Iot-based intelligent heating system anomaly monitoring method and system

Abstract

The application discloses an Internet of Things-based intelligent heat supply system abnormality monitoring method and system, relates to the technical field of abnormality monitoring, and collects multi-dimensional sensor data of a target user in real time through the Internet of Things, extracts key features representing system operation states therefrom, constructs a dynamic judgment interval in combination with a user real-time heat use state, identifies abnormalities by comparison with normal ranges in a background database. For abnormal events, further extract multi-dimensional information such as water inlet temperature, indoor and outdoor temperature difference, dynamic heat dissipation coefficient change rate and spatial correlation abnormality degree, and input a pre-trained multi-classification model for deep analysis to obtain abnormality diagnosis results containing prediction reasons, reason probabilities and suggestions. In this way, hidden faults can be captured and targeted treatment suggestions can be provided, the abnormality identification accuracy and fault processing efficiency are significantly improved, the fine management and active operation and maintenance of the heat supply system are effectively supported, and the heat supply efficiency and user experience are comprehensively improved.

Description

Technical Field

[0001] This application relates to the field of anomaly monitoring technology, and more specifically, to an anomaly monitoring method and system for a smart heating system based on the Internet of Things. Background Technology

[0002] Against the backdrop of global energy structure transformation and the growing demand for refined management, improving energy efficiency has become a core issue spanning from industrial production to public services. This is especially true in heavy industrial cities whose economies rely heavily on oil extraction and processing, where energy consumption is enormous, making efficiency optimization across the entire energy supply chain crucial. In the oil industry, from exploration and extraction to refining and transportation, complex IoT-based monitoring systems are widely used. These systems enable real-time, high-precision monitoring and anomaly diagnosis of widely distributed pipeline networks, pressure vessels, and fluid conditions to ensure production safety, reduce energy consumption, and achieve intelligent operation and maintenance. This mature monitoring concept and technology system, developed in harsh industrial environments, provides valuable lessons for the modernization and upgrading of other critical urban infrastructure, particularly for urban centralized heating systems that also rely on large-scale pipeline systems and are crucial to energy end-use efficiency and public well-being. Applying industrial-grade intelligent monitoring solutions to the civilian heating sector to address the high energy consumption, low efficiency, and delayed fault response issues prevalent in traditional heating models has become an inevitable trend in promoting smart city construction and achieving energy conservation and emission reduction goals.

[0003] While the integration of IoT technology into heating systems to build smart heating has become an industry consensus, existing solutions mostly remain at a rudimentary level, such as remote meter reading, failing to leverage its data analysis potential. Their anomaly monitoring logic generally relies on simple fixed threshold alarms, setting fixed upper and lower limits for single parameters like temperature and pressure. This static approach ignores the dynamic and interconnected characteristics of the heating system, making it prone to false alarms due to fluctuations in normal operation. Furthermore, it fails to detect slow-developing, hidden faults such as pipe blockages and valve leaks, leading to energy waste and accumulated problems. In addition, even when alarms are triggered, existing systems only provide superficial information, unable to diagnose the root cause of the fault, ultimately requiring manual troubleshooting, which is inefficient and unable to achieve proactive maintenance.

[0004] Therefore, there is an urgent need for an optimized method and system for anomaly monitoring in smart heating systems based on the Internet of Things. Summary of the Invention

[0005] This application is made in order to solve the above-mentioned technical problems.

[0006] According to one aspect of this application, a method for anomaly monitoring of a smart heating system based on the Internet of Things is provided, comprising: acquiring real-time raw sensor data of a target user, the real-time raw sensor data including inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature; extracting core features from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient; extracting the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient of the target user from a background database; determining whether the real-time dynamic heat dissipation coefficient is within the range of the normal lower limit and the normal upper limit of the dynamic heat dissipation coefficient, and if not, calculating an anomaly score and generating an anomaly event; in response to receiving the anomaly event, inputting the real-time raw sensor data into a pre-trained multi-classification model to obtain an anomaly diagnosis result, the anomaly diagnosis result including predicted cause, cause probability, and suggestions; and customizing and displaying alarm information based on the anomaly diagnosis result; wherein, the core features are extracted from the real-time raw sensor data to obtain the real-time dynamic heat dissipation coefficient using the following formula: ;in, For heat capacity, For the specific heat capacity of water, For traffic, For inlet water temperature, For return water temperature, For indoor and outdoor temperature difference, For indoor temperature, outdoor temperature and This is the dynamic heat dissipation coefficient.

[0007] According to another aspect of this application, an IoT-based smart heating system anomaly monitoring system is provided, comprising: a raw sensor data acquisition module for acquiring real-time raw sensor data of a target user, the real-time raw sensor data including inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature; a core feature extraction module for extracting core features from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient; a dynamic heat dissipation coefficient retrieval module for extracting the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient of the target user from a background database; and an anomaly judgment module for judging the real-time dynamic heat dissipation coefficient. Whether the dynamic heat dissipation coefficient is within the normal lower limit and the normal upper limit of the dynamic heat dissipation coefficient; if not, calculate the anomaly score and generate an anomaly event; an anomaly diagnosis module, used to respond to the received anomaly event by inputting the real-time raw sensor data into a pre-trained multi-classification model to obtain anomaly diagnosis results, the anomaly diagnosis results including predicted causes, cause probabilities, and suggestions; an alarm information generation module, used to customize and display alarm information based on the anomaly diagnosis results; wherein, the real-time dynamic heat dissipation coefficient is obtained by extracting core features from the real-time raw sensor data using the following formula: ;in, For heat capacity, For the specific heat capacity of water, For traffic, For inlet water temperature, For return water temperature, For indoor and outdoor temperature difference, For indoor temperature, outdoor temperature and This is the dynamic heat dissipation coefficient.

[0008] Compared with existing technologies, this application provides an IoT-based smart heating system anomaly monitoring method and system. It collects multi-dimensional sensor data from target users in real time via the IoT, extracts key features characterizing the system's operating status, and constructs a dynamic judgment range based on the user's real-time heating status. Anomalies are identified by comparing this range with the normal range in the background database. For abnormal events, it further extracts multi-dimensional information such as inlet water temperature, indoor-outdoor temperature difference, dynamic heat dissipation coefficient change rate, and spatial correlation anomaly degree, and inputs this information into a pre-trained multi-classification model for deep analysis, obtaining anomaly diagnosis results including predicted causes, cause probabilities, and suggestions. This approach can capture hidden faults and provide targeted handling suggestions, significantly improving the accuracy of anomaly identification and fault handling efficiency. It provides strong support for the refined management and proactive operation and maintenance of heating systems, comprehensively improving heating efficiency and user experience. Attached Figure Description

[0009] The above and other objects, features, and advantages of this application will become more apparent from the more detailed description of the embodiments of this application in conjunction with the accompanying drawings. The drawings are provided to further illustrate the embodiments of this application and form part of the specification. They are used together with the embodiments of this application to explain this application and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 This is a flowchart of an anomaly monitoring method for an IoT-based smart heating system according to an embodiment of this application.

