A method and system for analyzing operation data of a heating system

Abstract

The application provides a kind of operation data analysis method and system of heating system, it is related to building environment and energy application engineering technical field, the method includes: if diagnosis result is no exception, based on the building historical room temperature in multi-source fusion data set, building characteristics and real-time meteorological parameters, generate different operation strategy;Obtain the meteorological prediction data of preset time period;Based on the different operation strategy and meteorological prediction data, the change of heat load demand is predicted by long short-term memory network;Based on the change of heat load demand, with room temperature standard and minimum energy consumption as optimization goal, the target output parameter set value is calculated, and the set value is sent to the boiler control system to execute.The application realizes closed-loop intelligent operation from abnormal accurate diagnosis to demand prediction regulation through multi-source data fusion and real-time analysis, so as to comprehensively improve the reliability, energy efficiency and comfort of heating system.

Description

Technical Field

[0001] This invention relates to the field of building environment and energy application engineering technology, and in particular to a method and system for analyzing the operation data of a heating system. Background Technology

[0002] With the acceleration of urbanization and the continuous expansion of centralized heating in my country, the structure of heating systems is becoming increasingly complex. The traditional operation and management model, which relies on manual experience, can no longer meet the needs of refined and intelligent operation. Although some systems have deployed automated monitoring equipment and achieved basic data collection, due to insufficient data processing and analysis capabilities, the systems still have defects, which restrict the improvement of operational efficiency and service quality.

[0003] Taking the heating system of an old residential community in northern China as an example, it serves 288 households (8 buildings with 6 floors), uses a 2.8MW gas-fired hot water boiler, and deploys three types of data collection equipment: the boiler side collects water temperature, pressure, etc. every minute (local storage), the pipeline side collects the building's supply and return water parameters every 5 minutes (stored on the property management platform), and 10% of the residents upload room temperature data every 30 minutes (including complaints via the APP). The system has the following defects: First, poor integration of multi-source data. Boiler, pipeline, and user data are stored in three independent systems with different formats (binary, Excel, JSON) and no unified time sequence alignment. When user room temperature is abnormal, manual data comparison is required, which takes more than 2 hours and makes fault location difficult. Second, insufficient real-time data analysis. It adopts offline batch processing every morning. Minor pipeline leaks (pressure drop of 0.08MPa in 4 hours) need to be detected the next day. During this period, heat is lost and user room temperature drops, resulting in a passive response. Third, lack of personalized control basis. It operates uniformly based on the average parameters of the community (such as boiler water temperature of 75℃), without considering the differences in heat loss of 13% to 16% for buildings facing the street and the difference in room temperature of 2 to 2.5℃ lower for top-floor buildings. This may lead to overheating of middle buildings and only 72% compliance rate for peripheral buildings, resulting in 9% gas waste. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method and system for analyzing the operation data of a heating system. Through multi-source data fusion and real-time analysis, a closed-loop intelligent operation is realized from accurate diagnosis of anomalies to demand prediction and regulation, thereby comprehensively improving the reliability, energy efficiency and comfort of the heating system.

[0005] To solve the above-mentioned technical problems, the technical solution of the present invention is as follows:

[0006] In a first aspect, a method for analyzing operational data of a heating system, the method comprising:

[0007] Real-time acquisition of equipment operation data, pipeline operation data, and user-side data; time-series alignment of the data based on a unified time base and feature dimensionality reduction to construct a multi-source fusion dataset;

[0008] Based on a multi-source fusion dataset, key performance indicators of each building's thermal inlet are calculated in real time. Based on these key performance indicators, multiple distributed reference nodes are dynamically determined through the pipeline topology and real-time hydraulic conditions. A hydraulic balance reference model is constructed based on the data from the reference nodes, and its state characteristics are extracted. Dynamic calibration coefficients are generated based on these state characteristics. The key performance indicators are calibrated using the dynamic calibration coefficients to obtain the calibrated key performance indicators.

[0009] The real-time values ​​of the calibrated key performance indicators are compared with the preset dynamic threshold range to obtain the indicator deviation state vector; based on the indicator deviation state vector, abnormal data clusters are identified to obtain diagnostic results containing the abnormality type and location.

[0010] If the diagnosis result is no abnormality, a differentiated operation strategy is generated based on the building's historical room temperature, building characteristics and real-time meteorological parameters in the multi-source fusion dataset.

[0011] Acquire meteorological forecast data for a preset time period; based on differentiated operation strategies and meteorological forecast data, predict changes in heat load demand through a long short-term memory network; based on changes in heat load demand, calculate the target output parameter setpoint with the goal of achieving room temperature standard and minimizing energy consumption, and send the setpoint to the boiler control system for execution.

[0012] Secondly, a system for analyzing operational data of a heating system includes:

[0013] The data acquisition module is used to collect equipment operation data, pipeline operation data, and user-side data in real time; it performs time-series alignment of the data based on a unified time benchmark and performs feature dimensionality reduction to construct a multi-source fusion dataset.

[0014] The calibration module is used to calculate the key performance indicators of each building's thermal inlet in real time based on a multi-source fusion dataset; based on the key performance indicators, it dynamically determines multiple distributed reference nodes through the pipeline topology and real-time hydraulic conditions; it constructs a hydraulic balance reference model based on the data of the reference nodes and extracts their state characteristics, and generates dynamic calibration coefficients based on these state characteristics; it uses the dynamic calibration coefficients to calibrate the key performance indicators and obtain the calibrated key performance indicators.

[0015] The identification module is used to compare the real-time values ​​of the calibrated key performance indicators with the preset dynamic threshold range to obtain the indicator deviation state vector; based on the indicator deviation state vector, abnormal data clusters are identified to obtain diagnostic results including the abnormality type and location.

[0016] The strategy generation module is used to generate differentiated operation strategies based on the building's historical room temperature, building characteristics, and real-time meteorological parameters in the multi-source fusion dataset if the diagnosis result is no abnormality.

[0017] The optimization control module is used to acquire meteorological forecast data for a preset time period; based on the differentiated operation strategy and meteorological forecast data, it predicts changes in heat load demand through a long short-term memory network; based on changes in heat load demand, with the goal of achieving room temperature standard and minimizing energy consumption, it calculates the target output parameter setpoint and sends the setpoint to the boiler control system for execution.

[0018] Thirdly, a computing device, comprising:

[0019] One or more processors;

[0020] A storage device for storing one or more programs that, when executed by one or more processors, cause the one or more processors to implement the method.

[0021] Fourthly, a computer-readable storage medium storing a program that, when executed by a processor, implements the method.

[0022] The above-described solution of the present invention has at least the following beneficial effects:

[0023] Real-time acquisition of multi-source data and time-series alignment based on a unified time base eliminates data time discrepancies; dimensionality reduction through principal component analysis reduces data redundancy, constructs a high-quality multi-source fusion dataset, and improves data integration and utilization efficiency; key performance indicators are calculated in real-time based on the fusion dataset, and reference nodes are dynamically determined by combining pipeline topology and hydraulic conditions, a hydraulic model is built to generate dynamic calibration coefficients, and indicators are calibrated to improve indicator accuracy; deviation vectors are obtained by comparing calibrated indicators with dynamic thresholds, and abnormal data clusters are identified by cluster analysis to determine the type and location of anomalies, improving the targeting of anomaly diagnosis; when no anomalies are found, historical room temperature, building characteristics, and real-time meteorological parameters of buildings in multi-source data are used to generate differentiated operation strategies through a pre-trained neural network, making the strategies fit the differences between buildings; meteorological forecast data is acquired, and combined with the strategies, long short-term memory networks are used to predict heat load changes, with the goal of achieving room temperature standards and minimizing energy consumption, and control parameters are calculated and sent to the system for execution, realizing data-driven dynamic regulation and improving system operational adaptability. Attached Figure Description

[0024] Figure 1 This is a flowchart illustrating a method for analyzing the operation data of a heating system according to an embodiment of the present invention.

[0025] Figure 2 This is a schematic diagram of an operation data analysis system for a heating system provided in an embodiment of the present invention. Detailed Implementation

[0026] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0027] like Figure 1 As shown in the figure, an embodiment of the present invention proposes a method for analyzing the operating data of a heating system, the method comprising the following steps:

[0028] Step 100: Collect equipment operation data, pipeline operation data, and user-side data in real time; perform time-series alignment of the data based on a unified time benchmark and perform feature dimensionality reduction to construct a multi-source fusion dataset;

[0029] Step 200: Based on the multi-source fusion dataset, calculate the key performance indicators of the thermal inlet of each building in real time; based on the key performance indicators, dynamically determine multiple distributed reference nodes through the pipeline topology and real-time hydraulic conditions; construct a hydraulic balance reference model based on the data of the reference nodes, extract its state characteristics, and generate dynamic calibration coefficients based on the state characteristics; use the dynamic calibration coefficients to calibrate the key performance indicators to obtain the calibrated key performance indicators.

[0030] Step 300: Compare the real-time values ​​of the calibrated key performance indicators with the preset dynamic threshold range to obtain the indicator deviation state vector; Based on the indicator deviation state vector, identify abnormal data clusters and obtain diagnostic results including abnormality type and location.

[0031] Step 400: If the diagnosis result is no abnormality, then generate a differentiated operation strategy based on the building's historical room temperature, building characteristics and real-time meteorological parameters in the multi-source fusion dataset.

[0032] Step 500: Obtain meteorological forecast data for a preset time period; based on the differentiated operation strategy and meteorological forecast data, predict changes in heat load demand through a long short-term memory network; based on changes in heat load demand, calculate the target output parameter setpoint with the goal of achieving room temperature standard and minimizing energy consumption, and send the setpoint to the boiler control system for execution.

[0033] In this embodiment of the invention, multiple types of data are collected in real time, and time-series alignment is achieved through a unified time base to eliminate time deviations from data from different sources. Principal component analysis is used for feature dimensionality reduction, which reduces data dimensions while retaining core information, thereby improving data processing efficiency. The constructed multi-source fusion dataset can integrate multi-dimensional information from the user side of the equipment pipeline network. Key performance indicators are calculated based on multi-source data, and reference nodes are dynamically determined by combining pipeline topology and real-time operating conditions, making the reference nodes more consistent with the actual operating state. A hydraulic balance model is constructed using reference node data, and dynamic calibration coefficients are generated to calibrate key indicators, improving their accuracy and relevance. The calibrated indicators are compared with dynamic thresholds to obtain deviation state vectors, quantifying the degree of deviation of the indicators. Cluster analysis identifies anomalous data clusters, aggregating anomalous data with similar characteristics and clearly presenting the clustering features of anomalies, facilitating accurate determination of the type and location of anomalies. Based on historical room temperature, building characteristics, and real-time meteorological parameters from multi-source data, a pre-trained neural network generates differentiated strategies, fully utilizing the correlations between multi-dimensional data to adapt the strategies to the building characteristics and heating needs of different buildings, achieving targeted regulation. Meteorological forecast data is acquired and combined with differentiated strategies to predict heat load changes through a long short-term memory network, effectively capturing the correlation between heat load and meteorological factors and time series. Output parameters are calculated with the goals of achieving room temperature standards and minimizing energy consumption, ensuring that parameter settings meet user heating needs while controlling energy consumption.

[0034] In a preferred embodiment of the present invention, step 100 includes:

[0035] Step 101 involves collecting equipment operation data, pipeline operation data, and user-side data in parallel from three independent data sources: the boiler control system, the property monitoring platform, and the user service system. Specifically, this includes: initiating a multi-source parallel data acquisition process, simultaneously connecting to these three independent data sources; collecting equipment operation data from the boiler control system, including boiler outlet water temperature, boiler operating pressure, instantaneous gas consumption, and boiler operating status parameters, maintaining the same acquisition frequency as the system's original data acquisition, i.e., acquiring data once per minute, and fully recording the corresponding acquisition time information; and collecting pipeline operation data from the property monitoring platform, specifically including data from each building. The supply and return water temperatures, pressures, and flow rates at the heating inlet are collected every 5 minutes on the platform, with the building number and collection time recorded synchronously. User-side data is collected from the user service system, specifically including indoor temperature data of users with installed room temperature data loggers and user complaints about excessively hot or cold rooms submitted via the APP. This data is collected every 30 minutes, with the building and floor information of the user recorded. During the collection process, the original data formats of each data source are initially identified and saved to ensure that binary boiler data, tabular network data, and JSON-formatted user-side data can be obtained completely.

