing, used to execute the above method, includes: The User-Device Dual Causal Network Model Construction Module is used to construct a dual causal network model that includes a set of user behavior variables, a set of system state variables, and their causal relationships. The anomaly detection and attribution module is used to analyze real-time data from the heat meter using an anomaly attribution algorithm based on counterfactual reasoning, and output anomaly detection results and explanations of their causes. An interpretable anomaly report generation module is used to transform attribution results into human-understandable explanatory reports; The predictive maintenance recommendation module is used to analyze historical anomaly data and equipment status using predictive maintenance algorithms and output predictive maintenance recommendations. The edge computing processing unit is used to execute the functions of the above modules locally on the heat meter, reducing the data transmission burden.

[0014] The beneficial effects of this invention are as follows: Significantly improved anomaly detection accuracy: By distinguishing between abnormal user behavior and equipment malfunction, the false alarm rate is reduced by 85%, and the diagnostic accuracy is increased to over 92%.

[0015] Provide interpretable diagnostic results: The system can provide human-understandable explanations of anomalies, enabling maintenance personnel to clearly understand the causes of anomalies and improve processing efficiency.

[0016] Reduce maintenance costs: By accurately identifying the type of anomaly, unnecessary on-site maintenance is reduced, and the efficiency of maintenance human resources is improved by 60%.

[0017] Reduce data transmission burden: Utilize edge computing capabilities to complete most analysis tasks locally, transmitting only the processed results, reducing data transmission volume by 95%.

[0018] Supports predictive maintenance: The system can predict potential failure risks, enabling a shift from reactive response to proactive prevention, reducing unexpected downtime by up to 70%.

[0019] Knowledge accumulation and self-evolution: The system has the ability to learn from historical cases, and its diagnostic accuracy and efficiency will continue to improve as it is used over time.

[0020] Furthermore, the method provided in this application has good scalability and adaptability, and can be flexibly configured and optimized according to the needs of different application scenarios. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a bar chart comparing the accuracy of anomaly detection. Figure 3 It is a line graph showing the effect of reducing the false alarm rate; Figure 4 It is a pie chart showing improvements in resource efficiency. Figure 5 It is an area graph showing the effect of reducing data transmission burden. Detailed Implementation

[0022] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, some features described in the examples may be combined in other examples.

[0023] Before providing a detailed description of the present invention, the terms used in this application will be explained below to aid in understanding the technical solutions of the present invention: Bicausal network: refers to a causal graph model that jointly models user behavior variables and system physical state variables, used to describe the causal relationships between variables and their mutual influence.

[0024] Causal relationship: refers to the cause-and-effect relationship between events, where one event (cause) leads to the occurrence of another event (effect). In this application, it specifically refers to the relationship between abnormal heat meter data and its potential causes.

[0025] Counterfactual reasoning: refers to the reasoning process that uses hypothetical questions such as "If X had not happened, what would Y have been like?" to verify and confirm the existence and strength of causal relationships.

[0026] Edge computing refers to a computing model that performs data processing and analysis at the network edge, close to the data source, reducing data transmission volume and improving response speed.

[0027] Anomaly attribution: refers to the process of tracing observed abnormal phenomena back to their root causes and identifying the specific factors or events that led to the anomaly.

[0028] Existing technical problems The technical solution of this invention is mainly applied to remote data acquisition and fault diagnosis scenarios of smart heat meters, and is particularly suitable for the following application environments: Central heating systems in large residential communities are equipped with a large number of smart heat meters, which require real-time monitoring and anomaly detection of each household's heating usage.

[0029] Zoned heating systems in commercial or office buildings need to accurately distinguish between changes in heating patterns caused by shifts in office hours and equipment malfunctions.

[0030] Industrial heating systems, where heat meters operate in complex environments and experience diverse types of malfunctions, require high-precision anomaly detection and diagnosis systems.

[0031] For heating stations that are remote or difficult to reach, it is necessary to reduce unnecessary on-site maintenance and improve maintenance efficiency and accuracy.

