A method and system for monitoring and analyzing data of a heating network based on big data

Abstract

The application discloses a kind of based on big data's heating network management data monitoring analysis method and system, comprising the following steps: step one: constructing pipe network topology diagram;Step two: constructing space-time dynamic baseline model, generating should be running track;Step three: obtaining net residual sequence;Step four: should be running track and net residual sequence are input into the improved USAD model with baseline conditionization countermeasure reconstruction mechanism, reconstruct error is obtained by double self-encoder structure, and abnormal score is generated based on the reconstruct error;Step five: according to the total abnormal score of monitoring node in current operation cycle, alarm level is judged;Step six: when triggering alarm, locate abnormal source node and influence range, and generate abnormal evidence chain;Step seven: according to alarm level and abnormal evidence chain, generate control instruction set.The present application fuses space-time baseline modeling and improved USAD detection, realizes the accurate monitoring and efficient control of heating network operation state.

Description

Technical Field

[0001] This invention relates to the field of heating system operation monitoring technology, and in particular to a method and system for monitoring and analyzing heating network management data based on big data. Background Technology

[0002] With the rapid development of smart heating and the urban energy internet, technologies for monitoring and intelligently analyzing multi-source operational data of heating networks are receiving increasing attention. Existing centralized heating network operation monitoring systems largely rely on single sensor data or simple threshold judgments for anomaly detection and control command generation, but in actual operation, they generally suffer from the following problems:

[0003] The monitoring data exhibits significant periodic and seasonal fluctuations. During the heating season and under different meteorological conditions, traditional methods struggle to effectively remove the periodic background from normal operating modes, leading to the masking of abnormal signals or confusion with load changes. Significant hydraulic and thermal coupling exists between monitoring nodes in the pipeline network, and anomalies may spread along the network topology. Single-node analysis neglects spatial correlation, easily missing source nodes or misjudging the scope of influence. Actual operating data exhibits significant noise, and the anomaly patterns are nonlinear and non-stationary. Traditional time-series analysis methods and single autoencoder reconstruction models have low sensitivity when dealing with slow drift, small fluctuations, or multi-node linkage anomalies. Existing methods lack a hierarchical control strategy based on the evidence chain of anomaly characteristics, making it impossible to achieve targeted adjustments to valve opening, pump frequency, and heat source temperature.

[0004] Therefore, how to provide a data monitoring and analysis method and system for heating network management based on big data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a data monitoring and analysis method and system for heating network pipelines based on big data. This invention integrates spatiotemporal baseline modeling and improved USAD detection technology. By constructing a comprehensive feature set of multi-source operational data and pipeline topology information, a spatiotemporal dynamic baseline model of the monitoring nodes is established. Combined with a baseline conditional adversarial reconstruction mechanism, it achieves accurate monitoring and anomaly identification of the operating status of the heating network. It can effectively locate the anomaly source nodes and the scope of influence under complex operating conditions, and generate an executable set of control instructions, thereby significantly improving the operational safety, anomaly response speed, and control accuracy of the heating network pipeline. It has the advantages of strong real-time performance, high adaptability, and excellent reliability.

[0006] A data monitoring and analysis method for heating network management based on big data according to an embodiment of the present invention includes the following steps:

[0007] Step 1: Collect heating network operation data, meteorological data, and network topology information to construct a network topology map;

[0008] Step 2: Construct a spatiotemporal dynamic baseline model for each monitoring node in the pipeline network topology diagram, and generate the expected operating trajectory of the monitoring node in the current and prediction periods;

[0009] Step 3: Calculate the difference between the measured value of the monitoring node and the expected operating trajectory, and perform periodic and seasonal decomposition to obtain the net residual sequence;

[0010] Step 4: Input the expected running trajectory and the net residual sequence into the improved USAD model with a baseline conditional adversarial reconstruction mechanism, obtain the reconstruction error through a dual autoencoder structure, and generate anomaly scores based on the reconstruction error;

[0011] Step 5: Set multi-level alarm thresholds based on the distribution of abnormal scores in the same historical time period, and determine the alarm level according to the total abnormal score of the monitoring node in the current operating cycle;

[0012] Step 6: When an alarm is triggered, analyze the spatiotemporal distribution of the anomaly score based on the pipeline topology diagram, locate the anomaly source node and its impact range, and generate an anomaly evidence chain;

[0013] Step 7: Generate a control instruction set based on the alarm level and the abnormal evidence chain. The control instruction set includes valve opening adjustment, pump frequency adjustment and heat source outlet water temperature adjustment.