[0036] Step 102: Based on a unified time base, timestamps are aligned for equipment operation data, pipeline operation data, and user-side data to obtain a timestamp-aligned data sequence. The timestamp-aligned data sequence is then normalized using a dynamic time warping algorithm to obtain an intermediate time series dataset with consistent time resolution. Specifically, this includes: first, establishing a unified time base, using millisecond-level precision standard timestamps as a unified reference; extracting the original acquisition time information from the equipment operation data, pipeline operation data, and user-side data collected from the three data sources; matching the original time information of each data set with the unified millisecond-level timestamps; correcting data with time discrepancies to ensure that all three types of data correspond to a unified time base, forming... The data sequence is timestamped and aligned. To address the time resolution differences between different data sources within the timestamped data sequence, a dynamic time warping algorithm is initiated. Using the time resolution of the equipment operation data per minute as the target resolution, interpolation is performed on the pipeline operation data every 5 minutes. The algorithm calculates the intermediate time value between two adjacent collection points, adjusting the pipeline data time interval to 1 minute. The user-side data is resampled every 30 minutes. Following the path matching rules of the dynamic time warping algorithm, the user-side data within 30 minutes is evenly distributed across the minute-level time nodes, ultimately resulting in an intermediate time-series dataset where all data has a uniform time resolution and time scale.

[0037] Step 103: Based on the intermediate time-series dataset, calculate its eigenvectors and eigenvalues ​​using principal component analysis to determine the main characteristic components and their contributions; filter the main characteristic components according to a preset variance contribution rate threshold to obtain the main feature subspace, specifically including: performing standardization preprocessing on all data in the intermediate time-series dataset to eliminate the influence of dimensional differences between different types of data (such as temperature, pressure, and flow rate), ensuring that all types of data are at the same numerical level; based on the standardized intermediate time-series dataset, apply principal component analysis to calculate the variance contribution rate of all characteristic variables in the dataset. The covariance matrix is ​​then used to solve for the eigenvectors and corresponding eigenvalues ​​of the covariance matrix through matrix operations. Based on the magnitude of the calculated eigenvalues, the contribution of each eigenvector to the original data information is determined; the larger the eigenvalue, the higher the contribution of the corresponding eigencomponent. A pre-set variance contribution rate threshold is set, such as a cumulative variance contribution rate of no less than 85%. Each eigencomponent is sorted in descending order of contribution, and the variance contribution rate of each eigencomponent is gradually accumulated. When the accumulated variance contribution rate reaches the preset threshold, the selection of eigencomponents is stopped, and the selected eigencomponents are used to form the main feature subspace.

[0038] Step 104 involves performing a geometric projection transformation on the feature data in the main feature subspace, mapping it to a low-dimensional feature space to obtain dimensionality-reduced feature projection data. Specifically, this includes: determining the direction of the feature vectors corresponding to each feature component in the main feature subspace, using these feature vectors as coordinate axes in the low-dimensional feature space to construct a coordinate system; extracting all feature data from the main feature subspace, treating each feature data point as a data point in a high-dimensional space; projecting each data point in the high-dimensional space into the low-dimensional feature space composed of the coordinate axes of the main feature components through a geometric projection transformation operation, maintaining the relative positional relationship between data points consistent with the original high-dimensional space during the projection process to ensure that the data distribution characteristics are not lost; verifying the projected data to check whether the data completely retains the core information in the main feature subspace; if information is missing, adjusting the projection transformation parameters and reprojecting; and obtaining the dimensionality-reduced feature projection data after successful verification.

[0039] Step 105: Based on the dimensionality-reduced feature projection data, integrate the feature information of multi-source data streams using a feature weighted fusion method to construct a multi-source fusion dataset with temporal consistency, feature orthogonality, and elimination of multicollinearity. Specifically, this includes: performing feature importance analysis on the dimensionality-reduced feature projection data; combining this with the heating system's operational logic to determine the degree of influence of different data source features on the system's operational status assessment. For example, boiler outlet water temperature features in equipment operation data, supply and return water pressure difference features in pipeline operation data, and indoor temperature features in user-side data are more critical to system analysis and are assigned higher weights, while other auxiliary features are assigned relatively lower weights; based on the determined feature weights, initiate feature weighted fusion. The process involves weighting the feature projection data corresponding to equipment operation data, pipeline operation data, and user-side data according to their weights to obtain preliminary fused feature data. A correlation test is then performed on the preliminary fused feature data, calculating the linear correlation coefficient between any two features. If the correlation coefficient exceeds a preset threshold (e.g., 0.8), multicollinearity is identified, and features with lower contributions are removed. This correlation test and feature removal process is repeated until the linear correlation coefficient between all remaining features is below the preset threshold. The resulting multi-source fused dataset maintains consistency across all data over time, achieves orthogonality between features, and eliminates the interference of multicollinearity on subsequent analysis.

[0040] In a preferred embodiment of the present invention, step 200 includes:

[0041] Step 201: Based on the multi-source fusion dataset, calculate in real time the key performance indicators of instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate at the heating inlets of each building. Specifically, this includes: extracting real-time supply and return water pressure data, supply and return water temperature data, and actual flow rate data of each building's heating inlet from the multi-source fusion dataset. The supply and return water pressure and temperature data are derived from the pipeline operation data collected by the original property monitoring platform, and the actual flow rate data is the real-time monitoring data integrated after multi-source fusion. For the eight residential buildings in an old community in northern China, calculate the instantaneous supply and return water pressure difference at the heating inlet of each building, specifically the difference between the supply water pipe pressure and the return water pipe pressure at the same time. Calculate the instantaneous supply and return water pressure difference at the heating inlet of each building. Temperature difference is specifically calculated as the difference between the supply water pipe temperature and the return water pipe temperature at the building's heating inlet at the same time. The instantaneous flow deviation rate of each building's heating inlet is calculated as the difference between the actual flow rate and the design flow rate at the building's heating inlet at the same time, divided by the design flow rate. The design flow rate is predetermined based on the rated output of the 2.8MW gas-fired hot water boiler in the community and the heat load demand of each building. During the calculation process, the time resolution is kept consistent with the multi-source fusion dataset (i.e., calculated once per minute) to ensure real-time capture of changes in the operating parameters of each building's heating inlet, fully covering all heating inlets of the 8 residential buildings, forming a data sequence of instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate for each building per minute.

[0042] Step 202: Based on the key performance indicators, combined with the pipeline network topology and real-time hydraulic condition data, extract the characteristic parameters of the pipeline network nodes; standardize the key performance indicators and node characteristic parameters to generate a standardized feature matrix, specifically including: extracting key performance indicators (i.e., instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate at the heating inlet of each building) from the multi-source fusion dataset, and simultaneously retrieving the topology data of the heating pipeline network of the community, including the pipeline connection relationship of the heating inlets of the 8 residential buildings, the length and diameter parameters of each pipe section, the distribution of pipeline branches, and real-time hydraulic condition data, including the current boiler outlet pressure, total pipeline flow, and medium velocity in the pipes of each building; extracting the characteristic parameters of the pipeline network nodes based on the above data, specifically including the pipeline connection of the heating inlet nodes of each building. The number of nodes, their hierarchical location in the pipeline network, the resistance coefficient of the corresponding pipeline, and the distance between the node and the boiler heat source are all considered. The extracted key performance indicators and node characteristic parameters are standardized. The supply and return water pressure difference (range 0.1 to 0.5 MPa), supply and return water temperature difference (range 5 to 20℃), flow deviation rate (range -20% to 20%), and node characteristic parameters (such as pipeline resistance coefficient) are all mapped to a numerical range of 0 to 1 through linear transformation to eliminate the influence of dimensional differences between different parameters. The standardized key performance indicators and node characteristic parameters are then organized by rows and columns, with each row corresponding to a building's heating inlet node and each column corresponding to a standardized parameter, constructing a standardized feature matrix to ensure that the matrix covers complete standardized data for all heating inlet nodes in all eight buildings.

[0043] Step 203: Using the standardized feature matrix as input, calculate the weighted Euclidean distance between each data sample using a clustering analysis algorithm, and use the K-means clustering method to divide the pipeline network nodes into multiple clusters to obtain the cluster division results. Specifically, this includes: using the standardized feature matrix as input data, determining the weight coefficients of each parameter, where the supply and return water pressure difference and supply and return water temperature difference have a significant impact on the hydraulic state, and their weight coefficients are set to 0.3; the flow deviation rate weight coefficient is set to 0.2; the pipe resistance coefficient and node branch position weight coefficients among the node feature parameters are set to 0.15; and the weight coefficients of the remaining parameters are set to 0.05; and calculating the heat inlet of each building based on the set weight coefficients. The weighted Euclidean distance between node data samples is calculated by weighting the differences in each parameter dimension and then calculating the square root of the sum of squares. Based on the pipe network distribution characteristics of the 8 buildings in the community, the number of K-means clusters is preset to 3, corresponding to the street-facing buildings, the middle buildings, and the top-floor concentrated buildings, respectively. The centroids of the 3 clusters are initialized, and each data sample is assigned to the cluster containing the nearest centroid. The mean of all samples in each cluster is calculated as the new centroid. The process of sample assignment and centroid update is repeated until the change in centroid in two consecutive iterations is less than a preset threshold, such as 0.001. The iteration is then stopped, and the final cluster division result is obtained, determining the cluster to which the thermal inlet node of each building belongs.

[0044] Step 204: Based on the cluster division results and combined with the connectivity analysis of the pipeline network topology, select the representative node closest to the centroid in each cluster as the distributed reference node to complete the dynamic identification and determination of the reference node. Specifically, this includes: based on the cluster division results, retrieving the topology diagram of the heating pipeline network of the community, analyzing the pipeline connectivity of the heating inlet nodes of each building in each cluster, determining whether the nodes are in the same pipeline branch, whether there is a direct pipeline connection, and whether they share the same water supply riser, and excluding isolated nodes with poor connectivity to other nodes in the cluster; for each cluster, calculate... Calculate the Euclidean distance from all nodes in the cluster to the cluster centroid, and select the 3 to 4 nodes closest to the centroid. From these nearby nodes, and considering the representativeness requirements of the pipeline network operation, select nodes located at key positions in the cluster's pipeline network branches. For example, if a cluster contains 3 buildings facing the street, select the building's thermal inlet node located in the middle of the cluster's pipeline network branch to ensure that the node can reflect the hydraulic state of other nodes in the cluster. Complete the above screening for each cluster, and finally determine 2 to 3 distributed reference nodes, covering all pipeline areas where the 8 buildings in the community are located, thus completing the dynamic identification and determination of reference nodes.

[0045] Step 205: Collect real-time operating data from the distributed reference nodes; using the real-time operating data, establish a hydraulic balance reference model reflecting the overall hydraulic state through a multiple linear regression method. Specifically, this includes: initiating the data acquisition process; for the identified distributed reference nodes (covering the key area of ​​the pipe network where 8 residential buildings in an old community in northern China are located), collecting real-time operating data from each reference node at a frequency of once per minute. This includes the real-time temperature of the water supply pipe, the real-time temperature of the return pipe, the real-time pressure of the water supply pipe, the real-time pressure of the return pipe, the instantaneous flow rate of the medium in the pipe, and the real-time velocity of the medium in the pipe. During the acquisition process, the millisecond-level timestamp corresponding to each set of data is recorded synchronously to ensure that the timestamp is consistent with the unified time reference of the multi-source fusion dataset. Continuous acquisition is performed for 30 minutes to obtain 30 sets of complete real-time operating data.

[0046] Based on the rated output parameters of the 2.8MW gas-fired hot water boiler used in the community's heating system, and combined with the design heat load of each building (pre-calculated based on building area, insulation coefficient, and apartment layout), the design hydraulic parameters of each distributed reference node are determined. Specifically, these include a design supply and return water pressure difference of 0.3MPa (to ensure sufficient medium circulation power), a design supply and return water temperature difference of 15℃ (to match the heat exchange requirements under the boiler's rated output), and a design flow rate of 8m³ / h. 3 / h (determined based on the total flow rate for balanced heat load distribution across 8 buildings); a hydraulic balance reference model is constructed based on the real-time operating data of 30 sets of distributed reference nodes. First, the independent and dependent variables of the model are selected. The independent variable is the supply and return water pressure difference between each reference node (including the pressure difference between the supply pipes of adjacent reference nodes and the pressure difference between the supply and return water pipes of the same reference node itself). The dependent variable is the instantaneous flow rate of each reference node (directly reflecting the hydraulic output state of the node and is the core indicator for judging hydraulic balance). The correlation between the two is established using the multiple linear regression method. First, the independent and dependent variables in the 30 sets of data are sorted out, and abnormal data that exceed the normal range are removed (such as data where the instantaneous flow rate suddenly becomes 0 or the pressure difference suddenly rises or falls) to ensure the validity of the data. Then, based on the sorted data, a preliminary correlation framework between the independent and dependent variables is established, forming the basic structure of the model.