[0032] In these scenarios, heat meters are typically equipped with processing units that have edge computing capabilities, enabling them to perform preliminary data processing and analysis locally. They also have a certain storage capacity to save historical data and models, and can exchange necessary data with a central server via a wireless network.

[0033] Example This embodiment provides a remote data acquisition method for smart heat meters based on edge computing, which mainly includes the following steps: Step 1: Construct a user-device dual-causal network model According to an embodiment of this application, this step utilizes a user-device dual-causal network model construction algorithm to process multi-source data collected by the heat meter, generating a structured network model describing the causal relationship between user behavior and system state. Specifically, it includes the following sub-steps: Step 1.1: Define the dual-causal network structure First, we define the basic structure of a two-causal network, namely: in, This represents a set of user behavior variables, including variables such as heating time, heating intensity, and heating patterns. It represents the set of system state variables, including variables such as flow rate, temperature, pressure, and power; This indicates the internal causal relationship between user behavior variables; This represents the internal causal relationship between system state variables; This represents the interactive causal relationship between user behavior variables and system state variables.

[0034] The actual structure of a dual-causal network model can be represented as a directed acyclic graph, where nodes represent variables and edges represent causal relationships. In heat meter data analysis, this model has a two-layer structure: the upper layer is the user behavior layer, and the lower layer is the system physical layer, with interactive connections between the two layers. For example, in a residential heating scenario, the user behavior layer may include nodes such as "wake-up time," "departure time," and "return time," while the system physical layer may include nodes such as "supply water temperature," "return water temperature," and "flow rate." The "user valve adjustment behavior" node connects the two layers, influencing multiple variables in the system physical layer.

[0035] Optionally, in some embodiments, the dual-causal network structure can be further refined into multiple functional subnetworks, such as temporal dependency subnetworks, spatial correlation subnetworks, and environmental influence subnetworks, to more accurately capture different types of causal relationships.

[0036] Step 1.2: User behavior variable extraction and modeling First, the historical heat meter data is preprocessed, including: (1) Handling missing values: For short-term data missing, linear interpolation is used to fill in the missing values; for long-term missing values, historical data from the same period are used to replace them. (2) Outlier removal: Use the sliding window midpoint filtering method to remove obvious outliers; (3) Data standardization: Transform data of different dimensions to a uniform scale and standardize them using z-score or min-max methods; (4) Time synchronization: Ensure that the timestamps of data from different sensors are aligned and achieve a unified time series sampling frequency through linear interpolation.

[0037] Then, time series analysis was performed on the processed historical heat meter data to extract user heat consumption behavior characteristics, including: Temporal characteristics: distribution of heat usage time points, duration, and periodic patterns; Intensity dimension characteristics: including heat magnitude, fluctuation range, and peak characteristics; Correlation dimension characteristics: the correlation between thermal patterns and external factors (such as weather and temperature); Then, cluster analysis is performed on the extracted features to identify typical user heating patterns, establish a user behavior model, and define a set of user behavior variable nodes. .

[0038] In some implementations, the system can employ adaptive learning methods to continuously update the user behavior model. For example, an online incremental clustering algorithm can be used to automatically adjust the cluster centers and number as new data accumulates, enabling the model to adapt to seasonal changes or long-term evolution of user behavior.

[0039] Step 1.3: Extraction and Modeling of System Physical State Variables Collect various physical state data of the heat meter system, including: Fluid characteristics: flow rate, velocity, direction, etc. Thermal characteristics: supply and return water temperature, temperature difference, heat capacity, etc.; Equipment characteristics: valve status, sensor readings, battery level, etc.; Based on thermodynamic principles and equipment operating principles, a physical model of the system is established, and a set of system state variable nodes is defined. .

[0040] Step 1.4: Causal Relationship Learning and Network Construction Before performing causal relationship learning, data preprocessing is first performed on different types of variables: Standardization of numerical variables: For continuous variables with different dimensions such as flow rate, temperature, and pressure, the z-score standardization method (i.e., subtracting the mean and dividing by the standard deviation) is used to make the mean 0 and the standard deviation 1, thus eliminating the influence of dimensions.