[0014] Optionally, the heating network operation data includes temperature data, pressure data, flow data, valve opening data, pump frequency data, and heat source outlet water temperature of each monitoring node in the network;

[0015] Based on the geographical information of the pipeline network, pipeline connection relationships, and equipment layout information, the connection relationships between monitoring nodes are determined. Monitoring nodes that are directly connected to each other are defined as hydraulically adjacent nodes, and the fluid transmission path direction, pipeline length, and pipe diameter information between adjacent monitoring nodes are recorded in the connection edges of the topology map. By analyzing the correlation of temperature, pressure, and flow rate changes of adjacent monitoring nodes in historical operating data, coupling degree information reflecting the degree of correlation between the operating states of monitoring nodes is obtained, and the coupling degree information is associated and labeled with the connection edges to form a pipeline network topology map.

[0016] Optionally, step two specifically includes:

[0017] Determine the topological relationships between the target monitoring node and its upstream and downstream adjacent monitoring nodes in the pipeline network topology diagram;

[0018] The historical heating network operation data of the target monitoring node and adjacent monitoring nodes are obtained, and time alignment processing is performed to remove missing data and fill in missing values.

[0019] Time series features are extracted from the processed historical heating network operation data, and the connection relationships between nodes, pipe length, pipe diameter and flow direction information are extracted from the topology as spatial structure features.

[0020] The time series features and the spatial structure features are fused together. The fusion process includes encoding the time series features in the time dimension, encoding the spatial structure features in the topological relationship, and concatenating the encoded features in a unified feature space to generate a comprehensive feature set.

[0021] The comprehensive feature set is divided into a training dataset and a validation dataset according to the time sequence and the distribution of monitoring nodes. A fitting algorithm based on iterative optimization is used to establish the mapping relationship between the operating status of monitoring nodes and the comprehensive feature set, so as to obtain a spatiotemporal dynamic baseline model.

[0022] Input the current heating network operation data and meteorological data into the spatiotemporal dynamic baseline model, and output the expected operating trajectory within the current period and the preset prediction period.

[0023] Optionally, step three specifically includes:

[0024] The difference between the measured value of the monitoring node in the current period and the corresponding expected operating trajectory is calculated to obtain the deviation sequence;

[0025] A time series dataset is constructed from the deviation sequence in chronological order, and the average value and fluctuation range of each time period are calculated based on a sliding time window;

[0026] A periodic fitting sequence is generated based on the recurring variation pattern of the average value within adjacent time windows; a seasonal fitting sequence is generated based on the variation pattern of the average value showing a continuous upward or downward trend under different seasons and climatic conditions, and the fluctuation range remaining within a preset fluctuation threshold.

[0027] The periodic fitting sequence and the seasonal fitting sequence are respectively mapped to the deviation sequence according to the time index. The net residual sequence is obtained by removing the periodic fitting sequence and the seasonal fitting sequence from the deviation sequence.

[0028] Optionally, step four specifically includes:

[0029] The expected operating trajectory and the net residual sequence are aligned according to the monitoring node number and time order. The expected operating trajectory is processed by feature encoding to obtain a baseline condition vector. The net residual sequence is processed by feature encoding to obtain a residual signal vector. The baseline condition vector and the residual signal vector are concatenated in the feature dimension to form a conditional input feature set.

[0030] The conditional input feature set is input into an improved USAD model with a baseline conditional adversarial reconstruction mechanism, which includes an encoding stage and a decoding stage.

[0031] The encoding stage simultaneously receives the baseline condition vector and the residual signal vector in a shared weight manner, so that the encoding result retains the alignment relationship between the normal operation mode and the deviation mode in the latent space.