[0047] Based on the principle of hydraulic balance in pipe networks, and combined with the topology of the pipe network in this community (the connection relationships of each reference node, the direction of the pipes, and the branching situation), a set of equilibrium equations for a hydraulic balance reference model is constructed. The set of equations contains three core relationships: First, the inflow and outflow of each reference node are equal. For example, if a reference node is connected to two input pipes and one output pipe, then the sum of the real-time flow of the two input pipes must equal the real-time flow of the one output pipe to ensure the conservation of flow at a single node. Second, the linear relationship between pressure difference and flow between reference nodes. For each pair of adjacent reference nodes, a linear correlation equation is established between the change in supply and return water pressure difference and the change in flow within the pipe, reflecting the driving effect of pressure difference on flow. Third, the overall flow balance relationship of the pipe network, that is, the total real-time flow of the boiler outlet water supply pipe is equal to the sum of the instantaneous flow of all distributed reference nodes, ensuring the balance of flow in and out of the entire pipe network system. These three relationships together constitute the equation system of the model.

[0048] Subsequently, model training and parameter determination were carried out. Thirty sets of continuously collected real-time operational data were substituted one by one into the constructed equilibrium equation set. First, the preliminary values ​​of each coefficient in the equation set (such as the proportional coefficient of the linear relationship between pressure difference and flow rate, and the node flow distribution coefficient) were calculated for each set of data. Then, the theoretical flow rate calculated by the equation set after substituting each set of data was compared with the actual instantaneous flow rate value collected, and the relative deviation between the two was calculated (the difference between the theoretical value and the actual value divided by the actual value). The coefficient parameters in the equation set were adjusted according to the magnitude of the relative deviation. If the deviation exceeded the preset allowable range of 5%, the corresponding coefficient was increased or decreased to reduce the deviation. The process of data substitution, deviation calculation, and coefficient adjustment was repeated until the relative deviation between all theoretical flow rate values ​​and actual flow rate values ​​in the 30 sets of data was less than 5%. When the deviation met the requirements, the final values ​​of each coefficient in the equation set were determined. At this point, the correlation between the independent and dependent variables established by the multiple linear regression, together with the adjusted equilibrium equation set, was integrated into a complete hydraulic balance reference model. This model can accurately reflect the overall hydraulic state of the heating network in the community, completing the model construction and training.

[0049] Step 206: Extract the pressure distribution features and flow balance features from the hydraulic balance reference model as state features. Specifically, this includes: calling the established hydraulic balance reference model and extracting the pressure distribution features from the model, including the real-time supply and return water pressure values ​​of each distributed reference node, the supply water pressure difference and return water pressure difference between adjacent reference nodes, the pressure gradient from the reference node to the boiler heat source (i.e., the pressure difference divided by the pipe length between the node and the boiler), and the pressure fluctuation amplitude of the pipe network branch where each reference node is located; extracting the flow balance features from the model, specifically including the real-time flow values ​​of each distributed reference node, the proportion of the flow of each reference node to the total flow of the pipe network, the flow distribution ratio of each reference node within the same pipe network branch, and the flow trend data of the reference node over time; organizing the extracted pressure distribution features and flow balance features to ensure that each feature corresponds to a specific value or data sequence and can reflect the overall hydraulic operation status of the heating pipe network of the community.

[0050] Step 207: Perform a correlation analysis between the state characteristics and the standard reference values ​​to obtain the correlation analysis results; based on the correlation analysis results, fit the mathematical relationship between the state characteristics and the reference values ​​using the least squares method to obtain the fitting results. Specifically, this includes: determining the standard reference values ​​for the state characteristics according to the design specifications and operating standards of the heating system in this community, wherein the standard reference values ​​for pressure distribution characteristics include a design supply and return water pressure difference of 0.3 MPa, a pressure gradient between adjacent nodes of 0.02 MPa / m, and a pressure fluctuation amplitude not exceeding ±5%; and the standard reference values ​​for flow balance characteristics include a design flow rate of 8 m³ / m³. 3 / h, the reference node traffic share is 12.5% ​​(averaged across 8 buildings), and the traffic distribution ratio deviation does not exceed ±10%; the real-time status feature data extracted in step 206 is correlated with the above standard reference value, and the difference, ratio and correlation coefficient between each real-time status feature and the corresponding standard reference value are calculated to determine the numerical correlation between the two; based on the data obtained from the correlation analysis (including the corresponding data of multiple sets of real-time status features and standard reference values), with real-time status features as independent variables and standard reference values ​​as dependent variables, the least squares method is used for fitting; the coefficients of the fitting relationship are determined through iterative calculation to minimize the sum of squared errors between the predicted value calculated by the fitting relationship and the standard reference value, and the final fitting result is obtained, that is, the mathematical relationship expression between the real-time status features and the standard reference value.

[0051] Step 208: Based on the fitting results, the weight coefficients of each state feature are obtained; dynamic calibration coefficients are generated based on the weight coefficients, specifically including: based on the fitting results obtained in step 207, i.e., the mathematical relational expression, the coefficients corresponding to each real-time state feature in the expression are extracted, and these coefficients are used as the weight coefficients of each state feature. Among them, the weight coefficients corresponding to state features that have a greater impact on the standard reference value (such as the pressure gradient of adjacent nodes and the flow ratio of the reference node) are higher, and the weight coefficients corresponding to features that have a smaller impact (such as the pressure fluctuation amplitude and the flow change trend) are lower; according to the degree of deviation between the real-time monitoring value and the standard reference value of each state feature, the deviation coefficient is calculated, i.e., (real-time value - standard value) / standard value; the weight coefficient of each state feature is multiplied by the corresponding deviation coefficient to obtain the local calibration coefficient corresponding to the state feature; the local calibration coefficients of all state features are weighted and summed to generate the final dynamic calibration coefficient. The key performance indicators of each building's heating inlet correspond to a dynamic calibration coefficient that matches the state of its own pipe network area, ensuring that the calibration coefficient can be adjusted with the real-time hydraulic state of the pipe network.

[0052] Step 209: Apply the dynamic calibration coefficient to the key performance indicators (KPIs) for dynamic calibration and correction to obtain calibrated KPIs. Specifically, this includes: retrieving the instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate data for each building's heating inlet, and simultaneously retrieving the corresponding dynamic calibration coefficients for each building; multiplying the instantaneous supply and return water pressure difference value for each building by the corresponding dynamic calibration coefficient to obtain the calibrated instantaneous supply and return water pressure difference; multiplying the instantaneous supply and return water temperature difference value for each building by the corresponding dynamic calibration coefficient to obtain the calibrated instantaneous supply and return water temperature difference; multiplying the instantaneous flow deviation rate value for each building by the corresponding dynamic calibration coefficient to obtain the calibrated instantaneous flow deviation rate; during the calibration process, synchronously recording the numerical changes before and after calibration and the corresponding timestamps to ensure data traceability; after completing the calibration of the KPIs for all 8 buildings' heating inlets, compiling them into a calibrated KPI dataset.

[0053] In this embodiment of the invention, the instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate at the building's heating inlet are calculated in real time, providing key data support, capturing dynamic changes in operating status, and ensuring that the data reflects the current real situation. Node features are extracted by combining the pipe network topology and hydraulic conditions to fit the actual scenario. Standardization processing eliminates dimensional differences, generating a regular feature matrix, improving the efficiency and accuracy of subsequent algorithm processing. Weighted Euclidean distance highlights the influence of important features, and K-means clustering divides nodes into clusters based on data similarity, reducing redundancy in indiscriminate analysis and providing a basis for selecting reference nodes. Node selection is combined with pipe network connectivity, conforming to the physical structure, and representative nodes near the centroid of the cluster are selected. The system dynamically adapts to changes in operating conditions; it uses reference node data for modeling to reduce computational load; it establishes a hydraulic balance model using multiple linear regression to comprehensively characterize the overall hydraulic state; it extracts core state features of pressure and flow from the model, eliminates irrelevant interference, and provides accurate basis for subsequent correlation analysis and calibration coefficient generation; the correlation analysis quantifies the relationship between state features and standard values, and the least squares method fits the mathematical relationship, laying the foundation for weight calculation; the weight coefficients are calculated to reflect the degree of influence of features on hydraulic balance, generating dynamic calibration coefficients to adapt to the actual hydraulic state; the dynamic calibration coefficients correct key performance indicators, adapt to the actual state of the pipeline network, improve the accuracy of indicators, and provide reliable data support for subsequent anomaly diagnosis.

[0054] In a preferred embodiment of the present invention, step 300 includes:

[0055] Step 301: Based on the calibrated key performance indicators, compare their real-time values ​​with preset dynamic threshold ranges item by item, calculate the relative deviation value of each indicator, and obtain the indicator deviation state vector. Specifically, this includes: extracting the instantaneous supply and return water pressure difference, instantaneous supply and return water temperature difference, and instantaneous flow deviation rate data per minute for each of the eight residential buildings in an old residential community in northern China from the calibrated key performance indicator dataset. The calibrated range for the supply and return water pressure difference is 0.12 to 0.48 MPa, the calibrated range for the supply and return water temperature difference is 6 to 19℃, and the calibrated range for the flow deviation rate is -14% to 14%. The preset dynamic threshold ranges for each indicator are determined based on the design parameters and historical normal operation data of the 2.8MW gas-fired hot water boiler in the community. The threshold range for the supply and return water pressure difference is 0.15 to 0.45 MPa, the threshold range for the supply and return water temperature difference is 7 to 18℃, and the threshold range for the flow deviation rate is... The range is from -12% to 12%. For each building, the three calibrated indicators per minute are compared with the corresponding dynamic threshold ranges to determine whether the real-time values ​​are within the threshold ranges. The relative deviation values ​​of each indicator are calculated. The relative deviation value of the supply and return water pressure difference is (real-time calibration value - median of the threshold range) divided by the median of the threshold range. The relative deviation value of the supply and return water temperature difference is calculated in the same way. The relative deviation value of the flow rate deviation rate is (real-time calibration value - median of the threshold range) divided by the median of the threshold range. The median values ​​of the threshold ranges are 0.3 MPa for the supply and return water pressure difference, 12.5℃ for the supply and return water temperature difference, and 0% for the flow rate deviation rate. The relative deviation values ​​of the supply and return water pressure difference, supply and return water temperature difference, and flow rate deviation rate per minute for each building are arranged in order to form the indicator deviation state vector of that building at that moment. This covers the data of all 8 buildings at all times, ensuring that each data sample corresponds to a unique vector.

[0056] Step 302: Use the index deviation state vector as input data; calculate the reachability distance between each data sample point by performing a density clustering algorithm on the input data; identify high-density areas and outliers based on the distribution characteristics of the reachability distance to obtain the data point classification results; based on the data point classification results, form a preliminary abnormal data cluster, specifically including: collecting all index deviation state vectors generated in step 301 and using them as input data for the density clustering algorithm. The data samples cover the vectors corresponding to each minute of 8 buildings over a continuous 24-hour period (a total of 24×60×8=11520 samples); preprocessing the input data to remove null samples caused by data collection interruptions, ensuring that the remaining samples all contain the three complete relative deviation values; executing the density clustering algorithm, first setting the algorithm parameters, the neighborhood radius is determined to be 0.08 based on the average distance between samples, and the minimum number of samples is set to 5 (i.e., a region must contain at least 5 samples to be considered high-density). (Degree region); calculate the reachability distance between each data sample point and all other sample points, that is, the farthest distance that a sample point can indirectly reach through other samples in its neighborhood; divide the region according to the distribution characteristics of the reachability distance, identify the region with the reachability distance less than the neighborhood radius and the number of samples exceeding the minimum number of samples as high-density region, and identify the sample with the reachability distance greater than the neighborhood radius and the number of surrounding samples less than the minimum number of samples as outlier point; based on the above identification results, obtain the data point classification results, the samples in the high-density region are classified into the normal data point set, and the outlier point and a small number of surrounding samples are classified into the potential abnormal data point set; group the potential abnormal data point set according to the building corresponding to the sample and the time correlation, group the potential abnormal samples in the same building within 5 consecutive minutes into one group, and group the potential abnormal samples in different buildings but in the same pipeline branch and with similar time into another group, forming a preliminary abnormal data cluster, each cluster containing the number of samples, the corresponding building number, and the time range information.