[0041] Categorical variable coding: Categorical variables such as valve status (open / closed) and operating mode are converted into numerical representations using one-hot coding.

[0042] Time feature conversion: Converting point-in-time information into periodic features, such as converting hours into continuous features using sine and cosine functions to preserve the periodicity of time.

[0043] Then, causal discovery algorithms are used to learn the causal relationships between variables from the preprocessed historical data, constructing a complete bicausal network. This mainly includes the following methods: Conditional independence test: Determine the conditional independence relationship between variables through statistical tests; Intervention data analysis: Identifying causal directions using system intervention records (such as maintenance and debugging); Causal discovery in time series: analyzing causal relationships in time series data based on Granger causality test; Expert knowledge integration: Integrating prior causal knowledge provided by domain experts to improve the network structure; Ultimately, a complete bicausal network model is formed, incorporating both user behavior and system state. .

[0044] Step 2: Implement an anomaly detection and attribution mechanism based on causal reasoning According to an embodiment of this application, this step utilizes an anomaly attribution algorithm based on counterfactual reasoning to analyze real-time data from the heat meter and outputs anomaly detection results and explanations of their causes. Specifically, it includes the following sub-steps: Step 2.1: Construction of the Anomaly Pattern Feature Library Collect and classify common anomaly patterns in calorie meters, and establish an anomaly feature database, including: Abnormal user behavior: such as sudden heating demand, long-term lack of heating usage, etc. Equipment malfunctions: such as sensor drift, valve jamming, communication interruption, etc. Abnormalities in the heating system: such as pipe leaks, heat source failures, water quality problems, etc. For each type of anomaly, a feature description and feature vector are established as a reference benchmark for anomaly detection.

[0045] Step 2.2: Real-time Data Anomaly Detection Preprocessing of real-time data collected by the heat meter specifically includes: (1) Signal smoothing: The sensor data noise is smoothed by using an exponentially weighted moving average method with a small window (such as 3-5 time points); (2) Feature calculation: Extract time-domain features (mean, standard deviation, kurtosis, skewness, etc.) and frequency-domain features (main frequency components and their energy) from the original signal; (3) Feature normalization: The extracted feature vectors are transformed to a uniform scale through a real-time adaptive normalization method to adapt to the dynamic changes in data distribution; (4) Dimension reduction: Use principal component analysis or autoencoder methods to reduce feature dimensions and improve anomaly detection efficiency.

[0046] Then, feature vectors are extracted from the preprocessed data, and anomaly detection is performed using the following methods: Deviation detection based on user behavior models: comparing the differences between current heat usage patterns and historical user behavior models; Constraint violation detection based on physical laws: whether the detection data violates the fundamental laws of thermodynamics and the operating principles of the equipment; Conditional probabilistic anomaly detection based on user-device dual causal network: evaluating the conditional probability of observed data under the user-device dual causal network model; When an anomaly is detected, an anomaly event A_{event} is generated, and the anomaly attribution step is initiated.

[0047] Step 2.3: Causal Attribution and Explanation Generation Before performing causal attribution, the abnormal events and possible causal variables are first preprocessed: Abnormal event feature standardization: Convert different types of abnormal events (such as high temperature, flow fluctuations, etc.) into a uniform standard score for comparison in attribution calculation.

[0048] Causal variable normalization: Normalize possible causal variables (such as sensor offset, valve opening, etc.) to the [0,1] interval to make causal variables of different natures comparable.

[0049] Discrete variable processing: For discrete causal variables (such as equipment status: normal / fault), a probability encoding method is used to represent them as the probability value of a specific state occurring.