[0011] Figure 2 This is a data flow diagram of an anomaly monitoring method for an IoT-based smart heating system according to an embodiment of this application.

[0012] Figure 3 This is a flowchart of sub-step S3 of the IoT-based smart heating system anomaly monitoring method according to an embodiment of this application.

[0013] Figure 4 This is a flowchart of sub-step S5 of the IoT-based smart heating system anomaly monitoring method according to an embodiment of this application.

[0014] Figure 5 This is a block diagram of an IoT-based smart heating system anomaly monitoring system according to an embodiment of this application. Detailed Implementation

[0015] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0016] To address the problems mentioned above in the background technology, this application proposes an anomaly monitoring method for smart heating systems based on the Internet of Things. Figure 1 This is a flowchart of an anomaly monitoring method for an IoT-based smart heating system according to an embodiment of this application. Figure 2 This is a data flow diagram of an anomaly monitoring method for an IoT-based smart heating system according to an embodiment of this application. (See diagram below.) Figure 1 and Figure 2As shown, the IoT-based smart heating system anomaly monitoring method includes the following steps: S1, acquiring real-time raw sensor data of the target user, including inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature; S2, extracting core features from the real-time raw sensor data to obtain the real-time dynamic heat dissipation coefficient; S3, extracting the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient of the target user from the background database; S4, determining whether the real-time dynamic heat dissipation coefficient is within the range of the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient. If not, calculating an anomaly score and generating an anomaly event; S5, in response to receiving the anomaly event, inputting the real-time raw sensor data into a pre-trained multi-classification model to obtain an anomaly diagnosis result, including predicted cause, cause probability, and suggestions; S6, customizing and displaying alarm information based on the anomaly diagnosis result.

[0017] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S1 involves acquiring real-time raw sensor data from the target user. This real-time raw sensor data includes inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature. It should be understood that due to the dynamic complexity of the heating system's operating status and heat transfer process, this application obtains real-time raw sensor data that accurately reflects the current operating status of the heating system and the user's heating environment by collecting multi-dimensional sensor data in real time. Specifically, the inlet and return water temperatures reflect the heat loss of the heating medium, the flow rate reflects heat transfer efficiency, the indoor temperature directly relates to the user's actual heating experience, and the outdoor temperature affects the building's heat dissipation. This provides a reliable data source for subsequent in-depth analysis, avoiding misjudgments caused by missing or delayed data.

[0018] Specifically, in one possible embodiment, step S1 is implemented as follows: First, temperature sensors are installed at the inlet and outlet of the target user's heating pipe to collect inlet and outlet water temperature data in real time. Second, flow sensors are installed on the heating pipe to monitor the flow rate of hot water through the pipe. Simultaneously, indoor temperature sensors are deployed at suitable locations within the user's room, and outdoor temperature sensors are installed in open outdoor areas to acquire indoor and outdoor temperature data, respectively. All sensors are connected to an edge computing gateway via an IoT communication module. The sensors collect data at 1-minute intervals and send the raw data to the edge computing gateway via encrypted transmission. Then, the edge computing gateway performs preliminary verification of the received sensor data, such as removing outliers that are clearly outside the physical range. Afterward, the verified real-time raw sensor data is transmitted to a backend database for storage via a 4G / 5G network. The backend system reads the latest sensor data of the target user from the database through a scheduled task, forming a real-time raw sensor dataset for subsequent feature extraction and anomaly analysis modules to use.

[0019] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S2 involves extracting core features from the real-time raw sensor data to obtain the real-time dynamic heat dissipation coefficient. It should be understood that, because the operating state of a heating system is affected by multiple dimensions of parameters such as inlet water temperature, return water temperature, flow rate, and indoor-outdoor temperature difference, single raw data is insufficient to directly characterize the dynamic characteristics of system heat dissipation. However, the dynamic heat dissipation coefficient can be used to construct a comprehensive index reflecting the actual heat transfer efficiency of the system by integrating the aforementioned parameters. Therefore, this application extracts the real-time dynamic heat dissipation coefficient from real-time raw sensor data, transforming multi-dimensional raw data into a key indicator characterizing the system's heat dissipation efficiency, thereby quantifying anomalies in the heating system and providing more accurate feature inputs for anomaly judgment.

[0020] Specifically, step S2 includes: extracting core features from the real-time raw sensor data using the following formula to obtain the real-time dynamic heat dissipation coefficient, the formula being: ;in, For heat capacity, For the specific heat capacity of water, For traffic, For inlet water temperature, For return water temperature, For indoor and outdoor temperature difference, For indoor temperature, outdoor temperature and This is the dynamic heat dissipation coefficient.

[0021] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S3 involves extracting the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient for the target user from the background database. Specifically, by extracting the normal lower and upper limits of the dynamic heat dissipation coefficient for the target user, and based on the statistical characteristics of the user's historical heating data, such as the mean and standard deviation obtained by fitting a Gaussian distribution, a dynamic threshold range is constructed. This allows the system to compare the real-time calculated dynamic heat dissipation coefficient with this personalized normal range, thereby accurately identifying abnormal states that deviate from the user's actual operating patterns. This avoids misjudgments caused by fixed thresholds and lays the foundation for in-depth diagnosis and precise handling of subsequent abnormal events.

[0022] In particular, in one specific embodiment, Figure 3 This is a flowchart of sub-step S3 of the IoT-based smart heating system anomaly monitoring method according to an embodiment of this application. Figure 3 As shown, step S3 includes: S31, extracting historical data of the dynamic heat dissipation coefficient of the target user from the background database; S32, processing the historical data of the dynamic heat dissipation coefficient using a Gaussian distribution fitting method to obtain the average value and standard deviation of the dynamic heat dissipation coefficient; S33, subtracting the standard deviation of the dynamic heat dissipation coefficient from the average value of the dynamic heat dissipation coefficient to obtain the normal lower limit of the dynamic heat dissipation coefficient; S34, adding the standard deviation of the dynamic heat dissipation coefficient to the average value of the dynamic heat dissipation coefficient to obtain the normal upper limit of the dynamic heat dissipation coefficient.