[0057] Step 303: Based on the spatial distribution characteristics of the preliminary abnormal data clusters and combined with the pipeline network topology, the abnormal data clusters are merged and optimized to determine the final abnormal data cluster classification, specifically including:

[0058] Extract the spatial distribution characteristics of each preliminary abnormal data cluster, and count the number of buildings covered by each cluster, the specific building number, and the corresponding floor information. For example, a preliminary cluster covers Building 1 (floors 1 to 6) and Building 2 (floors 1 to 6), a preliminary cluster only covers one sample point on the 5th floor of Building 3, and a preliminary cluster covers Building 5 (floors 2 to 5) and Building 6 (floors 1 to 4). Retrieve the topological structure data of the heating network of the community, including the connection relationship between the heating inlets of the 8 buildings and the water supply risers and branch pipes, as well as the pipe routing and length parameters. It was determined that Building 1 and Building 2 share a water supply riser (riser number L1), Building 3 is independently connected to a water supply branch (branch number F1), Building 4 is independently connected to a water supply branch (branch number F2), Building 5 and Building 6 share a water supply riser (riser number L2), and Building 7 and Building 8 share a water supply riser (riser number L3). Based on this data, a pipe network topology diagram was constructed, with the building heating inlets, water supply risers, and branch pipes as nodes in the diagram, and the pipe connections as edges between nodes.

[0059] Map the building heating inlet node corresponding to each preliminary abnormal data cluster to the pipe network topology diagram to determine the specific location of each preliminary cluster in the topology diagram and the associated riser / branch nodes; start the spatial topology clustering process to calculate the topological distance between each preliminary cluster. This distance is defined as the shortest path length (in terms of pipe segments) of the corresponding nodes of two preliminary clusters connected by pipes in the topology diagram. For example, the nodes of the preliminary cluster covering Building 1 and the preliminary cluster covering Building 2 are both connected to riser L1, and the topological distance is 1 pipe segment (riser segment). The nodes of the preliminary cluster covering Building 3 and the preliminary cluster covering Building 4 belong to different branches, and the topological distance is 3 pipe segments (F1 branch segment + main pipe segment + F2 branch segment).

[0060] Combining the pipeline network topology, a topological connectivity analysis is performed on the preliminary abnormal data clusters to determine whether all building heat inlet nodes within the same preliminary cluster belong to the topological subgraph corresponding to the same riser or branch. For example, the nodes of the preliminary cluster covering buildings 5 ​​and 6 all belong to the topological subgraph corresponding to riser L2, and are therefore determined to be topologically connected. It is also determined whether the nodes of different preliminary clusters are in the same topological subgraph and the topological distance is less than a preset threshold (set to 2 pipe segments). For example, the preliminary cluster covering building 1 and the preliminary cluster covering building 2 are in the same topological subgraph and the topological distance is 1, which meets the threshold requirement.

[0061] Preliminary anomalous data clusters that meet the topological connectivity requirements and have overlapping time ranges (overlap duration greater than or equal to 5 minutes) are merged. For example, the preliminary cluster covering Building 1 (time 8:00 to 8:10) and the preliminary cluster covering Building 2 (time 8:02 to 8:11) are merged into a new cluster because they are in the same topological subgraph, have a topological distance of 1 segment of pipeline, and have a time overlap of 9 minutes. Preliminary clusters with a topological distance greater than the threshold, or are in different topological subgraphs, or have no adjacent topological nodes for support are filtered. If a preliminary cluster contains only a single sample point and its corresponding node... If a point has no adjacent abnormal nodes in the topology graph (neither adjacent nodes have abnormal data), such as a preliminary cluster that only covers the 5th floor of Building 3, its corresponding node has no other abnormal nodes in branch F1, and is therefore determined to be a topologically isolated cluster and removed. If a preliminary cluster contains multiple sample points but is in an independent topology subgraph and has no associated abnormalities, such as a preliminary cluster that only covers Building 4 (5 sample points), it is also determined to be a topologically isolated cluster and removed because branch F2 has no other abnormal nodes and is far from other branches in the topology. Combined with the historical fault data of this community (branch F2 rarely has concentrated abnormalities), it is also determined to be a topologically isolated cluster and removed.

[0062] After merging and eliminating data, the final abnormal data cluster classification is determined. Each final cluster needs to determine the building range and floors covered, the corresponding riser / branch (topology affiliation), the specific time interval, the number of samples, and the information of the topology-related nodes. For example, final cluster 1 covers building 1 (floors 1 to 6) and building 2 (floors 1 to 6), corresponding riser L1, time 8:00 to 8:11, 22 samples, and the topology-related nodes are riser L1 and the pipes connected at both ends; final cluster 2 covers building 5 (floors 2 to 5) and building 6 (floors 1 to 4), corresponding riser L2, time 14:30 to 14:40, 20 samples, and the topology-related nodes are riser L2 and the pipes connected at both ends.

[0063] Step 304: Based on the final abnormal data cluster classification, extract the statistical characteristic parameters of each abnormal data cluster, including the average deviation of the index within the cluster, spatial distribution density, and duration characteristics. Specifically, this includes: for each final abnormal data cluster, extracting the index deviation data of all samples within the cluster, namely, the relative deviation values ​​of the supply and return water pressure difference, the relative deviation values ​​of the supply and return water temperature difference, and the relative deviation values ​​of the flow rate deviation rate; calculating the average deviation of the index within the cluster, specifically calculating the arithmetic mean of the relative deviation values ​​of the supply and return water pressure difference, the relative deviation values ​​of the supply and return water temperature difference, and the relative deviation values ​​of the flow rate deviation rate for all samples within the cluster. For example, the average relative deviation of the supply and return water pressure difference for final cluster 1 is -0.2, the average relative deviation of the supply and return water temperature difference is 0.15, and the average relative deviation of the flow rate deviation rate is -0.3; calculating the cluster... The spatial distribution density is calculated by determining the proportion of buildings covered by the final cluster to the total number of buildings in the community (8 buildings), and the proportion of floors covered to the total number of floors in the corresponding buildings (6 floors). The average of these two values ​​is taken as the spatial distribution density. For example, if final cluster 1 covers 2 buildings and each building covers 6 floors, the spatial distribution density is (2 / 8 + 6 / 6) / 2 = 0.625. The duration feature of the cluster is extracted by determining the time difference between the first and last occurrence of the sample in the final cluster. For example, if the first sample time of final cluster 1 is 8:00 and the last sample time is 8:10, the duration feature is 10 minutes. The average deviation of the intra-cluster index, spatial distribution density, and duration feature of each final abnormal data cluster are compiled into a set of statistical feature parameters for that cluster, ensuring that each parameter corresponds to a specific value.

[0064] Step 305: Based on the statistical feature parameters, a pre-defined anomaly type discrimination rule is matched using a rule-based reasoning method to determine the anomaly type and its location, obtaining a diagnostic result including the anomaly type and location. Specifically, this includes: pre-defined anomaly type discrimination rules, formulated based on common heating fault types and historical fault data in the community. The rules include: for localized pipe network blockage, the rule is that the average relative deviation of the flow rate deviation rate within the cluster is less than or equal to -0.2, the average relative deviation of the supply and return water pressure difference is less than or equal to -0.15, and the duration is greater than or equal to 5 minutes; for minor pipe network leakage, the rule is that the average relative deviation of the supply and return water pressure difference within the cluster is less than or equal to -0.2, the average relative deviation of the supply and return water temperature difference is greater than or equal to 0.1, and the duration is greater than or equal to 8 minutes; for decreased boiler thermal efficiency, the rule is that the average relative deviation of the supply and return water temperature difference in the final cluster of all buildings is less than or equal to -0.1, and the average relative deviation of the flow rate deviation rate is between -0.05 and 0.05. The process involves retrieving the statistical characteristic parameters of each final abnormal data cluster obtained in step 304, and matching the parameters of each cluster with preset rules one by one. For example, the statistical characteristic parameters of final cluster 1 are: average relative deviation of flow deviation rate -0.3, average relative deviation of supply and return water pressure difference -0.2, and duration of 10 minutes, which meet the discrimination rules for local blockage of the pipeline network. Therefore, the abnormality type corresponding to this cluster is determined to be local blockage of the pipeline network. Based on the building range covered by the final abnormal data cluster and the corresponding pipeline network topology, the location of the abnormality is determined. For example, if final cluster 1 covers buildings 1 and 2 and corresponds to the water supply risers of buildings 1 and 2, the location of the abnormality is determined to be the shared water supply riser of buildings 1 and 2 and the heat inlet of the two buildings. The abnormality type and location corresponding to each final abnormal data cluster are compiled and summarized to form a diagnostic result containing all abnormal information. The abnormality type, buildings involved, specific pipeline network location, and abnormal duration are determined and labeled to provide a basis for subsequent fault handling.

[0065] In a preferred embodiment of the present invention, step 400 includes:

[0066] Step 401: If the diagnostic result is no abnormality, extract the historical room temperature, building characteristics, and real-time meteorological parameters of each building from the multi-source fusion dataset. Specifically, this includes: first, obtaining the diagnostic result output in step 305; if the result shows that the heating system of 8 residential buildings in an old community in northern China is normal (i.e., no pipe blockage, leakage, or other faults), then initiating the data extraction process; extracting the historical room temperature data of each building from the multi-source fusion dataset, specifically the indoor temperature data of 10% of residents collected every 30 minutes over the past 30 days (summarized by building, such as the average, highest, and lowest room temperature values ​​of Building 1 from 6:00 to 22:00 daily), covering the complete historical room temperature sequence of each of the 8 buildings; extracting the building characteristic data of each building, including... The following parameters were used to determine the building's construction date (all built around 2000), external wall insulation coefficient (0.6 W / (㎡·K) for buildings 1-2 and 7-8 facing the street, and 0.8 W / (㎡·K) for buildings 3-6 in the middle), street facing status (buildings 1-2 and 7-8 facing the street, buildings 3-6 in the middle), number of floors (all 6 floors), and unit size (average 80㎡ per unit). Real-time meteorological parameters were extracted, including current outdoor temperature (e.g., -8℃), outdoor wind speed (e.g., 2.5 m / s), and solar radiation intensity (e.g., 200 W / ㎡). The data came from real-time data from a local meteorological platform that was centrally integrated from multiple sources, ensuring that the three types of data extracted corresponded one-to-one with each building without data confusion.

[0067] Step 402 involves combining the historical room temperature, building characteristics, and real-time meteorological parameters of the buildings to construct a feature vector for each building. Specifically, this includes: preprocessing the extracted data for each building; calculating the average daily room temperature over the past 7 days, the diurnal temperature fluctuation range (the difference between the highest and lowest daily values), and the average room temperature during the low-temperature period (6:00 to 8:00); converting building characteristic data into quantifiable parameters, with street-facing attributes represented by 1 (buildings 1 to 2, 7 to 8), and 0 representing intermediate buildings (buildings 3 to 6); retaining the original insulation coefficient value; and uniformly assigning 6 floors to the current real-time meteorological parameters. The raw values ​​of outdoor temperature and wind speed, and the average solar radiation intensity over the past 2 hours are used. A feature vector is constructed for each building according to historical room temperature characteristics, building characteristic parameters, and real-time meteorological parameters. For example, the feature vector for Building 1 (street-facing) includes: average daily room temperature over the past 7 days (18.5℃), diurnal fluctuation range (3.2℃), average low temperature period (16.8℃), insulation coefficient (0.6W / (㎡·K), street-facing sign (1), current outdoor temperature (-8℃), current wind speed (2.5m / s), and average solar radiation over the past 2 hours (200W / ㎡). Each of the 8 buildings has its own independent feature vector, ensuring that each vector contains 8 dimensions of feature data with no missing or redundant dimensions.