[0032] The decoding stage introduces the baseline condition vector into the decoding layer and fuses it with the latent space features to guide the reconstruction result to fit the expected running trajectory;

[0033] The improved USAD model includes a first autoencoder and a second autoencoder. The first autoencoder performs self-reconstruction based on the conditional input feature set, and the second autoencoder performs cross-reconstruction based on the output of the first autoencoder. The residual adjustment branch of the baseline conditional vector is retained in the cross-reconstruction path to enhance the anomalous amplification effect.

[0034] The reconstruction errors of the self-reconstruction path and the intersection reconstruction path are calculated separately. The reconstruction error of the intersection reconstruction path is amplified according to a preset weight. The amplified reconstruction error of the intersection reconstruction path is then weighted and fused with the reconstruction error of the self-reconstruction path to obtain the anomaly score of the monitoring node.

[0035] Optionally, step five specifically includes:

[0036] Obtain abnormal score records of the target monitoring node in the same historical time period, wherein the same time period includes operating cycles with the same season, consistent meteorological conditions, and consistent heat load levels.

[0037] The abnormal score records are organized according to time sequence and monitoring node number to construct a historical abnormal score distribution dataset;

[0038] Statistical analysis was performed on the historical abnormal score distribution dataset to calculate the abnormal score values ​​corresponding to multiple quantiles;

[0039] A multi-level alarm threshold is set based on the abnormal scores of the multiple quantiles. The multi-level alarm threshold includes a first alarm threshold and a second alarm threshold. When the total abnormal score of the monitoring node in the current operating cycle is greater than or equal to the first alarm threshold, the alarm level is the third level; when the total abnormal score of the monitoring node in the current operating cycle is less than the first alarm threshold but greater than the second alarm threshold, the alarm level is the second level; when the total abnormal score of the monitoring node in the current operating cycle is less than or equal to the second alarm threshold, the alarm level is the first level.

[0040] Optionally, step six specifically includes:

[0041] When an alarm is triggered, obtain the abnormal scores of all monitoring nodes in the current period, and map the abnormal scores to the corresponding monitoring node positions in the pipeline topology diagram;

[0042] Based on the spatial connection relationship and anomaly score of the monitoring nodes in the pipeline topology diagram, the monitoring nodes are traversed in descending order of anomaly score, and the first monitoring node whose anomaly score exceeds the first anomaly threshold within a preset number of consecutive sampling periods is taken as the anomaly source node.

[0043] Starting from the abnormal source node, adjacent monitoring nodes are sequentially searched along the upstream and downstream directions of the pipeline topology. Monitoring nodes with abnormal scores greater than the second abnormal threshold are added to the abnormal influence range set until the abnormal scores of adjacent nodes are lower than the second abnormal threshold or there are no further adjacent nodes. The first abnormal threshold is greater than the second abnormal threshold.

[0044] An abnormal evidence chain is generated based on each monitoring node in the set of abnormal source nodes and abnormal influence ranges, combined with the pipeline network topology connection relationship.

[0045] Optionally, the specific rules for generating the control instruction set are as follows:

[0046] The maximum temperature deviation value of the monitoring nodes in the abnormal evidence chain is defined as follows: The maximum pressure deviation is The maximum flow deviation value is And set the temperature regulation threshold. Pressure regulation threshold and flow regulation threshold ;

[0047] When the alarm level is Level 1, and the abnormal evidence chain satisfies:

[0048] and and ;

[0049] Furthermore, there is no abnormal impact range distributed across different branch numbers, and the output maintains the standard operating control commands with the original operating parameters;

[0050] When the alarm level is level two, and the abnormal evidence chain satisfies:

[0051] ;

[0052] If any of the conditions in the above conditions are met, and there is no abnormal influence range distributed across different branch numbers, output a single-parameter regulating control command, where: Adjust the valve opening as needed. Adjust the pump set frequency as needed. Adjust the outlet water temperature of the heat source;

[0053] When the alarm level is level 3, and the abnormal evidence chain satisfies:

[0054] ;

[0055] If any two or more of the conditions are met, and the abnormal influence range is distributed across different branch numbers, output a multi-parameter joint regulation control command, simultaneously adjusting the valve opening, pump frequency, and heat source outlet water temperature.