[0023] Specifically, step S31 involves retrieving historical data of the target user's dynamic heat dissipation coefficient from the background database. This should be understood. Because the target user's heat consumption behavior, outdoor ambient temperature, and other factors are time-varying and vary from person to person, the real-time dynamic heat dissipation coefficient at a single point in time cannot reflect the regularity and fluctuation range of its normal operation. Therefore, this application, by extracting historical data of the target user's dynamic heat dissipation coefficient, such as time series data from the past month, can analyze the central trend and fluctuation range of its daily operation based on the statistical characteristics of the data, thereby constructing a dynamic normal range adapted to the individual characteristics of the user.

[0024] Specifically, in one possible embodiment, step S31 is implemented as follows: First, a query request is sent to the database, with the query condition set to extract historical dynamic heat dissipation coefficient data of the target user within the past month, with a time granularity of 1 minute / record. After receiving the query instruction, the database quickly locates the record corresponding to the target user through the index and filters out all dynamic heat dissipation coefficient data within 30 days prior to the current time. After data extraction is completed, the data is deduplicated and formatted through an extraction, transformation, and loading process to form an ordered historical data list of dynamic heat dissipation coefficients, which is stored in a temporary cache for subsequent processing and analysis.

[0025] Specifically, in step S32, the historical data of the dynamic heat dissipation coefficient is processed using a Gaussian distribution fitting method to obtain the average value and standard deviation of the dynamic heat dissipation coefficient. It should be understood that the Gaussian distribution fitting method is a statistical modeling method based on the characteristics of a normal distribution. It can determine the probability distribution parameters of a set of data through mathematical fitting, making the fitted theoretical distribution as close as possible to the actual data distribution. This application uses the Gaussian distribution fitting method to model the historical data of the dynamic heat dissipation coefficient. The calculated average value of the dynamic heat dissipation coefficient reflects the central trend of normal operation, and the standard deviation of the dynamic heat dissipation coefficient reflects the range of data fluctuation. This provides a quantitative basis for constructing the lower and upper limits of dynamic normality, enabling the normal range to adapt to the natural fluctuations of the data, thereby providing statistical support for setting reasonable dynamic thresholds and improving the accuracy of anomaly identification.

[0026] Specifically, in one possible embodiment, step S32 is implemented as follows: After obtaining the historical data list of the target user's dynamic heat dissipation coefficient, a statistical analysis library is used to fit the data to a Gaussian distribution. Specifically, the historical data is first imported into an array structure, for example, an array containing 43,200 data points (30 days × 24 hours × 60 minutes). Then, the array is fitted using maximum likelihood estimation to calculate the optimal mean and standard deviation parameters. For example, assuming the historical data mean is 250 W / ℃ and the standard deviation is 15 W / ℃, the fitted Gaussian distribution parameters are μ = 250 and σ = 15. Finally, the fitted mean and standard deviation are stored in the configuration table of the corresponding user in the database for subsequent calculation of the dynamic upper and lower limits.

[0027] Specifically, steps S33 and S34 involve subtracting the standard deviation of the dynamic heat dissipation coefficient from the average value of the dynamic heat dissipation coefficient to obtain the lower limit of the dynamic heat dissipation coefficient, and adding the standard deviation of the dynamic heat dissipation coefficient to the average value to obtain the upper limit of the dynamic heat dissipation coefficient. In other words, by using the average value and standard deviation of the dynamic heat dissipation coefficient, a personalized normal operating range is constructed, enabling the system to compare the real-time dynamic heat dissipation coefficient with this range, accurately identifying abnormal states that deviate from the user's actual operating patterns, and providing a dynamic and user-specific threshold standard for anomaly judgment. The resulting upper limit of the dynamic heat dissipation coefficient can adaptively reflect the reasonable fluctuation range under the user's daily heating patterns and environmental changes, significantly improving the targeting and accuracy of anomaly identification.

[0028] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S4 involves determining whether the real-time dynamic heat dissipation coefficient falls within the normal lower limit and upper limit of the dynamic heat dissipation coefficient. If not, an anomaly score is calculated and an anomaly event is generated. Specifically, determining whether the real-time dynamic heat dissipation coefficient is within the normal lower limit and upper limit of the dynamic heat dissipation coefficient can accurately identify abnormal states that deviate from the user's actual operating patterns. When the real-time value exceeds the range, it indicates that the system's operating state has deviated from the normal fluctuation range, posing an anomaly risk.

[0029] Specifically, step S4 includes: if the real-time dynamic heat dissipation coefficient is less than the normal lower limit of the dynamic heat dissipation coefficient, then the anomaly score is calculated using the following formula, wherein the formula is: If the real-time dynamic heat dissipation coefficient is greater than the normal upper limit of the dynamic heat dissipation coefficient, the anomaly score is calculated using the following formula, wherein the formula is: ;in, For abnormal scores, For real-time dynamic heat dissipation coefficient, The lower limit of the dynamic heat dissipation coefficient and This represents the normal upper limit of the dynamic heat dissipation coefficient.

[0030] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S5, in response to receiving the abnormal event, involves inputting the real-time raw sensor data into a pre-trained multi-classification model to obtain anomaly diagnosis results. These results include predicted causes, probability of causes, and recommendations. Specifically, this application extracts features from real-time raw sensor data input into a multi-classification model and constructs anomaly state vectors. Then, it utilizes the model's classification capabilities to obtain anomaly diagnosis results containing predicted causes, probability of causes, and handling recommendations. This enables accurate identification of different anomaly types, such as pipe blockages, valve leaks, and radiator failures, providing precise root cause analysis and handling guidance for abnormal events.

[0031] In particular, in one specific embodiment, Figure 4 This is a flowchart of sub-step S5 of the IoT-based smart heating system anomaly monitoring method according to an embodiment of this application. Figure 4As shown, step S5 includes: S51, extracting inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature from the real-time raw sensor data; S52, calculating the indoor-outdoor temperature difference based on the indoor temperature and the outdoor temperature; S53, calculating the rate of change of the dynamic heat dissipation coefficient; S54, calculating the spatial correlation anomaly degree of the target user; S55, arranging the inlet water temperature, return water temperature, flow rate, indoor temperature, outdoor temperature, indoor-outdoor temperature difference, rate of change of the dynamic heat dissipation coefficient, and spatial correlation anomaly degree into an anomaly state vector; S56, inputting the anomaly state vector into the pre-trained multi-classification model to obtain the anomaly diagnosis result.