[0068] Step 403 involves inputting the feature vectors into a pre-trained neural network and obtaining the initial heating demand prediction values ​​for each building through forward propagation calculations. Specifically, this includes: firstly, constructing and training a pre-trained neural network model. The model construction phase is based on the operational data of an old residential community in northern China over the past two heating seasons, collecting daily building feature vectors and actual heating demand data for eight buildings during this period. The dimensions of the building feature vectors are consistent with those of the vectors constructed in Step 402, including eight dimensions: average daily room temperature over the past seven days, diurnal room temperature fluctuation range, average room temperature during low-temperature periods, insulation coefficient, street-facing signs, average daily outdoor temperature, average daily wind speed, and average daily solar radiation. The actual heating demand data is the actual heat consumption (unit: kW) per hour for each building, calculated using boiler gas consumption, supply and return water temperature difference, and flow rate data to ensure that each feature vector corresponds to a unique actual heat consumption label.

[0069] The collected sample data were preprocessed to remove abnormal heat consumption data caused by equipment failure (such as data where heat consumption suddenly increased to more than 3 times the normal range). Missing feature values ​​were filled with the average value of the same building and the same time period. The preprocessed samples were divided into training set and validation set in a 7:3 ratio. The training set was used for model parameter learning, and the validation set was used to monitor the model's generalization ability.

[0070] The neural network model structure is designed with 8 neurons in the input layer to match the dimension of the feature vector; 3 hidden layers are set, with 64 neurons in the first layer, 32 in the second layer, and 16 in the third layer. Each hidden layer uses the ReLU activation function for feature transformation to capture nonlinear features; the output layer has 1 neuron, and the output value is the predicted value of the building's initial heating demand (in kW). A linear activation function is used to ensure that the output value conforms to the numerical range of heat consumption.

[0071] The model training process is initiated, with mean squared error chosen as the loss function to measure the difference between the model's predicted values ​​and the actual heat consumption. A gradient descent optimizer is employed, with an appropriate learning rate set (initially 0.001, gradually decreasing with each training iteration). Training set samples are input into the model in batches, with each batch containing 32 samples. Predicted values ​​are calculated via forward propagation, and the weights and bias parameters of neurons in each layer are updated via backpropagation. After each training round (traversing all training set samples), the validation loss of the model is calculated using the validation set. If the validation loss does not decrease for five consecutive rounds, training is stopped to avoid overfitting. After training, the optimal parameters of the model are saved, forming a pre-trained neural network model that can be directly used.

[0072] The pre-trained neural network model, which has been trained as described above, is retrieved. The feature vectors of the eight buildings constructed in step 402 are input into the model one by one, and the forward propagation calculation is started. The feature vectors first pass through the input layer into the first hidden layer. After being transformed by the ReLU activation function, high-dimensional features are extracted. Then, they are passed to the second hidden layer to further compress the feature dimension, and then to the third hidden layer to complete the final feature extraction. Finally, the initial heating demand prediction value of each building is output through the output layer, in kW. For example, the initial prediction value of Building 1 (facing the street) is 120kW, the initial prediction value of Building 3 (middle) is 95kW, and the initial prediction value of Building 5 (middle) is 100kW. The entire calculation process ensures that the prediction value of each building is generated based on its own unique feature vector, which fully reflects the differences in heating demand caused by different buildings due to building characteristics, historical heating conditions, and real-time weather differences.

[0073] Step 404: Based on the pipeline network topology, perform hydraulic coupling analysis on the initial heating demand forecasts for each building, and calculate the heat loss correction during pipeline transmission. Specifically, this includes: retrieving the topology data of the heating pipeline network for the community, and determining the connection relationship between the 8 buildings and the boiler heat source: Buildings 1 and 2 share a single water supply riser (approximately 80m from the boiler, pipe diameter DN100); Building 3 has an independent water supply pipe (approximately 60m from the boiler, diameter DN80); Building 4 has an independent water supply pipe (approximately 65m from the boiler, diameter DN80); Buildings 5 ​​and 6 share a single water supply riser (approximately 50m from the boiler, diameter DN100); Buildings 7 and 8 share a single water supply riser (approximately 90m from the boiler, diameter DN100). Based on the topology, hydraulic coupling analysis is conducted to calculate the heat loss of the water supply pipelines in each building. The heat loss calculation takes into account the pipeline length (the longer the pipeline, the greater the loss), diameter (the smaller the diameter, the greater the loss), and outdoor ambient temperature (the lower the outdoor ambient temperature, the greater the loss). For example, the heat loss rate of the water supply pipeline in Building 1 is 8%, in Building 3 it is 6%, and in Building 7 it is 9%. Based on the initial heating demand forecast and the heat loss rate, the heat loss correction amount for each building is calculated. The correction amount = initial forecast value × heat loss rate. For example, the correction amount for Building 1 = 120kW × 8% = 9.6kW, the correction amount for Building 3 = 95kW × 6% = 5.7kW, and the correction amount for Building 7 = 110kW × 9% = 9.9kW, ensuring that the correction amount closely matches the actual transmission loss of the pipeline network.

[0074] Step 405 involves superimposing the initial heating demand forecast value with the heat loss correction amount to calculate the differentiated heating demand target value for each building. Specifically, for each building, the initial heating demand forecast value obtained in step 403 is superimposed with the heat loss correction amount obtained in step 404. The superposition formula is: Differentiated heating demand target value = Initial forecast value + Heat loss correction amount. The calculation is completed for each of the eight buildings. For example, the target value for building 1 is 120kW + 9.6kW = 129.6kW, the target value for building 3 is 95kW + 5.7kW = 100.7kW, the target value for building 5 is 100kW + 6.2kW = 106.2kW, and the target value for building 7 is 110kW + 9.9kW = 119.9kW. After the calculation is completed, a list of differentiated heating demand target values ​​for the eight buildings is compiled, with each building number and corresponding target value marked to ensure that the target value includes both its own heating demand and pipeline transmission losses, thus conforming to actual operating needs.

[0075] Step 406: Based on the differentiated heating demand target values ​​for each building, calculate the corresponding target water supply temperature setpoints to form a differentiated operation strategy. Specifically, this includes: establishing a correlation between the differentiated heating demand target values ​​and the target water supply temperature based on the rated parameters of the 2.8MW gas-fired hot water boiler in the community (rated outlet water temperature 85℃, rated return water temperature 60℃, rated heat load 2800kW). That is, based on the proportion of the target value to the total boiler heat load, combined with the pipe network return water temperature (set at the historical average of 58℃), calculate the target water supply temperature for the corresponding building; for example, Building 1's target demand of 129.6kW accounts for approximately 4.63% of the total load. The calculated target water supply temperature is 78℃; Building 3 has a target demand of 100.7kW, accounting for approximately 3.59%, with a target water supply temperature of 72℃; Building 5 has a target demand of 106.2kW, accounting for approximately 3.79%, with a target water supply temperature of 73℃; Building 7 has a target demand of 119.9kW, accounting for approximately 4.28%, with a target water supply temperature of 76℃. The numbers of the eight buildings and their corresponding target water supply temperatures are compiled into a table to determine the water supply temperature setpoint that needs to be implemented at the heat inlet of each building. At the same time, the strategy execution cycle is marked (fine-tuned once per hour based on real-time meteorological parameters), forming a complete differentiated operation strategy that can be directly used to guide the control and operation of boilers and pipe networks.

[0076] In this embodiment of the invention, when the diagnostic results are normal, the data extraction process is initiated to avoid unnecessary resource consumption. Simultaneously, three key information categories—historical room temperature, building characteristics, and real-time weather—are extracted from the multi-source fusion dataset to provide comprehensive and building-related data support for subsequent processing. These three types of data are combined into structured feature vectors specific to each building to avoid data confusion and improve processing standardization and targeting. The specific feature vectors are input into a pre-trained neural network, which uses existing model patterns for rapid calculation. Initial heating demand predictions are generated through forward propagation, reflecting individual building differences. Combining network topology analysis with building location and pipe connections, heat loss corrections are calculated to reflect the differences in losses among different buildings, improving accuracy. The initial predictions and corrections are superimposed to integrate theoretical demand with actual losses, making the differentiated heating demand target values ​​closer to reality and improving practicality. The target water supply temperature for each building is calculated based on the target values, making the strategy implementable. The temperature settings are aligned with the actual needs of the buildings and can directly guide boiler and network regulation, improving the strategy's operability.

[0077] In a preferred embodiment of the present invention, step 500 includes:

[0078] Step 501: Obtain meteorological forecast data for a future preset time period, including outdoor temperature, wind speed, and solar radiation intensity parameters. Specifically, this includes: determining the future preset time period as 24 hours, setting the time resolution to once per hour to ensure real-time tracking of the impact of meteorological changes on heat load; obtaining the meteorological forecast data for the area where the old northern residential community is located for the next 24 hours through the open data interface of the local meteorological platform; the outdoor temperature forecast data covers the range of -15℃ to -8℃, with one forecast value recorded every hour, such as -10℃ at 0:00, -11℃ at 1:00, -15℃ at 6:00, and -8℃ at 14:00; the outdoor wind speed forecast data covers 1.5m / The wind speed is predicted to be slightly higher in the area of ​​buildings facing the street (Buildings 1 to 2 and Buildings 7 to 8) within the range of 3.0 m / s, with the hourly predicted value being 0.3 to 0.5 m / s higher than that of the middle buildings (Buildings 3 to 6). The predicted solar radiation intensity covers the range of 50 W / m² to 300 W / m², with non-zero values ​​only between 6:00 and 18:00, reaching a peak of 300 W / m² around 12:00. The three types of meteorological parameters are organized in chronological order to form structured data for each hour. Each data entry includes a timestamp, outdoor temperature, corresponding area wind speed, and solar radiation intensity, ensuring that the data fully covers the next 24 hours and matches the distribution of buildings in the community.

[0079] Step 502: Based on the differentiated operation strategy and the meteorological forecast data, construct time-series input features, specifically including: extracting the target water supply temperature setpoints for each building from the differentiated operation strategy, wherein the target water supply temperature for street-facing buildings (Buildings 1 to 2, Buildings 7 to 8) is 76℃ to 78℃, and the target water supply temperature for intermediate buildings (Buildings 3 to 6) is 72℃ to 73℃; constructing time-series input features with each hour of the next 24 hours as a time step; the features of each time step include two parts, the first part being the meteorological forecast data corresponding to that time step (i.e., the hourly outdoor temperature, wind speed of the corresponding building area, and solar radiation collected in Step 501). The first part is the solar radiation intensity), and the second part is the target water supply temperature of each building corresponding to this time step. The features of the 24 time steps are arranged in chronological order to form the time series input features. The feature dimension of each time step is 5 (1 for outdoor temperature, 1 for wind speed, 1 for solar radiation intensity, 1 for target water temperature of buildings facing the street, and 1 for target water temperature of buildings in the middle). For example, the features of the time step at 0:00 in the future are: outdoor temperature -10℃, wind speed in the area facing the street 2.8m / s, wind speed in the middle area 2.5m / s, solar radiation intensity 0W / ㎡, target water temperature of buildings facing the street 78℃, and target water temperature of buildings in the middle 73℃. This ensures that the features of each time step can be associated with meteorological conditions and building operation strategies, and that the temporal logic is coherent.