[0056] A heating network data monitoring and analysis system based on big data according to an embodiment of the present invention includes the following modules:

[0057] The data acquisition module is used to collect heating network operation data, meteorological data, and network topology information, and generate a network topology map.

[0058] The spatiotemporal baseline modeling module is used to construct a spatiotemporal dynamic baseline model and output the expected trajectory.

[0059] The deviation decomposition module is used to calculate the difference between the measured value of the monitoring node and the expected operating trajectory and to perform periodic and seasonal decomposition to obtain the net residual sequence.

[0060] An improved USAD detection module is used to input the expected running trajectory and the net residual sequence into an improved USAD model with a baseline-conditional adversarial reconstruction mechanism to obtain anomaly scores;

[0061] The threshold alarm module is used to set multi-level alarm thresholds and trigger alarms based on historical anomaly score distribution;

[0062] The anomaly localization module is used to locate the source node and the scope of influence of anomalies based on the pipeline network topology map and anomaly score, and generate anomaly evidence chain;

[0063] The control instruction generation module is used to generate control instruction sets based on alarm levels and abnormal evidence chains.

[0064] The beneficial effects of this invention are:

[0065] This invention constructs a spatiotemporal dynamic baseline model that integrates time-series features and spatial structural features. Addressing the periodic fluctuations, seasonal trends, and spatial correlation coupling issues in heating network operation data, it employs an iterative optimization-based fitting algorithm to establish a mapping relationship between the operating status of monitoring nodes and the comprehensive feature set. This generates a dynamically updated expected operating trajectory that adapts to weather conditions and load changes, achieving high-precision modeling of normal operating modes. In the anomaly detection stage, an improved USAD model with a baseline conditionalized adversarial reconstruction mechanism is proposed. The expected operating trajectory and net residual sequence are encoded as baseline conditional vectors and residual signal vectors, respectively, and then concatenated along the feature dimensions before being input into a dual autoencoder structure. Through self-reconstruction, the model learns the low-error features of the normal mode and performs cross-validation. By reconstructing and combining residual adjustment branch amplification with anomalous signals showing significant baseline differences, and utilizing an adversarial discriminator to enhance the model's sensitivity to nonlinear and non-stationary small-amplitude fluctuations, periodic and seasonal background interference is effectively suppressed. In the anomaly location and handling stage, the spatiotemporal distribution of anomaly scores is analyzed in conjunction with the pipeline topology map. Based on the continuous over-threshold rule, the anomaly source and its impact range are accurately identified, generating anomaly evidence chains. Multi-level alarm thresholds are set according to the historical anomaly score distribution, and differentiated control instruction sets for valve opening adjustment, pump frequency adjustment, and heat source outlet water temperature adjustment are automatically generated according to the alarm level. This achieves closed-loop optimization of the entire chain from anomaly detection and source tracing to graded handling, significantly improving the accuracy of anomaly identification, location precision, and the level of fine-grained operation control. Attached Figure Description

[0066] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0067] Figure 1 This is an overall flowchart of a big data-based heating network data monitoring and analysis method proposed in this invention.

[0068] Figure 2 This is a schematic diagram of the structure of a heating network data monitoring and analysis system based on big data proposed in this invention. Detailed Implementation

[0069] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0070] refer to Figure 1 A data monitoring and analysis method for heating network management based on big data includes the following steps:

[0071] Step 1: Collect heating network operation data, meteorological data, and network topology information to construct a network topology map;

[0072] Step 2: Construct a spatiotemporal dynamic baseline model for each monitoring node in the pipeline network topology diagram, and generate the expected operating trajectory of the monitoring node in the current and prediction periods, wherein the prediction period is set by the user;

[0073] Step 3: Calculate the difference between the measured value of the monitoring node and the expected operating trajectory, and perform periodic and seasonal decomposition to obtain the net residual sequence;

[0074] Step 4: Input the expected running trajectory and the net residual sequence into the improved USAD model with a baseline conditional adversarial reconstruction mechanism, obtain the reconstruction error through a dual autoencoder structure, and generate anomaly scores based on the reconstruction error;