[0032] Specifically, step S51 involves extracting the inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature from the real-time raw sensor data. It should be understood that the inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature are fundamental parameters reflecting the operating status of the heating system and the user's heating environment. This application extracts these raw parameters from the real-time raw sensor data, transforming the raw data collected by the sensors into structured features suitable for in-depth analysis. This provides foundational data for subsequently constructing abnormal state vectors, enabling multi-classification models to predict the causes of anomalies based on a complete feature set.

[0033] Specifically, in one possible embodiment, step S51 is implemented as follows: In response to receiving the abnormal event, the real-time raw sensor data record of the target user is read from the real-time data buffer. The data record is in JSON format, containing a timestamp and various parameter values. For example, the record for 10:00 AM on June 27, 2025 is read: inlet water temperature 45℃, return water temperature 38℃, flow rate Fr = 200 kg / h, indoor temperature 20℃, and outdoor temperature 5℃. Then, by parsing the JSON fields, the parameter values ​​are stored in temporary variables, and the timestamp is recorded synchronously. After extraction, the data is validated, such as checking whether the temperature is within the sensor's range of -20℃ to 100℃ and whether the flow rate is positive. If the validation passes, the subsequent processing flow is initiated; otherwise, the data is marked as invalid and re-acquired.

[0034] Specifically, steps S52 and S53 calculate the indoor-outdoor temperature difference based on the indoor and outdoor temperatures, and calculate the rate of change of the dynamic heat dissipation coefficient. It should be understood that by calculating the indoor-outdoor temperature difference, the indoor and outdoor temperatures are transformed into quantitative indicators reflecting the building's heat load. Simultaneously, by combining this with the rate of change of the dynamic heat dissipation coefficient, multi-dimensional features encompassing the spatial thermal environment and temporal trends can be provided to the multi-classification model. This enables the model to distinguish between instantaneous environmental fluctuations and gradual faults, achieving accurate identification of hidden anomalies, such as a slow decline in radiator efficiency.

[0035] Specifically, in one possible embodiment, steps S52 and S53 are implemented as follows: First, the current indoor temperature is extracted from the real-time sensor data. and outdoor temperature Perform temperature difference calculation Simultaneously, the current DHC value and the DHC value from one hour ago are retrieved from the database, and the rate of change is calculated using a formula: .

[0036] Before calculation, the temperature data needs to be validated, such as checking if it falls within the range of -20℃ to 100℃, and the DHC historical data needs to be timestamped to ensure that the data interval is 1 hour. The temperature difference and rate of change results are stored in a temporary feature table with fields including user ID, timestamp, temperature difference value, and rate of change, awaiting further processing.

[0037] Specifically, step S54 involves calculating the spatial correlation anomaly degree of the target user. It should be understood that the spatial correlation anomaly degree is a quantitative indicator used to measure the degree of correlation between abnormal events of a target user in a spatial dimension. By statistically analyzing the distribution density of similar anomalies among other users in the same building unit, it is possible to determine whether the anomaly is an individual fault or a systemic problem, thereby significantly narrowing the scope of fault investigation, improving location efficiency, and reducing manual investigation costs.

[0038] Specifically, in one embodiment, step S54 includes: querying the building ID of the target user based on the target user's user ID; querying all user IDs under the building ID other than the target user; counting the total number of user IDs with the same abnormal events as the target user as the total number of abnormal neighbors; and dividing the total number of abnormal neighbors by the total number of users to obtain the spatial correlation anomaly degree. It should be understood that since users in the same building share the same riser and have a direct hydraulic relationship, when a systemic problem occurs, such as a valve being accidentally closed or a main pipeline failure, it will affect multiple users in the same building in a short period of time, leading to a concentrated occurrence of similar abnormal events. Therefore, by querying the building ID of the target user and other user IDs in that building, counting the total number of abnormal neighbors with similar abnormal events, quantifying the spatial distribution density of abnormal events, and generating a spatial correlation anomaly degree index, it is possible to determine whether anomalies have spatial correlation, thereby distinguishing between individual equipment failures and systemic failures, providing maintenance personnel with more targeted troubleshooting directions, reducing the scope of manual troubleshooting, and improving processing efficiency.

[0039] Specifically, in one possible embodiment, step S54 is implemented as follows: For example, if the target user's user ID is "U-00123", firstly, the user's corresponding building ID "Unit B-3" is retrieved from the user table. Next, all user IDs under "Unit B-3" are queried. After excluding the target user "U-00123", other user IDs in the same unit are obtained as "U-00124" to "U-00139", a total of 16 users. Then, a query is performed in the abnormal event history record to filter out records with building ID "Unit B-3", user ID not equal to "U-00123", event timestamp within the past 2 hours, and anomaly type the same as the target user. In one possible embodiment, the user IDs that meet the conditions are "U-00125", "U-00127", "U-00130", and "U-00135", a total of 4. Finally, the total number of abnormal neighbors, 4, is divided by the total number of users in the unit, 16, to obtain a spatial association anomaly degree of 0.25. This value indicates that 25% of users in this unit experienced the same type of anomaly within the same time period. Based on this, the system judges that there may be systemic problems such as abnormal riser flow or local pipe blockage, providing key spatial features for subsequent multi-classification model diagnosis.

[0040] Specifically, in step S55, the inlet water temperature, return water temperature, flow rate, indoor temperature, outdoor temperature, indoor-outdoor temperature difference, rate of change of dynamic heat dissipation coefficient, and spatial correlation anomaly degree are arranged into an abnormal state vector. In other words, by orderly arranging multi-dimensional features into an abnormal state vector, a standardized feature matrix that meets the input requirements of a pre-trained multi-classification model is constructed. This enables the model to achieve deep semantic understanding of abnormal events based on basic parameters such as inlet water temperature and return water temperature, combined with derived indicators such as indoor-outdoor temperature difference and rate of change of dynamic heat dissipation coefficient, as well as spatial correlation features, providing multi-dimensional data support for anomaly root cause analysis. It should be understood that the abnormal state vector integrates the thermodynamic parameters, environmental parameters, dynamic change characteristics, and spatial propagation characteristics of the heating system. Inputting it into a multi-classification model can significantly improve the ability to identify hidden faults. For example, when the inlet water temperature is normal but the return water temperature is low, the flow rate is reduced, the dynamic heat dissipation coefficient decreases, and the spatial correlation anomaly degree is high, the model can accurately determine that it is a partial blockage in the riser pipe, rather than a fault in a single user device, thus providing accurate fault location basis for operation and maintenance.