[0080] Step 503: Input the time series input features into a Long Short-Term Memory (LSTM) network. Process the data through the network's temporal memory units to predict the future trend of heat load demand. Specifically, this includes: first, constructing and training a pre-trained LSM network model. The model construction phase uses hourly operating data from the past two heating seasons of an old residential community in northern China as training samples; collecting hourly meteorological data during this period, including outdoor temperature, outdoor wind speed (distinguishing between street-facing and intermediate building areas), and solar radiation intensity; collecting the actual hourly heat load values ​​for each building, and analyzing the hourly gas consumption of the 2.8MW gas-fired hot water boiler, the temperature difference between the supply and return water in the pipeline network, and... The flow data is calculated, for example, based on the gas consumption (m³) and the gas calorific value (approximately 36 MJ / m³), the total heat is calculated, and then the heat load is divided according to the proportion of the street-facing buildings (Building 1 to Building 2, Building 7 to Building 8) and the intermediate buildings (Building 3 to Building 6) to obtain the actual hourly heat load values ​​for each type of building; the collected sample data is preprocessed to remove abnormal heat load values ​​caused by boiler failure or data collection interruption, such as data exceeding twice the normal range, the meteorological data and heat load data are aligned by timestamp to ensure that the hourly samples contain complete meteorological parameters and corresponding building heat load values, and then divided into training set and validation set in a 7:3 ratio. 3 / h 3

[0081] The Long Short-Term Memory (LSTM) network model was designed with 5 neurons in the input layer to match the dimensions of each time step of the subsequent time series input features (outdoor temperature, wind speed in the street-facing area, wind speed in the central area, solar radiation intensity, and water temperature correlation features of the building). Two LSTM unit layers were set in the hidden layer, each with 32 neurons, to capture long-term dependencies in the data over time. The output layer had 2 neurons, corresponding to the hourly heat load demand values ​​of the street-facing and central buildings, respectively, ensuring the output could distinguish the load differences between the two types of buildings. A Dropout mechanism (with a ratio of 0.2) was used between each LSTM unit layer to prevent overfitting, and a linear activation function was used in the output layer to match the range of heat load values.

[0082] The model training process is initiated, with mean squared error chosen as the loss function to measure the difference between the model's predicted and actual heat load values. The Adam optimizer is used, with an initial learning rate of 0.001, which is adjusted by decreasing by 10% every 50 training iterations. Training set samples are input into the model in batches of 32. The predicted heat load values ​​are calculated via forward propagation, and the model's weights and bias parameters are updated via backpropagation. After each training iteration (traversing all training set samples), the validation loss of the model is calculated using the validation set. If the validation loss does not decrease for five consecutive iterations (fluctuation less than 0.001), training is stopped to avoid overfitting. After training, the optimal parameters of the model are saved, forming a pre-trained long short-term memory network model that can be directly accessed.

[0083] Retrieve the pre-trained Long Short-Term Memory (LSTM) network model that has been trained as described above, and input the 24-hour time series input features constructed in step 502 (each time step includes outdoor temperature, wind speed in the street area, wind speed in the central area, solar radiation intensity, and building target water temperature correlation features) into the model as a whole, and start the model's temporal memory unit processing; the temporal memory unit retains key information of historical time steps (such as the continuous impact of the previous hour's low outdoor temperature on the current heat load, and the lagging adjustment effect of the previous 3 hours' solar radiation intensity on the heat load) through a gating mechanism of input gate, forget gate, and output gate, while updating the feature information of the current time step, according to time. The input data of 24 time steps are processed sequentially and the temporal correlation features are extracted step by step. After processing, the model outputs the trend of heat load demand changes every hour for the next 24 hours through the output layer. For example, at 6 o'clock in the future (outdoor temperature -15℃, wind speed in the street area 3.0m / s), the heat load demand of the street-facing buildings is 130kW and that of the middle buildings is 105kW. At 14 o'clock in the future (outdoor temperature -8℃, solar radiation intensity 280W / ㎡), the heat load demand of the street-facing buildings is 110kW and that of the middle buildings is 90kW. This shows the fluctuation pattern of heat load with meteorological conditions and accurately distinguishes the heat load difference between the street-facing buildings and the middle buildings.

[0084] Step 504: Based on the aforementioned trend of heat load demand changes, a multi-objective optimization function is adopted, using the goal of maintaining the user's room temperature to meet the standard and minimizing total energy consumption. The final operating parameter combination is solved using a particle swarm optimization algorithm. Specifically, this includes: setting constraints for the multi-objective optimization function, where the room temperature compliance constraint is that the indoor temperature of all users in all buildings is maintained between 18℃ and 22℃, and the minimum total energy consumption constraint is that the hourly gas consumption of the 2.8MW gas-fired hot water boiler is minimized; constructing a multi-objective optimization function based on these constraints, which integrates the two optimization objectives of room temperature compliance and minimum total energy consumption into a quantifiable comprehensive index. This avoids excessive energy consumption due to solely pursuing room temperature compliance or room temperature failure due to solely controlling energy consumption, while eliminating the conflict between the two objectives and making the optimization direction more aligned with actual heating demand; the value generated by this function is a comprehensive score, which directly reflects the quality of a set of operating parameter combinations. Specifically, the higher the room temperature compliance rate and the lower the gas consumption, the higher the comprehensive score. This value is used to subsequently determine whether each parameter combination meets the optimization objectives.

[0085] Based on the changing trend of heat load demand, the range of operating parameters to be optimized is determined. The setpoint for water supply temperature is 65℃ to 85℃ (not exceeding the boiler's rated outlet water temperature of 85℃), and the setpoint for circulating pump frequency is 30Hz to 50Hz (ensuring normal operation of the circulating pump and not exceeding the rated frequency). A particle swarm optimization algorithm is initiated, with an initial particle swarm size of 50 particles. Each particle represents a set of parameter combinations for water supply temperature and circulating pump frequency. During algorithm iteration, each particle is ranked according to its own historical best position (i.e., the position corresponding to the parameter combination with the highest comprehensive score calculated by that particle in the past) and the group's best position (i.e., the parameter combination with the highest comprehensive score calculated by all particles in the past). The flight direction and speed are adjusted according to the corresponding position, and the objective function value corresponding to each particle is calculated, which is the comprehensive score mentioned above. After 100 iterations, if the change in the optimal position of the group is less than 0.1℃ (water supply temperature) and 0.5Hz (circulation pump frequency) in 10 consecutive iterations, the iteration is stopped. Finally, the parameter combination with the highest comprehensive score and the satisfaction of 100% room temperature compliance rate and the lowest gas consumption is selected. For example, when the heat load demand is 130kW, the water supply temperature is 78℃ and the circulation pump frequency is 45Hz, and this combination has the highest comprehensive score. When the heat load demand is 90kW, the water supply temperature is 72℃ and the circulation pump frequency is 38Hz, and this combination also meets the requirement of the best comprehensive score.

[0086] Step 505: Convert the final operating parameter combination into the target output parameter setpoints for the boiler, including the water supply temperature setpoint and the circulating pump frequency setpoint. Specifically, this includes: establishing a correspondence between the operating parameter combination and the target output parameters based on the rated operating parameters of the 2.8MW gas-fired hot water boiler (rated outlet water temperature 85℃, rated circulating water volume 50m³) and the rated performance parameters of the circulating pump (rated frequency 50Hz, rated flow rate 50m³); converting the final operating parameter combination obtained in step 504, wherein the water supply temperature setpoint directly adopts the optimized value without additional conversion, for example, the optimized 78℃ is directly used as the boiler water supply temperature setpoint; the circulating pump frequency setpoint is based on the optimized... The frequency value is directly determined; for example, the optimized 45Hz is directly used as the circulating pump frequency setting value. For the heat load demand at different times in the next 24 hours, the target output parameter setting values ​​for the corresponding time periods are converted to form a 24-hour parameter setting sequence. For example, at 6:00 AM, the corresponding water supply temperature is 78℃ and the circulating pump frequency is 45Hz; at 2:00 PM, the corresponding water supply temperature is 72℃ and the circulating pump frequency is 38Hz; and at 10:00 PM, the corresponding water supply temperature is 75℃ and the circulating pump frequency is 42Hz. The converted parameter setting values ​​are organized in chronological order, and the corresponding water supply temperature and circulating pump frequency for each time point are labeled to ensure that the parameter values ​​are within the rated operating range of the boiler and circulating pump and match the trend of heat load demand changes. 3 / h 3 / h

[0087] Step 506 involves sending the target output parameter setpoints to the boiler control system for real-time control. This includes: establishing a communication connection with the boiler control system, which includes a water temperature control unit and a circulating pump frequency converter control unit, ensuring the real-time performance and stability of data transmission; sequentially sending the target output parameter setpoints (water supply temperature, circulating pump frequency) for each time point according to a 24-hour time sequence to the boiler control system; employing a timed transmission mechanism, sending the parameter setpoints for that hour once per hour, for example, sending the water supply temperature of 76℃ and the circulating pump frequency of 43Hz for the period from 0:00 to 1:00 at 0:00 in the future, and sending the setpoints at 1:00... The system sends parameters from 1 to 2 hours in advance. After receiving the parameter settings, the boiler control system automatically adjusts the gas supply to maintain the set water temperature, and the circulating pump frequency converter automatically adjusts the motor frequency to match the set frequency. At the same time, the system feeds back the actual operating parameters (actual water temperature and actual circulating pump frequency) to the data processing module every 10 minutes to ensure the execution effect of the settings. If the actual parameters deviate from the settings by more than ±1℃ (water temperature) or ±1Hz (frequency), the mechanism of resending the settings is triggered to realize real-time control closed loop, avoid passive response problems, and ensure that the heating system operates stably according to the optimized parameters.

[0088] In this embodiment of the invention, multi-dimensional meteorological forecast data of outdoor temperature, wind speed, and solar radiation intensity for a preset future time period are acquired, covering key environmental factors affecting heat load demand. Time-series input features are constructed by combining differentiated operation strategies (including differences in building demand) with meteorological forecast data (including future environmental changes), integrating static strategy data and dynamic time-series data to ensure temporal correlation of the features. The resulting time-series features are adapted to the processing requirements of Long Short-Term Memory (LSTM) networks, laying the foundation for the network to capture data temporal correlations and improving the adaptability of input data and the model. The temporal memory units of the LTM network are used to process the time-series input features, effectively capturing the temporal dimension dependencies of the data (such as the lagging effect of meteorology on heat load and the correlation of heat load between time periods). The future heat load demand change trend is output, ensuring that the trend prediction reflects the long- and short-term correlations in the time series and aligns with actual heat demand. The system analyzes load variation patterns and uses room temperature compliance and minimum total energy consumption as multi-objective optimization functions to determine dual core optimization objectives. It employs a particle swarm optimization algorithm to solve for the combination of operating parameters, efficiently searching the optimal parameter space under multiple constraints to quickly find parameters that balance heat demand and energy consumption control, thus improving the overall rationality of parameters. The abstract operating parameter combinations obtained from particle swarm optimization are converted into specific setpoints for boiler-executable water supply temperature and circulating pump frequency. The optimization results are transformed from theoretical parameters into equipment operating values, eliminating the disconnect between parameters and equipment operation and improving the feasibility of optimization results. The target output parameter setpoints are directly sent to the boiler control system, achieving seamless integration of data processing results and real-time equipment control. This avoids delays in data processing and control execution, ensuring that optimized setpoints are applied to boiler operation promptly, responding quickly to changes in heat load, and improving the timeliness and effectiveness of regulation.

[0089] In a preferred embodiment of the present invention, step 503 above, which involves inputting the time-series input features into a long short-term memory network and processing them through the network's time-series memory units to predict the trend of heat load demand changes in the future time period, includes:

[0090] Step 5031 involves inputting the time series input features into the Long Short-Term Memory (LSTM) network. Specifically, this includes: first determining the time series input features to be input, which are structured data for each hour of the next 24 hours in an old residential community in northern China. Each time step contains five dimensions of information: the current hourly outdoor temperature (e.g., -10°C at 0:00, -15°C at 6:00, -8°C at 14:00), the wind speed in the area where the buildings along the street (Building 1 to Building 2, Building 7 to Building 8) are located (e.g., 3.0 m / s at 6:00, 2.0 m / s at 14:00), the wind speed in the area where the intermediate buildings (Building 3 to Building 6) are located (e.g., 2.7 m / s at 6:00, 1.7 m / s at 14:00), and the current hourly wind speed. The solar radiation intensity of the previous hour (e.g., 0 W / m² at 6:00 AM and 280 W / m² at 2:00 PM) and the corresponding target water supply temperature values ​​for buildings during the same time period (76-78℃ for buildings facing the street and 72-73℃ for buildings in the middle) are used to input the time series features into a feature sequence of 24 consecutive time steps. This sequence is then input into a pre-trained Long Short-Term Memory (LSTM) network. The network is trained based on hourly data from the past two heating seasons in the community. The samples include meteorological data from the same period and the actual heat load values ​​of each building (calculated from the gas consumption of the 2.8MW gas-fired hot water boiler and pipeline parameters). The parameters have been optimized and the network is suitable for the time series data processing needs of the community's heating system.