[0050] Then, for the detected abnormal event A_{event}, a causal reasoning algorithm is applied to perform attribution analysis to find the most likely set of causes C_{cause}. The specific implementation is as follows: in, Indicates the observed anomalous events, Represents the set of possible causal variables. This indicates a causal intervention operation. Let C represent the probability of observing A_{event} under the condition of intervention C_{cause}. Indicates the conciseness of the explanation (preferring the simplest explanation). Function This function generates the optimal explanation for an anomaly. Its input is an anomalous event A_{event}, and its output is the set of causes most likely leading to that anomaly. The function searches all possible subsets of causes C_{cause}, calculates the explanation score for each subset (determined by conditional probability and simplicity), and selects the set of causes with the highest score as the final explanation. In its implementation, this function employs a heuristic search algorithm, starting with a small set of causes and gradually expanding the search space to ensure that a near-optimal explanation is found even with limited computational resources.

[0051] It should be noted that the function It is a composite function, whose inner function is... This represents an intervention operation on node C_{cause} in the network. Specifically, it modifies the causal graph structure by deleting all edges pointing to C_{cause}, making C_{cause} an exogenous variable, unaffected by other variables. This intervention operation differs from traditional conditional probability. Unlike other things, it reflects the true causal relationship rather than just the correlation. The calculation is based on the modified graph structure and uses the truncated factorization formula: in This represents the set of all variables in the graph. Representing variables The set of parent nodes, This indicates that the summation is performed on all variables except A_{event} and C_{cause}.

[0052] In addition, the function The simplicity used for quantitative interpretation is achieved as follows: in This indicates the number of variables in the cause set C_{cause}. This represents the complexity of the relationships between variables in C_{cause} (such as distance between variables, connectivity, etc.). and These are the weighting coefficients. This function ensures that, given similar explanatory power, it prioritizes the set of causes with fewer variables and simpler structure, adhering to Occam's razor principle.

[0053] In practical applications, this causal reasoning algorithm is implemented as follows: For an observed anomalous event A_{event} (e.g., abnormally low water supply temperature), the system first identifies the set of all variables that may affect A_{event} in a bicausal network. Then, for each possible subset of causes C_{cause} (e.g., heat source failure, pump failure, etc.), the system calculates the probability of observing the current anomalous event under the condition of intervening in these variables (setting them to specific values). Intervention Operation This is achieved through graph cutting, which involves removing all edges pointing to variables in C_{cause} and then calculating the conditional probability. Finally, the system combines the probability value with the simplicity of the explanation (determined by the size and complexity of the cause set) to select the optimal explanation. For example, if the system observes multiple related sensor anomalies simultaneously but the physical system is normal, it is more likely to attribute the cause to a communication module failure rather than a simultaneous physical failure of multiple sensors.

[0054] Optionally, in resource-constrained edge computing environments, approximate computation methods can be used to reduce computational complexity, such as Markov blanket approximation or variable importance sampling.

[0055] Step 2.4: Counterfactual Verification To further verify the reliability of the attribution results, counterfactual reasoning was used for verification: Construct a counterfactual scenario: "If cause C does not exist, will exception A still occur?" The counterfactual conditional probability P(A_{event}|do(¬C_{cause})) is calculated using a dual-causal network model. Calculate the strength of causal effects; When the intensity of the causal effect exceeds the threshold, C_{cause} is confirmed as the true cause of A_{event}; When calculating the strength of the causal effect, the system first obtains the probability value of observing anomaly A_{event} when the intervention condition C_{cause} exists, and then obtains the probability value of observing anomaly A_{event} when the intervention condition C_{cause} does not exist. Next, the difference between the two probability values ​​is calculated; this difference reflects the strength of the causal influence of condition C_{cause} on anomaly A_{event}. When this difference is greater than a preset threshold (set by the system according to the application scenario, such as 0.3 or 0.5), the system confirms that C_{cause} is the true cause of A_{event}; otherwise, it continues to search for other possible causes.

[0056] The function here This represents the probability of observing anomaly A_{event} when intervention renders C_{cause} nonexistent (or negated). Its calculation method is similar to... Similarly, but with different intervention content, the value of C_{cause} is fixed to its complement or opposite.

[0057] It should be understood that the counterfactual verification step is a key step in ensuring the robustness of attribution results. In some implementations, the counterfactual verification method can be further extended, for example by considering multiple counterfactual scenarios or introducing temporal counterfactuals ("how would A change if C had been intervened earlier") to enhance the reliability of the verification.