[0041] Specifically, in one possible embodiment, step S55 is implemented as follows: First, extract the following parameters from the real-time raw sensor data: inlet water temperature 45℃, return water temperature 38℃, flow rate 200kg / h, indoor temperature 20℃, and outdoor temperature 5℃. Second, calculate the indoor-outdoor temperature difference as 20℃-5℃=15℃, and assume the current DHC value is 108.9W / ℃ and the DHC value one hour ago was 340W / ℃, calculate the rate of change of the dynamic heat dissipation coefficient as (108.9-340) / 340=-0.68. Next, obtain the spatial correlation anomaly degree as 0.25, such as in the previous example where 4 out of 16 households in the building exhibited the same anomaly. Finally, arrange the above inlet water temperature, return water temperature, flow rate, indoor temperature, outdoor temperature, indoor-outdoor temperature difference, rate of change of the dynamic heat dissipation coefficient, and spatial correlation anomaly degree in an ordered manner into an anomaly state vector [45,38,200,20,5,15,-0.68,0.25].

[0042] Specifically, in step S56, the abnormal state vector is input into the pre-trained multi-classification model to obtain the abnormal diagnosis result. Specifically, after the abnormal state vector is input into the model, the model can quickly calculate the probability values ​​of each possible cause based on learned knowledge, such as the probability distribution of "normal inlet water temperature but low return water temperature + decreased flow rate + decreased dynamic heat dissipation coefficient + high spatial correlation anomaly" corresponding to "half-open riser valve." It then outputs predicted probabilities for different abnormal causes and corresponding handling suggestions, prioritizing high-probability causes and handling measures, providing maintenance personnel with accurate fault location information.

[0043] Specifically, in one possible embodiment, step S56 is implemented as follows: First, the abnormal state vector is input into the pre-trained multi-classification model, which is an encoder-multi-classification head architecture. The encoder is responsible for extracting high-level, robust feature representations from the 8-dimensional input features. For example, it can be composed of multiple fully connected neural networks, with each layer followed by a ReLU activation function to enhance nonlinear expressive power. The multi-classification head receives the features output by the encoder and maps them to the probability distributions of five types of abnormal causes, such as half-open riser valve, slight pipe blockage, radiator dust accumulation, user-end valve not fully open, and heat source temperature fluctuation. During the model's forward propagation, the input features are first subjected to layer-by-layer feature learning and abstraction by the encoder to extract discriminative intermediate feature representations. Subsequently, these intermediate features are fed into the multi-classification head, and through its internal fully connected layers and Softmax activation function, the probability distributions of each abnormal cause are finally output. For example, if the output results are: riser valve half-open probability 0.85, pipe slight blockage probability 0.10, radiator dust accumulation probability 0.03, user-end valve not fully open probability 0.01, heat source temperature fluctuation probability 0.01, then the riser valve half-open corresponding to the highest probability value is taken as the predicted cause, and the abnormal diagnosis result is output.

[0044] It is worth noting that when the abnormal state vector is input into the multi-class classification model for classification, since the abnormal event state has been determined, the overall prior state of the classification mapping is determined. In the process of determining the prior state, the state determination is made by using the dynamic heat dissipation coefficient as the core indicator. Therefore, the rate of change of the dynamic heat dissipation coefficient becomes the focused prior value. On the other hand, since the spatial association anomaly of the target user will significantly affect the nature of the anomaly diagnosis result, the spatial association anomaly of the target user can also be classified as the focused prior value. Thus, the improvement of the multi-class classification task is a multi-class optimization based on a partial prior value under the determinacy of the overall prior state.

[0045] Specifically, in another possible embodiment, inputting the abnormal state vector into the pre-trained multi-classification model to obtain the abnormal diagnosis result includes: first, inputting the abnormal state vector into the encoder of the pre-trained multi-classification model to obtain an abnormal state latent encoding vector. It should be understood that the encoder of the pre-trained multi-classification model consists of a multi-layer fully connected neural network, with each layer followed by a ReLU activation function to enhance nonlinear expressive power. Its core responsibility is to extract high-level, robust feature representations from the original 8-dimensional input features (including inlet water temperature, return water temperature, flow rate, indoor temperature, outdoor temperature, indoor-outdoor temperature difference, rate of change of dynamic heat dissipation coefficient, and spatial correlation anomaly degree). Specifically, although the sensor data and derived features of the heating system are abundant, they may contain redundancy and complex nonlinear relationships, making direct use for diagnosis potentially inefficient. Transforming them into a more compact and discriminative latent encoding vector through the encoder can effectively filter out noise and capture the essential characteristics of abnormal states, such as deep patterns of overall system heat loss or local circulation obstacles.

[0046] Then, the hidden encoding vector of the abnormal state Local correlation aggregation is performed to obtain the local correlation aggregation encoding vector of the abnormal state, i.e., one-dimensional convolution is performed to obtain... ,in, This refers to one-dimensional convolution, where the kernel length is greater than a predetermined threshold, such as greater than five. In other words, many abnormal phenomena in heating systems, such as partial pipe blockage or radiator dust accumulation, often exhibit specific local correlation patterns in sensor data. For example, there may be specific linkages between inlet water temperature, return water temperature, and flow rate, or a correlation between the indoor / outdoor temperature difference and the rate of change of the heat dissipation coefficient. Through local correlation aggregation (e.g., using one-dimensional convolution operations with sufficiently long kernels), local focusing physical behavior can be simulated during weighted aggregation based on mid-range neighborhoods to fully capture the correlation patterns of some focusing prior feature values. This allows for the discovery of deep connections between local features, improving the perception of specific physical behaviors (such as increased local thermal resistance) and facilitating local classification decisions.

[0047] Next, the local correlation aggregation encoding vector of the abnormal state is subjected to propagation and diffusion based on focusing prior to obtain the a priori propagation and diffusion encoding vector of the abnormal state, that is, for eigenvalues and Perform the following operation, which is represented as: ;in, and These respectively represent the first and second parts of the local correlation aggregation encoding vector of the abnormal state. The and the first 1 eigenvalue, The first element representing the prior propagation diffusion encoding vector of the abnormal state Several eigenvalues. It should be understood that in the diagnosis of heating system anomalies, certain indicators are considered highly focused prior information, having a decisive impact on the diagnostic results. For example, a significant rate of change in the dynamic heat dissipation coefficient directly points to the overall thermal efficiency of the system, while a high degree of spatial correlation anomaly (multiple users in the same building exhibiting the same anomaly) strongly suggests a systemic fault such as a riser or main pipe, rather than a problem with a single user's equipment. These focused priors require stronger diagnostic influence. By using differential signal gradient propagation, these focused prior signals are diffused and strengthened in the locally correlated aggregated encoding vector, allowing this key information to more profoundly influence the representation of other features. Thus, based on a locally preset focused pattern, the differential gradient propagation of local focused prior eigenvalues ​​through an extension process amplifies the weight and influence of these key prior information in the overall feature representation, guiding the model to focus more on these highly diagnostic features, such as distinguishing between individual radiator faults and building riser problems.