[0091] Step 5032: Based on the time-series input features, calculate the retention weight of the current input information through the network's input gate structure, and calculate the decay weight of the historical memory state through the forget gate structure, and obtain the input gate output results and forget gate output results respectively. Specifically, this includes: based on the time-series input features, initiating the network's gate structure calculation for the data at each time step; analyzing the key information of the current time step through the input gate structure, for example, in the time step of 6 hours in the future (outdoor temperature -15℃, wind speed in the street area 3.0m / s), the input gate identifies low temperature and high wind speed as the core factors affecting the heat load, and calculates the indoor temperature during this period. The retention weights for external temperature (0.8), wind speed (0.7), and solar radiation intensity (0 W / m²) (0.1) are used to obtain the input gate output. The useful information in the historical memory state is analyzed using a forget gate structure. For example, when processing data for the next 6 hours, the forget gate retrieves heat load-related memories from similar cold wave periods in history (such as periods with temperatures below -14°C in past heating seasons). The attenuation weight for this type of effective historical memory is calculated to be 0.2 (i.e., retaining 80% of the effective information). Simultaneously, the attenuation weight for historical memories from non-cold wave periods (such as periods with temperatures below 10°C in the past) is calculated to be 0.8 (i.e., retaining only 20% of the redundant information), resulting in the forget gate output.

[0092] Step 5033: Based on the input gate output and forget gate output, the memory cell states of the Long Short-Term Memory (LSTM) network are weighted, fused, and updated to obtain the updated cell states. Specifically, this includes: obtaining the input gate output and forget gate output to update the memory cell states of the LTM network; multiplying the forget gate output with the memory cell states of the previous time step to retain valid historical information, such as retaining the associated memory of high load corresponding to low temperatures during past cold waves; multiplying the input gate output with the information of the current time step features after activation function processing to filter out the key feature information of the current time step, such as the heat load impact features corresponding to -15℃ + high wind speed in the next 6 hours; adding the above two results to complete the weighted fusion and update of the memory cell states; the updated cell states contain both the heat load patterns of historical cold wave periods and integrate the meteorological features of the current time step, for example, the updated cell states in the next 6 hours can reflect the specific patterns of high heat loss and high load requirements of buildings along the street in an environment of -15℃.

[0093] Step 5034: Based on the updated cell state, calculate the hidden state at the current moment through the output gate, and add this hidden state to the hidden state sequence to obtain the updated continuous hidden state sequence. Specifically, this includes: starting the output gate structure calculation based on the updated cell state; the output gate processes the updated cell state through an activation function to filter out the key features that need to be output at the current moment, such as the correlation between low temperature and high wind speed and high load on the street in the cell state at the next 6 hours, and the correlation between mild temperature + solar radiation and the intermediate building in the cell state at the next 14 hours (outdoor temperature -8℃, solar radiation 280W / ㎡). The correlation characteristics of low load; calculate the hidden state at the current moment based on the screening results, with each time step corresponding to a hidden state, and the hidden state contains the core temporal characteristics related to the heat load at that moment; add the hidden states calculated at each time step to the hidden state sequence in chronological order, and finally form a continuous hidden state sequence covering the next 24 hours. For example, the 6th element in the sequence (corresponding to the next 6 hours) is low temperature, high wind speed and high load characteristics, and the 14th element (corresponding to the next 14 hours) is mild temperature + solar radiation and low load characteristics, to ensure that the sequence can coherently reflect the changes in the influencing factors of heat load at different times.

[0094] Step 5035: Based on the continuous hidden state sequence, regression calculation is performed through a fully connected layer to predict the trend of heat load demand changes in the future time period. Specifically, this includes: inputting the continuous hidden state sequence into the fully connected layer of a Long Short-Term Memory network; the fully connected layer performs linear transformation and integration on each element in the hidden state sequence through a weight matrix, transforming the abstract temporal features in the hidden state into specific heat load values; during the transformation process, regression calculations are performed separately for street-facing buildings and intermediate buildings, taking into account the differences between buildings in the community. For example, for the element with low temperature and high wind speed in the hidden state sequence, regression is performed to obtain the street-facing... The heat load demand values ​​for buildings (e.g., 130kW at 6:00 AM) and intermediate buildings (e.g., 105kW at 6:00 AM) are calculated. For the elements of mild temperature and solar radiation, regression is performed to obtain the heat load demand values ​​for street-facing buildings (e.g., 110kW at 2:00 PM) and intermediate buildings (e.g., 90kW at 2:00 PM). The regression results for all time steps are organized in chronological order to form the hourly heat load demand trend for the next 24 hours. This trend reflects both the load fluctuations over time, such as high loads during low-temperature periods and low loads during high-temperature periods, and also distinguishes the load differences between street-facing and intermediate buildings.

[0095] In a preferred embodiment of the present invention, step 504 above, based on the trend of heat load demand changes, uses maintaining the user's room temperature to meet the standard and minimizing total energy consumption as a multi-objective optimization function, and employs a particle swarm optimization algorithm to solve for the final combination of operating parameters, including:

[0096] Step 5041: Based on the aforementioned trend of heat load demand changes, construct a multi-objective optimization function with the goal of minimizing the deviation of user room temperature from the target and minimizing total energy consumption. Specifically, this includes: first, determining the trend of heat load demand changes, which covers the hourly heat load of an old residential community in northern China over the next 24 hours. Specifically, the heat load demand of the street-facing buildings (Buildings 1-2 and 7-8) reaches 130kW during cold wave periods (e.g., when the outdoor temperature is -15℃ at 6 AM), while the heat load demand of the intermediate buildings (Buildings 3-6) is 105kW during the same period. The heat load gradually decreases as the outdoor temperature rises at other times. Based on this trend, construct a multi-objective optimization function. The function contains two core parts: the first part is minimizing the deviation of user room temperature from the target, with the goal of minimizing the deviation of user room temperature from the target. Based on room temperature data uploaded by 10% of residents in the district, the acceptable room temperature range was set at 18℃ to 22℃. The absolute deviation of the actual room temperature from this range was calculated. The deviation was calculated separately for buildings facing the street and buildings in the middle, and the average value was taken to ensure that the room temperature of all buildings was covered. The second part was to minimize total energy consumption, using the hourly gas consumption of a 2.8MW gas-fired hot water boiler as the metric. The consumption data was collected by the boiler gas flow sensor and converted into standard energy consumption units (m). To balance the two objectives, the weight of the room temperature compliance deviation was set at 0.6, and the weight of total energy consumption was set at 0.4. The two parts were weighted and summed to obtain the comprehensive output value of the optimization function. The smaller this value, the better the parameter combination, thus achieving joint optimization of room temperature compliance and energy consumption control. 3 / h

[0097] Step 5042: Based on the multi-objective optimization function, initialize the particle swarm algorithm parameters, set the particle swarm size, and randomly generate an initial position vector and an initial velocity vector for each particle, containing the boiler water supply temperature setpoint and the circulating pump frequency setpoint, to obtain the initialized particle swarm. Specifically, this includes: initializing the core parameters of the particle swarm algorithm based on the multi-objective optimization function; setting the particle swarm size to 50 particles, which can cover a sufficient parameter search space, avoiding incomplete searching due to an excessively small size, and avoiding an excessively large size that increases the computational load (adapting to the real-time control requirements of the community heating system); determining that the position vector of each particle contains two dimensions, namely the boiler water supply temperature setpoint and the circulating pump frequency setpoint, wherein the water supply temperature setpoint ranges from 65℃ to 85℃, and the circulating pump frequency setpoint ranges from 30Hz to 50Hz (specifically...). To ensure the normal operation of the circulating pump and adapt to water circulation needs under different heat loads through frequency adjustment, an initial position vector is randomly generated for each particle. During generation, the parameter differences between buildings facing the street and those in the middle must be considered. For example, for particles in buildings facing the street during high load periods, the initial water supply temperature can be randomly selected between 75℃ and 82℃, while for particles in buildings in the middle during low load periods, the initial water supply temperature can be randomly selected between 68℃ and 75℃. Simultaneously, an initial velocity vector is generated for each particle. The two dimensions of the velocity vector correspond to the adjustment step size of the water supply temperature and the circulating pump frequency, respectively. The water supply temperature velocity range is 0.5℃ to 2℃ / iteration, and the circulating pump frequency velocity range is 0.3Hz to 1Hz / iteration. This avoids the initial velocity being too large, causing the parameters to jump out of the feasible region, or too small, leading to low search efficiency. After generating the initial vectors for all particles, an initialized particle swarm is obtained.

[0098] Step 5043: Based on the initialized particle swarm, in each iteration, calculate the fitness value of each particle using a multi-objective optimization function based on its current position vector; based on the fitness value, update the individual historical final position of each particle and the global final position of the particle swarm; adjust the velocity vector and position vector of each particle based on the updated individual historical final position and global final position, specifically including: based on the initialized particle swarm, setting the total number of iterations to 100, and starting the iterative calculation; in each iteration, first obtain the velocity vector and position vector of each particle... The current position vector (i.e., a set of water supply temperature and circulating pump frequency parameters) is substituted into the multi-objective optimization function in step 5041 to calculate the fitness value corresponding to the particle. For example, if a particle's position vector is a water supply temperature of 78℃ and a circulating pump frequency of 45Hz, after substituting it into the function, the average deviation of room temperature from the target is 0.4℃ and the energy consumption is 8.2m. The comprehensive fitness value is calculated by weighting: 0.4×0.6+8.2×0.4. The smaller the fitness value, the better the parameter combination. Based on the calculated fitness value, the individual historical best position of each particle is updated, i.e. The fitness value of the current particle is compared with its best fitness value in all previous iterations. If the current value is better, the current position vector is updated to the individual's historical best position. Simultaneously, the global best position of the particle swarm is updated by comparing the fitness values ​​corresponding to the individual historical best positions of all particles and selecting the position vector with the smallest fitness value as the global best position. Based on the updated individual historical best position and the global best position, the velocity vector of each particle is adjusted, taking into account the directions of the individual and global best. For example, if a particle's current position is far from the global best position, the velocity component towards the global best direction is increased; if it is close to the global best position, the velocity component is decreased to avoid excessive particle oscillation. The position vector of the particle is then updated based on the adjusted velocity vector, ensuring that the new position vector remains within the feasible region of a water supply temperature of 65℃ to 85℃ and a circulation pump frequency of 30Hz to 50Hz. If it exceeds this range, the position vector is corrected to the boundary value of the feasible region. This calculation, update, and adjustment process is repeated until the current iteration is completed, ensuring that the overall fitness value of the particle swarm gradually decreases after each iteration, moving closer to the optimal parameter region. 3 / h

[0099] Step 5044: Based on the global final position, when the preset convergence condition is met, the parameter combination corresponding to the global final position is taken as the final operating parameter combination. Specifically, this includes: after each iteration, monitoring the change of the global optimal position of the particle swarm, and setting the preset convergence condition as follows: in 10 consecutive iterations, the change in water supply temperature corresponding to the global optimal position is less than 0.1℃ and the change in circulation pump frequency is less than 0.5Hz. This convergence accuracy can meet the requirements of room temperature control deviation less than 1℃ and energy consumption control accuracy. If the convergence condition is not met, the next iteration continues until the total number of iterations is reached or the convergence condition is met. When the convergence condition is met, the iteration is stopped, and the parameter combination corresponding to the global optimal position at this time is determined as the final operating parameter combination. For example, in the case of the next 6 hours (outdoor...) In the iteration with a temperature of -15℃ and a heat load of 130kW for buildings facing the street, the converged global optimal position is a water supply temperature of 78℃ and a circulation pump frequency of 45Hz. After substituting this parameter combination into the optimization function, the average deviation of room temperature compliance is 0.3℃ (meeting the range of 18℃ to 22℃ with a small deviation), and the energy consumption is 8.0m. In the iteration corresponding to the next 14 hours (outdoor temperature of -8℃ and heat load of 90kW for intermediate buildings), the converged global optimal position is a water supply temperature of 72℃ and a circulation pump frequency of 38Hz. This parameter combination can make the room temperature compliance deviation of intermediate buildings 0.5℃, with an energy consumption of 6.5m. The final operating parameter combinations corresponding to different heat load periods are compiled and summarized to form a parameter set covering the next 24 hours, ensuring that the parameter combination for each period can meet the goal of achieving room temperature compliance and minimizing total energy consumption. 3 / h 3 / h

[0100] like Figure 2 As shown, embodiments of the present invention also provide an operational data analysis system for a heating system, comprising:

[0101] The data acquisition module is used to collect equipment operation data, pipeline operation data, and user-side data in real time; it performs time-series alignment of the data based on a unified time benchmark and performs feature dimensionality reduction to construct a multi-source fusion dataset.