[0075] Step 5: Set multi-level alarm thresholds based on the distribution of abnormal scores in the same historical time period, and determine the alarm level according to the total abnormal score of the monitoring node in the current operating cycle;

[0076] Step 6: When an alarm is triggered, analyze the spatiotemporal distribution of the anomaly score based on the pipeline topology diagram, locate the anomaly source node and its impact range, and generate an anomaly evidence chain;

[0077] Step 7: Generate a control instruction set based on the alarm level and the abnormal evidence chain. The control instruction set includes valve opening adjustment, pump frequency adjustment and heat source outlet water temperature adjustment.

[0078] In this embodiment, the heating network operation data includes temperature data, pressure data, flow data, valve opening data, pump frequency data, and heat source outlet water temperature of each monitoring node in the network.

[0079] Based on the geographical information of the pipeline network, pipeline connection relationships, and equipment layout information, the connection relationships between monitoring nodes are determined. Monitoring nodes that are directly connected to each other are defined as hydraulically adjacent nodes, and the fluid transmission path direction, pipeline length, and pipe diameter information between adjacent monitoring nodes are recorded in the connection edges of the topology map. By analyzing the correlation of temperature, pressure, and flow rate changes of adjacent monitoring nodes in historical operating data, coupling degree information reflecting the degree of correlation between the operating states of monitoring nodes is obtained, and the coupling degree information is associated and labeled with the connection edges to form a pipeline network topology map.

[0080] In this embodiment, step two specifically includes:

[0081] Determine the topological relationships between the target monitoring node and its upstream and downstream adjacent monitoring nodes in the pipeline network topology diagram;

[0082] The historical heating network operation data of the target monitoring node and adjacent monitoring nodes are obtained, and time alignment processing is performed to remove missing data and fill in missing values.

[0083] Time series features are extracted from the processed historical heating network operation data, and the connection relationships between nodes, pipe length, pipe diameter and flow direction information are extracted from the topology as spatial structure features.

[0084] The time series features and the spatial structure features are fused together. The fusion process includes encoding the time series features in the time dimension, encoding the spatial structure features in the topological relationship, and concatenating the encoded features in a unified feature space to generate a comprehensive feature set.

[0085] The comprehensive feature set is divided into training and validation datasets according to time sequence and monitoring node distribution. An iterative optimization-based fitting algorithm is used to establish a mapping relationship between the operating status of monitoring nodes and the comprehensive feature set. Specifically, the comprehensive feature set is divided into training and validation datasets according to time sequence and monitoring node distribution. The training dataset is input into a prediction model containing a fusion layer of time series features and spatial structure features. This prediction model is a spatiotemporal prediction model combining a temporal convolutional network and a graph convolutional network. The temporal convolutional network is used to extract multi-scale time-dependent features from historical operating data, and the graph convolutional network is used to extract spatial correlation features between monitoring nodes based on the pipeline topology. The outputs of the two are concatenated and fused in the feature fusion layer and then input into the fully connected prediction layer. A nonlinear mapping relationship between the comprehensive feature set and the operating status of monitoring nodes is established through end-to-end training.

[0086] A loss function with parameter regularization and spatial smoothing constraints is constructed using the difference between the predicted output and the actual running state. The model parameters are iteratively updated using gradient descent-like optimization methods. After each iteration, the validation error is calculated on the validation dataset, and early stopping is determined based on the changes in the validation error. Training is terminated when the validation error does not decrease or decreases by less than a preset tolerance threshold in multiple consecutive iterations. The parameters corresponding to the minimum validation set error are taken as the final model parameters. This results in a spatiotemporal dynamic baseline model that can output the expected running trajectory of the current cycle and the predicted cycle based on the comprehensive feature set.

[0087] Input the current heating network operation data and meteorological data into the spatiotemporal dynamic baseline model, and output the expected operating trajectory within the current period and the preset prediction period.