[0048] Next, the overall prior comparison factor between the anomalous state prior propagation diffusion encoding vector and the anomalous state latent encoding vector is calculated, and is expressed as: ,in, This represents the prior propagation diffusion encoding vector of the abnormal state. The hidden encoding vector representing the abnormal state Norm 1 This represents vector subtraction. The L2 norm of a vector. This represents the overall prior comparison factor. It should be understood that while the anomalous state prior propagation and diffusion encoding vector and the anomalous state latent encoding vector can enhance specific information, for complex heating systems, overemphasizing local or single priors while detaching from the overall context may lead to misjudgment. Therefore, it is necessary to compare these processed vectors with the original, more general latent encoding vector to ensure that the enhanced feature representation is consistent with the overall anomalous state of the heating system. In this way, by using the overall prior state as a deterministic reference and focusing on the multidimensional response of the prior through comparison matching, while still using the prior state as an expected reference to provide a response reference point, the prior stability of the overall classification decision is promoted. This provides a quantified overall prior comparison factor that reflects the difference or consistency between the feature representation after prior propagation and diffusion and the original latent representation, providing important guidance for subsequent feature fusion and ensuring the stability and consistency of the diagnostic process.

[0049] Next, based on the overall prior comparison factor, the anomalous state prior propagation diffusion encoding vector and the anomalous state local correlation aggregation encoding vector are fused to obtain the optimized anomalous state latent encoding vector, expressed as: ,in, This indicates a positional dot product, i.e., a product of positions. Add the feature value at each position , This represents vector addition. This represents the local correlation aggregation encoding vector of the abnormal state. This represents the implicit encoded vector of the optimized abnormal state. It should be understood that heating system anomalies are a multifaceted issue requiring comprehensive consideration. Previously, abnormal states were encoded and enhanced at different levels (original feature extraction, local correlation, and focused prior propagation). Now, it is necessary to effectively integrate this information from different levels to form a final feature representation that includes comprehensive information while highlighting key priors. By using the overall prior comparison factor as a weight or modulation factor, the vector after prior propagation and the vector after local correlation aggregation can be intelligently fused. This allows the final optimized vector to take into account both local details and global prior guidance, thus more comprehensively reflecting the abnormal state of the heating system. In this way, focusing on partial prior values ​​and the determinism of the overall prior state promotes classification decisions, thereby improving the classification performance of multi-classification models.

[0050] Finally, the optimized abnormal state latent encoding vector is input into the multi-classifier head of the pre-trained multi-classifier model to obtain the abnormal diagnosis result. It should be understood that the multi-classifier head receives the optimized abnormal state latent encoding vector and maps it to a preset probability distribution of five types of heating system anomalies, such as half-open riser valves, minor pipe blockage, radiator dust accumulation, user-end valves not fully open, and heat source temperature fluctuations. Internally, it typically contains fully connected layers and a Softmax activation function, responsible for calculating the probability of each anomaly cause.

[0051] In the aforementioned IoT-based smart heating system anomaly monitoring method, step S6 involves customizing and displaying alarm information based on the anomaly diagnosis results. Specifically, by performing structured analysis and visualization of the anomaly diagnosis results, the abstract model output is transformed into standardized alarm information containing fault type, confidence level, and handling suggestions. Furthermore, differentiated display strategies, such as color coding and priority sorting, are adopted for different anomaly levels to provide maintenance personnel with precise fault handling guidance, thereby improving the automation and intelligence level of heating system operation and maintenance.

[0052] Specifically, in one possible embodiment, step S6 is implemented as follows: After the multi-classification model outputs anomaly diagnosis results for target user U-00123, the anomaly diagnosis results are first subjected to structured analysis. The results show that the predicted cause is "riser valve half-open" with a probability of 85%, and the secondary cause is "minor pipe blockage" with a probability of 10%, along with the suggestion to "check the opening degree of riser valve in unit B-3". Meanwhile, the key parameters in the anomaly state vector include inlet water temperature 45℃, return water temperature 38℃, flow rate 200kg / h, indoor temperature 20℃, outdoor temperature 5℃, indoor-outdoor temperature difference 15℃, dynamic heat dissipation coefficient change rate -0.68, and spatial correlation anomaly degree 0.25. Key information such as the main cause prediction, probability value, correlation suggestion, secondary cause, anomaly score, and spatial correlation anomaly degree are extracted to provide data support for subsequent alarm customization.

[0053] When determining the alarm level and display strategy, based on the conditions of a primary cause probability of 85% ≥ 80% and an anomaly score of 0.32 ≥ 0.3, this alarm was classified as an "urgent alarm." A visual identifier with a red background, a valve icon, and an exclamation mark was adopted, and it was set to the highest priority, displaying it prominently on the operations and maintenance platform homepage to remind operations and maintenance personnel. This differentiated display strategy ensures that urgent faults receive immediate attention, avoiding processing delays caused by information hierarchy confusion.

[0054] Subsequently, the system generates customized alarm content based on a preset template. This content, titled "Emergency Alarm: Heating Abnormality for User U-00123 in Unit B-3," presents the predicted cause ("Riser valve partially open (confidence level 85%)"), abnormal characteristics ("Inlet water temperature 45℃, return water temperature 38℃, flow rate 200kg / h, dynamic heat dissipation coefficient 108.9W / ℃, 25% of users in the unit experience similar abnormalities"), handling suggestions ("Immediately check the opening of the riser valve in Unit B-3; it is recommended to fully open the valve to restore normal heat flow"), and auxiliary information ("Secondary cause is minor pipe blockage; simultaneously check for debris accumulation in the pipes"). This template-based content generation method ensures information completeness while facilitating rapid retrieval of key instructions by maintenance personnel.