[0102] The calibration module is used to calculate the key performance indicators of each building's thermal inlet in real time based on a multi-source fusion dataset; based on the key performance indicators, it dynamically determines multiple distributed reference nodes through the pipeline topology and real-time hydraulic conditions; it constructs a hydraulic balance reference model based on the data of the reference nodes and extracts their state characteristics, and generates dynamic calibration coefficients based on these state characteristics; it uses the dynamic calibration coefficients to calibrate the key performance indicators and obtain the calibrated key performance indicators.

[0103] The identification module is used to compare the real-time values ​​of the calibrated key performance indicators with the preset dynamic threshold range to obtain the indicator deviation state vector; based on the indicator deviation state vector, abnormal data clusters are identified to obtain diagnostic results including the abnormality type and location.

[0104] The strategy generation module is used to generate differentiated operation strategies based on the building's historical room temperature, building characteristics, and real-time meteorological parameters in the multi-source fusion dataset if the diagnosis result is no abnormality.

[0105] The optimization control module is used to acquire meteorological forecast data for a preset time period; based on the differentiated operation strategy and meteorological forecast data, it predicts changes in heat load demand through a long short-term memory network; based on changes in heat load demand, with the goal of achieving room temperature standard and minimizing energy consumption, it calculates the target output parameter setpoint and sends the setpoint to the boiler control system for execution.

[0106] It should be noted that this system is a system corresponding to the above method. All implementation methods in the above method embodiments are applicable to this embodiment and can achieve the same technical effect.

[0107] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A method for analyzing operational data of a heating system, characterized in that, The method includes: Step 100: Collect equipment operation data, pipeline operation data, and user-side data in real time; perform time-series alignment of the data based on a unified time benchmark and perform feature dimensionality reduction to construct a multi-source fusion dataset; Step 200: Based on the multi-source fusion dataset, calculate the key performance indicators of the thermal inlets of each building in real time; based on the key performance indicators, dynamically determine multiple distributed reference nodes through the pipe network topology and real-time hydraulic conditions; construct a hydraulic balance reference model based on the data of the reference nodes, extract its state characteristics, and generate dynamic calibration coefficients based on these state characteristics; calibrate the key performance indicators using the dynamic calibration coefficients to obtain the calibrated key performance indicators, including: Based on the multi-source fusion dataset, the instantaneous supply and return water pressure difference, temperature difference, and flow deviation rate of each building's thermal inlet are calculated in real time. Based on the key performance indicators, combined with the pipeline network topology and real-time hydraulic condition data, characteristic parameters of pipeline network nodes are extracted; the key performance indicators and node characteristic parameters are standardized to generate a standardized feature matrix. Using the standardized feature matrix as input, the weighted Euclidean distance between each data sample is calculated through a clustering analysis algorithm, and the K-means clustering method is used to divide the pipeline nodes into multiple clusters to obtain the cluster division results. Based on the cluster division results and combined with the connectivity analysis of the pipeline topology, the node that is closest to the centroid and is representative in each cluster is selected as the distributed reference node to complete the dynamic identification and determination of the reference node. Collect real-time operating data from the distributed reference nodes; use the real-time operating data to establish a hydraulic balance reference model reflecting the overall hydraulic state through a multiple linear regression method; The pressure distribution characteristics and flow balance characteristics in the hydraulic balance reference model are extracted as state characteristics. The state characteristics are correlated with the standard reference values ​​to obtain the correlation analysis results; based on the correlation analysis results, the mathematical relationship between the state characteristics and the reference values ​​is fitted using the least squares method to obtain the fitting results; Based on the fitting results, the weight coefficients of each state feature are obtained; and dynamic calibration coefficients are generated based on the weight coefficients. The dynamic calibration coefficients are applied to key performance indicators and dynamically calibrated and corrected to obtain calibrated key performance indicators. Step 300: Compare the real-time values ​​of the calibrated key performance indicators with the preset dynamic threshold range to obtain the indicator deviation state vector; Based on the indicator deviation state vector, identify abnormal data clusters and obtain diagnostic results including abnormality type and location. Step 400: If the diagnosis result is no abnormality, then generate a differentiated operation strategy based on the building's historical room temperature, building characteristics and real-time meteorological parameters in the multi-source fusion dataset. Step 500: Obtain meteorological forecast data for a preset time period; based on the differentiated operation strategy and meteorological forecast data, predict changes in heat load demand through a long short-term memory network; based on changes in heat load demand, calculate the target output parameter setpoint with the goal of achieving room temperature standard and minimizing energy consumption, and send the setpoint to the boiler control system for execution.

2. The method for analyzing the operating data of a heating system according to claim 1, characterized in that, Step 100 includes: Data on equipment operation, pipeline operation, and user side are collected in parallel from three independent data sources: boiler control system, property monitoring platform, and user service system. Based on a unified time base, the equipment operation data, pipeline operation data, and user-side data are timestamped to obtain a timestamped data sequence. The timestamped data sequence is then normalized using a dynamic time warping algorithm to obtain an intermediate time series dataset with consistent time resolution. Based on the intermediate time series dataset, the eigenvectors and eigenvalues ​​are calculated using principal component analysis to determine the main feature components and their contributions; the main feature components are then filtered according to a preset variance contribution rate threshold to obtain the main feature subspace. The feature data in the main feature subspace are geometrically projected and mapped to a low-dimensional feature space to obtain the dimensionality-reduced feature projection data. Based on the dimensionality-reduced feature projection data, the feature information of multi-source data streams is integrated through the feature weighted fusion method to construct a multi-source fusion dataset with temporal consistency, feature orthogonality, and elimination of multicollinearity.

3. The method for analyzing the operating data of a heating system according to claim 2, characterized in that, Step 300 includes: Based on the calibrated key performance indicators, their real-time values ​​are compared with the preset dynamic threshold range item by item, the relative deviation value of each indicator is calculated, and the indicator deviation state vector is obtained. The deviation vector of the index is used as input data; a density clustering algorithm is performed on the input data to calculate the reachability distance between each data sample point; high-density areas and outliers are identified based on the distribution characteristics of the reachability distance to obtain the data point classification results; and preliminary abnormal data clusters are formed based on the data point classification results. Based on the spatial distribution characteristics of the preliminary abnormal data clusters and combined with the pipeline network topology, the abnormal data clusters are merged and optimized to determine the final abnormal data cluster classification. Based on the final abnormal data cluster classification, the statistical characteristic parameters of each abnormal data cluster are extracted, including the average deviation of the index within the cluster, spatial distribution density, and duration characteristics. Based on the statistical feature parameters, the system uses a rule-based reasoning method to match preset anomaly type discrimination rules, determines the anomaly type and its location, and obtains a diagnostic result that includes the anomaly type and location.

4. The method for analyzing the operating data of a heating system according to claim 3, characterized in that, Step 400 includes: If the diagnostic result is no abnormality, then the historical room temperature, building characteristics and real-time meteorological parameters of each building are extracted from the multi-source fusion dataset; The historical room temperature, building characteristics and real-time meteorological parameters of the buildings are combined to construct a feature vector for each building. The feature vectors are input into a pre-trained neural network, and the initial heating demand prediction values ​​for each building are obtained through forward propagation calculation of the neural network. Based on the pipeline network topology, a hydraulic coupling analysis is performed on the initial heating demand forecast of each building to calculate the heat loss correction during pipeline transportation. The initial heating demand forecast is combined with the heat loss correction to calculate the differentiated heating demand target value for each building. Based on the differentiated heating demand target values ​​of each building, the corresponding target water supply temperature setpoint is calculated to form a differentiated operation strategy.

5. The method for analyzing the operating data of a heating system according to claim 4, characterized in that, Step 500 includes: Obtain meteorological forecast data for a future preset time period, including outdoor temperature, wind speed, and solar radiation intensity parameters; Based on the differentiated operation strategy and the meteorological forecast data, time series input features are constructed; The time series input features are input into a long short-term memory network, and processed by the network's time-series memory units to predict the trend of heat load demand changes in future time periods. Based on the aforementioned trend of heat load demand changes, with the goal of maintaining the user's room temperature to the standard and minimizing total energy consumption as the multi-objective optimization function, the particle swarm optimization algorithm is used to solve for the final combination of operating parameters. The final combination of operating parameters is converted into target output parameter setpoints for the boiler, including water supply temperature setpoints and circulating pump frequency setpoints. The target output parameter setpoint is sent to the boiler control system for real-time control.

6. The method for analyzing the operating data of a heating system according to claim 5, characterized in that, The time-series input features are fed into a Long Short-Term Memory (LSTM) network, and processed by the network's temporal memory units to predict future trends in heat load demand, including: The time series input features are then input into a long short-term memory network. Based on the time series input features, the retention weight of the current input information is calculated through the input gate structure of the network, and the decay weight of the historical memory state is calculated through the forget gate structure, and the input gate output results and forget gate output results are obtained respectively. Based on the output results of the input gate and the output results of the forget gate, the memory cell states of the Long Short-Term Memory Network are weighted, fused, and updated to obtain the updated cell states. Based on the updated cell state, the hidden state at the current time step is calculated through the output gate, and the hidden state is added to the hidden state sequence to obtain the updated continuous hidden state sequence. Based on the continuous hidden state sequence, regression calculations are performed through a fully connected layer to predict the trend of heat load demand changes in the future time period.

7. The method for analyzing the operating data of a heating system according to claim 6, characterized in that, Based on the aforementioned trend in heat load demand, and taking maintaining the user's room temperature while minimizing total energy consumption as the multi-objective optimization function, a particle swarm optimization algorithm is used to solve for the final combination of operating parameters, including: Based on the aforementioned trend of heat load demand changes, a multi-objective optimization function is constructed with the goal of minimizing the deviation of user room temperature compliance and minimizing total energy consumption. Based on the multi-objective optimization function, the particle swarm algorithm parameters are initialized, the particle swarm size is set, and an initial position vector and an initial velocity vector containing the boiler water supply temperature setpoint and the circulating pump frequency setpoint are randomly generated for each particle to obtain the initialized particle swarm. Based on the initialized particle swarm, in each iteration, the fitness value of each particle is calculated using a multi-objective optimization function according to its current position vector; based on the fitness value, the individual historical final position of each particle and the global final position of the particle swarm are updated; based on the updated individual historical final position and global final position, the velocity vector and position vector of each particle are adjusted. Based on the global final position, when the preset convergence condition is met, the parameter combination corresponding to the global final position is used as the final running parameter combination.

8. A system for analyzing operational data of a heating system, wherein the system implements the method as described in any one of claims 1 to 7, characterized in that, include: The data acquisition module is used to collect equipment operation data, pipeline operation data, and user-side data in real time. The data is time-series aligned based on a unified time benchmark, and feature dimensionality reduction is performed through principal component analysis to construct a multi-source fusion dataset. The calibration module is used to calculate the key performance indicators of each building's thermal inlet in real time based on a multi-source fusion dataset; based on the key performance indicators, it dynamically determines multiple distributed reference nodes through the pipeline topology and real-time hydraulic conditions; it constructs a hydraulic balance reference model based on the data of the reference nodes and extracts their state characteristics, and generates dynamic calibration coefficients based on these state characteristics; it uses the dynamic calibration coefficients to calibrate the key performance indicators and obtain the calibrated key performance indicators. The identification module is used to compare the real-time values ​​of the calibrated key performance indicators with the preset dynamic threshold range to obtain the indicator deviation state vector; based on the indicator deviation state vector, the clustering analysis method is used to identify abnormal data clusters and obtain diagnostic results containing the abnormality type and location. The strategy generation module is used to dynamically generate differentiated operation strategies based on the building's historical room temperature, building characteristics, and real-time meteorological parameters in the multi-source fusion dataset, using a pre-trained neural network, if the diagnostic result is no abnormality. The optimization control module is used to acquire meteorological forecast data for a preset time period; predict changes in heat load demand using a long short-term memory network based on the differentiated operation strategies and meteorological forecast data; calculate the target output parameter setpoint based on the changes in heat load demand, with the optimization objective of achieving the room temperature standard and minimizing energy consumption, and send the setpoint to the boiler control system for execution.

Images