[0058] Figure 2 By comparing the accuracy of traditional threshold detection methods with the anomaly detection method based on dual causal networks of this invention in different scenarios, the technical achievement of "significantly improved anomaly detection accuracy: by distinguishing between abnormal user behavior and equipment failure, the diagnostic accuracy is increased to over 92%" is directly verified, specifically demonstrating the comprehensive effectiveness of various anomaly detection methods in step 2.2.

[0059] Figure 3 The figure demonstrates that as the amount of learning data increases, the false positive rate of the method described in this invention is significantly lower than that of traditional methods, and shows a continuous downward trend, verifying the specific data of "reducing the false positive rate by 85%" in the technical effect. At the same time, the figure also illustrates that the system has the ability to learn from historical cases, and its diagnostic accuracy and efficiency continuously improve with extended use, exhibiting a self-evolutionary characteristic.

[0060] Step 3: Develop an interpretable exception report generation system According to embodiments of this application, this step utilizes an interpretable anomaly report generation algorithm to transform the attribution results into a human-understandable explanatory report, outputting an anomaly report containing diagnostic results, explanations, and recommendations. Specifically, it includes the following sub-steps: Step 3.1: Anomaly Classification and Severity Assessment Classify and assess the severity of attribution results: Categorized by source: Abnormal user behavior, equipment failure, system failure; Classified by urgency: Urgent (requires immediate action), Important (requires planned action), Warning (requires close observation); Calculate the scope of impact: single-user impact, multi-user impact, system-level impact; Optionally, the system may also consider the economic impact of anomalies (such as the degree of energy waste and maintenance costs) and safety impacts (such as whether they may lead to equipment damage or safety hazards) as evaluation indicators.

[0061] Step 3.2: Natural Language Interpretation Generation Transform the results of technology attribution into natural language descriptions: Build an explanation template library containing explanation templates for various exceptions; Select an appropriate template based on the attribution results; Fill in the specific parameters and data to generate a natural language explanation; Add visual aids, such as trend charts and comparison charts.

[0062] Step 3.3: Maintenance Recommendation Generation Generate targeted maintenance recommendations based on attribution results: Abnormal user behavior: Provide suggestions for adjusting user behavior; Equipment failure: Generate specific repair plans and procedures; System anomaly: Propose system-level optimization suggestions.

[0063] Step 3.4: Pushing anomaly reports Choose the appropriate report delivery strategy based on the anomaly type and severity: Emergency fault: Pushed to maintenance personnel and management center; Abnormal user behavior: Push notifications to users and customer service; System-level anomaly: Pushed to system administrator and technical support team.

[0064] Step 4: Implement the predictive maintenance recommendation system According to an embodiment of this application, this step utilizes a predictive maintenance algorithm to analyze historical anomaly data and equipment status, and outputs predictive maintenance recommendations. Specifically, it includes the following sub-steps: Step 4.1: Equipment Health Status Assessment Assessing the current health status of devices based on a dual-causal network model: Analyze the historical trends of system state variables; Detect potential performance degradation patterns; Assess the remaining service life of each component of the equipment; In this step, the system uses a health status assessment function to calculate the device's health status: in, express The state vector of the device at any given time. Indicates the first The state variables are in The value at time, This represents the normal values ​​or distribution of the variable. The function calculates the degree of deviation between the current value and the normal value. These are the weighting coefficients of each variable. This represents the total number of state variables in the system state vector. It should be noted that in calculating... Previously, state variables with different dimensions (such as temperature in degrees Celsius, pressure in megapascals, and flow rate in cubic meters per hour) had to be properly standardized to ensure they could be compared and weighted on the same scale. Function The specific implementation is as follows: when the state variable follows a normal distribution, the Z-score is used for calculation, i.e. ,in and These are the mean and standard deviation of the state variable when it is operating normally. For non-normally distributed state variables, a percentile deviation calculation method based on the empirical cumulative distribution function is used to map the original values ​​to percentiles in the range of 0-1 to ensure the comparability of different types of variables. When the health score of a component falls below the threshold, the system will further assess the remaining service life of the component.