[0055] In terms of multi-terminal display adaptation, the operation and maintenance management platform presents complete alarm content in a card-style pop-up window, along with a nearly one-hour DHC value change curve and a heat map of abnormal user distribution in the unit. The curve on the left intuitively shows the deviation of the real-time DHC value from the normal range, while the heat map on the right marks the location of abnormal users in red. Hovering the mouse over the user allows viewing detailed data, providing operation and maintenance personnel with visualized fault analysis assistance. The mobile app pushes a simplified notification titled "[Urgent] Abnormal riser valve in Unit B-3," with some technical parameters omitted from the details page, highlighting "It is suspected that the riser valve is not fully open. It is recommended to check immediately. Currently, 25% of users are affected." It also provides one-click dialing and map navigation functions, facilitating rapid response by operation and maintenance personnel in mobile scenarios. For the user-side mini-program, the system displays a gentle prompt: "Heating in your area may be affected by pipeline adjustments. The temperature is temporarily low. Staff are investigating and it is expected to be restored within one hour." This avoids user anxiety while concealing technical details, ensuring appropriate information delivery.

[0056] After an alarm is generated, the system stores it in the backend database and performs multi-dimensional correlation. The correlation includes basic information such as the target user ID, diagnosis time, and abnormal state vector. It also automatically retrieves historical processing records for that unit; for example, if a case occurred three months ago where a riser valve was partially open due to rust, the system will prompt maintenance personnel to focus on checking the valve's corrosion. When maintenance personnel click the "Processed" button, they need to enter the actual processing result, such as "The valve was indeed not fully open; it has been adjusted to the fully open state." By automatically comparing the predicted cause with the actual cause, this data is included in the model training set to continuously optimize the accuracy of subsequent diagnoses.

[0057] In the closed-loop alarm management process, the system periodically checks the alarm handling status. If an alarm remains unhandled for more than 30 minutes, it is automatically escalated and pushed to the operations and maintenance supervisor. The "in progress" status displays the estimated completion time, while the "resolved" status records the handling time and the person responsible, and a recovery notification is pushed to the user's mobile app. For example, if operations and maintenance personnel complete valve adjustments at 13:00, the system will push a notification to the user at 13:05 stating "Heating has returned to normal," and mark the alarm status as green "Resolved" on the operations and maintenance platform. Through this closed-loop management throughout the entire process, every alarm is ensured to be followed up and handled promptly, avoiding the omission of abnormal events and effectively improving the stability and reliability of the heating system.

[0058] In summary, the IoT-based smart heating system anomaly monitoring method based on the embodiments of this application is explained. It collects multi-dimensional sensor data from target users in real time via the IoT, extracts key features characterizing the system's operating status, and constructs a dynamic judgment range based on the user's real-time heating status. Anomalies are identified by comparing this range with the normal range in the background database. For abnormal events, multi-dimensional information such as inlet water temperature, indoor-outdoor temperature difference, dynamic heat dissipation coefficient change rate, and spatial correlation anomaly degree are further extracted and input into a pre-trained multi-classification model for deep analysis, yielding anomaly diagnosis results including predicted causes, cause probabilities, and suggestions. This approach can capture hidden faults and provide targeted handling suggestions, significantly improving the accuracy of anomaly identification and fault handling efficiency. It provides strong support for the refined management and proactive operation and maintenance of the heating system, comprehensively improving heating efficiency and user experience.

[0059] Figure 5 This is a block diagram of an IoT-based smart heating system anomaly monitoring system according to an embodiment of this application. Figure 5 As shown, the IoT-based smart heating system anomaly monitoring system 100 according to an embodiment of this application includes: a raw sensor data acquisition module 110, used to acquire real-time raw sensor data of a target user, the real-time raw sensor data including inlet water temperature, return water temperature, flow rate, indoor temperature, and outdoor temperature; a core feature extraction module 120, used to extract core features from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient; a dynamic heat dissipation coefficient retrieval module 130, used to extract the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient of the target user from a background database; an anomaly judgment module 140, used to determine whether the real-time dynamic heat dissipation coefficient is within the range of the normal lower limit and normal upper limit of the dynamic heat dissipation coefficient, and if not, calculate an anomaly score and generate an anomaly event; an anomaly diagnosis module 150, used to, in response to receiving the anomaly event, input the real-time raw sensor data into a pre-trained multi-classification model to obtain an anomaly diagnosis result, the anomaly diagnosis result including predicted cause, cause probability, and suggestions; and an alarm information generation module 160, used to customize and display alarm information based on the anomaly diagnosis result.

[0060] As described above, the IoT-based smart heating system anomaly monitoring system 100 according to embodiments of this application can be implemented in various wireless terminals. In one possible implementation, the IoT-based smart heating system anomaly monitoring system 100 according to embodiments of this application can be integrated into the wireless terminal as a software module and / or hardware module. For example, the IoT-based smart heating system anomaly monitoring system 100 can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the IoT-based smart heating system anomaly monitoring system 100 can also be one of many hardware modules of the wireless terminal.

[0061] Alternatively, in another example, the IoT-based smart heating system anomaly monitoring system 100 and the wireless terminal can also be separate devices, and the IoT-based smart heating system anomaly monitoring system 100 can be connected to the wireless terminal via wired and / or wireless networks, and transmit interactive information in accordance with an agreed data format.

[0062] Here, those skilled in the art will understand that the specific operations of each step in the above-mentioned IoT-based smart heating system anomaly monitoring system have been referenced above. Figures 1 to 4 The description of the IoT-based smart heating system anomaly monitoring method is detailed here, and therefore, its repeated description will be omitted.

Claims

1. An abnormality monitoring method for an Internet of Things (IoT)-based intelligent heating system, characterized by comprising: The method comprises the following steps: acquiring real-time raw sensor data of a target user, the real-time raw sensor data comprising inlet water temperature, return water temperature, flow rate, indoor temperature and outdoor temperature; extracting core features from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient; extracting a dynamic heat dissipation coefficient lower limit and a dynamic heat dissipation coefficient upper limit of the target user from a background database; determining whether the real-time dynamic heat dissipation coefficient is within the range of the dynamic heat dissipation coefficient lower limit and the dynamic heat dissipation coefficient upper limit, and if not, calculating an abnormality score and generating an abnormal event; In response to receiving the abnormal event, inputting the real-time raw sensor data into a pre-trained multi-classification model to obtain an abnormal diagnosis result, the abnormal diagnosis result including a predicted cause, a cause probability and a suggestion, comprising: extracting a water inlet temperature, a water return temperature, a flow rate, an indoor temperature and an outdoor temperature from the real-time raw sensor data; calculating an indoor-outdoor temperature difference based on the indoor temperature and the outdoor temperature; calculating a change rate of a dynamic heat dissipation coefficient; calculating a spatial correlation abnormality degree of the target user; arranging the water inlet temperature, the water return temperature, the flow rate, the indoor temperature, the outdoor temperature, the indoor-outdoor temperature difference, the change rate of the dynamic heat dissipation coefficient and the spatial correlation abnormality degree into an abnormal state vector; inputting the abnormal state vector into the pre-trained multi-classification model to obtain the abnormal diagnosis result; based on the abnormal diagnosis result, customizing and displaying alarm information; wherein the core features are extracted from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient according to the following formula: wherein, is a heat capacity, is a specific heat capacity of water, is a flow rate, is a water inlet temperature, is a water return temperature, is an indoor-outdoor temperature difference, is an indoor temperature, is an outdoor temperature, and is a dynamic heat dissipation coefficient. 2.The IoT-based intelligent heat supply system anomaly monitoring method of claim 1, wherein, extracting a dynamic heat dissipation coefficient lower limit and a dynamic heat dissipation coefficient upper limit of the target user from a background database, comprising: extracting historical data of the dynamic heat dissipation coefficient of the target user from the background database; processing the historical data of the dynamic heat dissipation coefficient using a Gaussian distribution fitting method to obtain a dynamic heat dissipation coefficient mean value and a dynamic heat dissipation coefficient standard deviation; subtracting the dynamic heat dissipation coefficient standard deviation from the dynamic heat dissipation coefficient mean value to obtain the dynamic heat dissipation coefficient lower limit; adding the dynamic heat dissipation coefficient standard deviation to the dynamic heat dissipation coefficient mean value to obtain the dynamic heat dissipation coefficient upper limit. 3.The IoT-based intelligent heat supply system anomaly monitoring method of claim 2, wherein, The historical data of the dynamic heat dissipation coefficient is the dynamic heat dissipation coefficient of the target user in the past month. 4.The IoT-based intelligent heat supply system anomaly monitoring method of claim 1, wherein, determining whether the real-time dynamic heat dissipation coefficient is within a range of a lower limit of the normal dynamic heat dissipation coefficient and an upper limit of the normal dynamic heat dissipation coefficient, and if not, calculating an anomaly score and generating an anomaly event, including: if the real-time dynamic heat dissipation coefficient is less than the lower limit of the normal dynamic heat dissipation coefficient, calculating the anomaly score with the following formula, wherein the formula is: if the real-time dynamic heat dissipation coefficient is greater than the upper limit of the normal dynamic heat dissipation coefficient, calculating the anomaly score with the following formula, wherein the formula is: wherein, is the anomaly score, is the real-time dynamic heat dissipation coefficient, is the lower limit of the normal dynamic heat dissipation coefficient, and is the upper limit of the normal dynamic heat dissipation coefficient. 5.The IoT-based intelligent heating system anomaly monitoring method of claim 4, wherein, inputting the abnormal state vector into the pre-trained multi-classification model to obtain the abnormal diagnosis result, comprising: inputting the abnormal state vector into an encoder of the pre-trained multi-classification model to obtain an abnormal state implicit encoding vector; performing local correlation aggregation on the abnormal state implicit encoding vector to obtain an abnormal state local correlation aggregation encoding vector; performing focus prior-based propagation diffusion on the abnormal state local correlation aggregation encoding vector to obtain an abnormal state prior propagation diffusion encoding vector; calculating an overall prior comparison factor between the abnormal state prior propagation diffusion encoding vector and the abnormal state implicit encoding vector; fusing the abnormal state prior propagation diffusion encoding vector and the abnormal state local correlation aggregation encoding vector based on the overall prior comparison factor to obtain an optimized abnormal state implicit encoding vector; inputting the optimized abnormal state implicit encoding vector into a multi-classification head of the pre-trained multi-classification model to obtain the abnormal diagnosis result. 6.The IoT-based intelligent heat supply system anomaly monitoring method of claim 1, wherein, calculating a spatial correlation abnormality degree of the target user, comprising: querying a unit building ID in which the user ID of the target user is located based on the user ID of the target user; querying all user IDs under the unit building ID except the target user; counting the total number of user IDs having the same abnormal event as the target user among all user IDs as an abnormal neighbor total number; dividing the abnormal neighbor total number by the total number of users to obtain the spatial correlation abnormality degree.

7. An Internet of Things-based intelligent heating system anomaly monitoring system, characterized in that, The method comprises the following steps: an original sensor data acquisition module, configured to acquire real-time raw sensor data of a target user, the real-time raw sensor data comprising inlet water temperature, return water temperature, flow rate, indoor temperature and outdoor temperature; a core feature extraction module configured to extract core features from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient; a dynamic heat dissipation coefficient retrieval module configured to extract a dynamic heat dissipation coefficient lower limit and a dynamic heat dissipation coefficient upper limit of the target user from a background database; an anomaly judgment module configured to judge whether the real-time dynamic heat dissipation coefficient is within the range of the dynamic heat dissipation coefficient lower limit and the dynamic heat dissipation coefficient upper limit, and if not, to calculate an anomaly score and generate an anomaly event; The abnormality diagnosis module is configured to, in response to receiving the abnormal event, input the real-time raw sensor data into a pre-trained multi-classification model to obtain an abnormality diagnosis result, the abnormality diagnosis result including a predicted cause, a cause probability, and a suggestion, including: extracting a water inlet temperature, a water return temperature, a flow rate, an indoor temperature, and an outdoor temperature from the real-time raw sensor data; calculating an indoor-outdoor temperature difference based on the indoor temperature and the outdoor temperature; calculating a change rate of a dynamic heat dissipation coefficient; calculating a spatial correlation abnormality degree of the target user; arranging the water inlet temperature, the water return temperature, the flow rate, the indoor temperature, the outdoor temperature, the indoor-outdoor temperature difference, the change rate of the dynamic heat dissipation coefficient, and the spatial correlation abnormality degree into an abnormality state vector; inputting the abnormality state vector into the pre-trained multi-classification model to obtain the abnormality diagnosis result; and an alarm information generation module configured to customize and display alarm information based on the abnormality diagnosis result; wherein core features are extracted from the real-time raw sensor data to obtain a real-time dynamic heat dissipation coefficient according to the following formula: wherein, is a heat capacity rate, is a specific heat capacity of water, is a flow rate, is a water inlet temperature, is a water return temperature, is an indoor-outdoor temperature difference, is an indoor temperature, is an outdoor temperature, and is a dynamic heat dissipation coefficient.

Images