[0065] When assessing the remaining useful life of a device component, the system first obtains the component's current health index value, then calculates its rate of change over time (i.e., degradation rate), and combines this with the component's nominal useful life parameters to predict the time required for the component to reach a failure threshold from its current state using a regression model. This calculation considers multiple factors such as current health status, historical degradation patterns, and component type, enabling personalized lifetime predictions for different types of components.

[0066] In practical applications, the dual-causal network model serves as the core of health status assessment, functioning in several ways: First, the model stores the conditional probability distribution of each variable under normal equipment conditions; second, the model captures the strength of causal relationships between variables; and finally, the model includes anomaly propagation paths. The system inputs the current equipment state variables into the model, calculates the degree of deviation of the state variables, and combines this with the strength of causal relationships between variables to generate a health score. For example, for the flow sensor component in a heat meter, the system analyzes indicators such as reading fluctuations and response time changes. If it finds that the response time is gradually increasing but has not yet reached the fault threshold, it marks the component as "performance degradation in progress" and estimates its remaining service life under the current usage mode.

[0067] It should be noted that in some implementations, physical models can be introduced to assist in the health status assessment, such as using fluid dynamics models to analyze the performance of flow sensors or using thermodynamic models to assess heat exchange efficiency, thereby improving the accuracy and reliability of the assessment.

[0068] Step 4.2: Fault Risk Prediction Before performing fault risk prediction, the input data is preprocessed: Time series data standardization: Standardize the time series features (such as temperature change rate, pressure fluctuation, etc.) in historical fault data to eliminate the influence of seasonal and periodic factors.

[0069] Feature scale uniformity: The status indicators of different equipment components (such as temperature sensor accuracy, flow meter deviation, etc.) are unified to the [0,1] interval through min-max normalization to ensure comparability in risk calculation.

[0070] Fault type classification coding: Different types of faults (such as sensor failure, valve jamming, etc.) are individually encoded and converted into numerical feature vectors.

[0071] Then, using the preprocessed historical fault data and the current equipment status, future fault risks are predicted: Construct a failure risk prediction model; Calculate the probability of occurrence of various types of faults within a future time window; Assess the scope and severity of the potential failure; In some implementations, a user-device dual-causal network model can be used as the basis for risk prediction. Forward simulation methods can be used to predict future changes in the system state and assess possible failure scenarios.

[0072] Step 4.3: Maintenance Plan Optimization Based on the failure risk prediction results, an optimized maintenance plan is generated: Determine maintenance priorities and address high-risk, high-impact potential faults first; Develop a reasonable maintenance schedule to avoid peak user heating periods; Optimize the allocation of maintenance resources and improve maintenance efficiency; Generate detailed maintenance operation guidelines and technical support materials.

[0073] Figure 4 The system demonstrates the optimization of maintenance resource allocation after adopting the system of this invention, including the changes in the proportion of necessary maintenance, preventive maintenance and emergency maintenance, which directly corresponds to the maintenance plan optimization process described in lines 265-269, and quantitatively demonstrates the technical effect of improving the efficiency of maintenance human resources by 60%.

[0074] Figure 5 This diagram illustrates the change in system data transmission volume after combining edge computing with a dual-causal network model. It directly verifies the effectiveness of utilizing edge computing capabilities to complete most analysis tasks locally, transmitting only the processed results, thus reducing data transmission volume by 95%. This chart highlights the core feature of this invention—data processing and analysis at the edge, solving the problem of heavy data transmission burden.

[0075] The embodiments of the present invention have been described above. However, the embodiments are not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make more equivalent embodiments under the guidance of the present embodiments, and all of them are within the protection scope of the present embodiments.

Claims

1. A remote data acquisition method for smart heat meters based on edge computing, characterized in that, Includes the following steps: Construct a user-device dual-causal network model, which includes a set of user behavior variables, a set of system state variables, and the causal relationships between them; An anomaly attribution algorithm based on counterfactual reasoning is used to analyze real-time data from heat meters, and anomaly detection results and explanations of their causes are output. Transform attribution results into human-understandable explanatory reports, outputting anomaly reports that include diagnostic results, explanations, and recommendations; Utilize predictive maintenance algorithms to analyze historical anomaly data and equipment status, and output predictive maintenance recommendations; The anomaly attribution algorithm calculates: Determine the most likely cause of the abnormal event. Indicates the observed anomalous events, Represents the set of possible causal variables. This indicates a causal intervention operation. Indicating intervention Observed under the conditions The probability, Indicates the conciseness of the explanation.

2. The method according to claim 1, characterized in that, The steps for constructing the user-device dual causal network model include: Define a dual-causal network structure; Extract user behavior variables and build a user behavior model; Extract the system's physical state variables and establish a system physical model; By using causal discovery algorithms to learn causal relationships between variables from historical data, a complete bicausal network can be constructed.

3. The method according to claim 2, characterized in that, The dual-causal network structure is as follows: ,in Represents a set of user behavior variables. Represents the set of system state variables. This indicates the internal causal relationship between user behavior variables. This represents the internal causal relationship between system state variables. This represents the interactive causal relationship between user behavior variables and system state variables.

4. The method according to claim 1, characterized in that, The steps for analyzing real-time data from heat meters using an anomaly attribution algorithm based on counterfactual reasoning include: Build an abnormal pattern feature library; Anomaly detection is performed on the real-time data collected by the heat meter; Generate causal attributions and explanations for detected abnormal events; Counterfactual reasoning was used to verify the reliability of the attribution results.

5. The method according to claim 4, characterized in that, The counterfactual verification steps include: Constructing a counterfactual scenario: "If the cause..." Not present, abnormal Will it still happen? Calculating counterfactual conditional probabilities using a bicausal network model ; Calculate the causal effect strength; when the causal effect strength exceeds a preset threshold, confirm. for The real reason.

6. The method according to claim 1, characterized in that, The steps for transforming attribution results into human-understandable explanatory reports include: Classify and assess the severity of attribution results; Transform the results of technology attribution into natural language descriptions; Generate targeted maintenance recommendations based on attribution results; Choose an appropriate report delivery strategy based on the type and severity of the anomaly.

7. The method according to claim 1, characterized in that, The steps for analyzing historical anomaly data and equipment status using predictive maintenance algorithms include: Assess the current health status of the device based on a dual-causal network model; Predict future failure risks by using historical failure data and current equipment status; Based on the failure risk prediction results, an optimized maintenance plan is generated.

8. The method according to claim 7, characterized in that, The step of assessing the health status of the equipment employs a health status assessment function. ,in express The state vector of the device at any given time. Indicates the first The state variables are in The value at time, This represents the normal values ​​or distribution of the variable. The function calculates the degree of deviation between the current value and the normal value. These are the weighting coefficients of each variable.

9. The method according to claim 1, characterized in that, The method is executed on the edge computing unit of the heat meter, and only the processed anomaly detection results, interpretation reports and maintenance suggestions are transmitted to the cloud server, reducing the data transmission burden.

10. A remote data acquisition system for a smart heat meter based on edge computing, used to execute the method described in any one of claims 1-9, characterized in that, include: The User-Device Dual Causal Network Model Construction Module is used to construct a dual causal network model that includes a set of user behavior variables, a set of system state variables, and their causal relationships. The anomaly detection and attribution module is used to analyze real-time data from the heat meter using an anomaly attribution algorithm based on counterfactual reasoning, and output anomaly detection results and explanations of their causes. An interpretable anomaly report generation module is used to transform attribution results into human-understandable explanatory reports; The predictive maintenance recommendation module is used to analyze historical anomaly data and equipment status using predictive maintenance algorithms and output predictive maintenance recommendations. The edge computing processing unit is used to execute the functions of the above modules locally on the heat meter, reducing the data transmission burden.

Images