[0088] In this embodiment, step three specifically includes:

[0089] The difference between the measured value of the monitoring node in the current period and the corresponding expected operating trajectory is calculated to obtain the deviation sequence;

[0090] A time series dataset is constructed from the deviation sequence in chronological order, and the average value and fluctuation range of each time period are calculated based on a sliding time window;

[0091] A periodic fitting sequence is generated based on the recurring variation pattern of the average value within adjacent time windows; a seasonal fitting sequence is generated based on the variation pattern of the average value showing a continuous upward or downward trend under different seasons and climatic conditions, and the fluctuation range remaining within a preset fluctuation threshold.

[0092] The periodic fitting sequence and the seasonal fitting sequence are respectively mapped to the deviation sequence according to the time index. The net residual sequence is obtained by removing the periodic fitting sequence and the seasonal fitting sequence from the deviation sequence.

[0093] In this embodiment, step four specifically includes:

[0094] The expected operating trajectory and the net residual sequence are aligned according to the monitoring node number and time order. The expected operating trajectory is processed by feature encoding to obtain a baseline condition vector. The net residual sequence is processed by feature encoding to obtain a residual signal vector. The baseline condition vector and the residual signal vector are concatenated in the feature dimension to form a conditional input feature set.

[0095] The conditional input feature set is input into an improved USAD model with a baseline conditional adversarial reconstruction mechanism, which includes an encoding stage and a decoding stage.

[0096] The encoding stage simultaneously receives the baseline condition vector and the residual signal vector in a shared weight manner, so that the encoding result retains the alignment relationship between the normal operation mode and the deviation mode in the latent space.

[0097] In the baseline conditional adversarial reconstruction mechanism of the present invention, the encoding stage simultaneously inputs the expected running trajectory and the net residual sequence, so that the latent space representation simultaneously contains the feature distribution of the normal operation mode and the deviation mode. Thus, in the decoding stage, the difference of the deviation mode can be highlighted while maintaining the alignment of the normal mode, thereby achieving effective separation and amplification of abnormal signals.

[0098] The decoding stage introduces the baseline conditional vector into the decoding layer and fuses it with the latent space features to guide the reconstruction result to fit the expected running trajectory; during the training process, an adversarial discriminator is introduced to distinguish the decoded output from the real signal, thereby improving the model's sensitivity to bias patterns.

[0099] The improved USAD model includes a first autoencoder and a second autoencoder. The first autoencoder performs self-reconstruction based on the conditional input feature set, and the second autoencoder performs cross-reconstruction based on the output of the first autoencoder. The residual adjustment branch of the baseline conditional vector is retained in the cross-reconstruction path to enhance the anomalous amplification effect.

[0100] In the improved USAD model, self-reconstruction refers to inputting the conditional input feature set into the first autoencoder, mapping the input features to the latent space representation through the encoding layer, and then restoring the latent space representation through the decoding layer to obtain a reconstructed output that is aligned with the original input feature set in both the time and spatial dimensions. The purpose of self-reconstruction is to enable the first autoencoder to learn the correspondence between the normal operating mode and the baseline conditions, thereby generating a lower reconstruction error when facing data that conforms to the baseline mode and a higher reconstruction error when facing abnormally deviating inputs.

[0101] Inter-reconstruction refers to using the reconstruction output of the first autoencoder as the input of the second autoencoder. During the encoding stage, features are re-extracted and the baseline conditional vector is fused again in the latent space. During the decoding stage, an inter-reconstruction output corresponding to the original input feature set is generated. The design intent of inter-reconstruction is to amplify anomalous signals present in the input feature set by introducing information loss and nonlinear transformation through the two-stage encoding and decoding process. To further enhance the amplification effect on anomalies, a residual adjustment branch of the baseline conditional vector is retained in the inter-reconstruction path. When generating the inter-reconstruction output, the residual adjustment branch is element-wise weighted and superimposed with the decoding result, so that features that differ greatly from the baseline conditions occupy a higher weight in the final output, thereby significantly improving the sensitivity of the inter-reconstruction error to anomalies.

[0102] The reconstruction errors of the self-reconstruction path and the intersection reconstruction path are calculated separately. The reconstruction error of the intersection reconstruction path is amplified according to a preset weight. The amplified reconstruction error of the intersection reconstruction path is then weighted and fused with the reconstruction error of the self-reconstruction path to obtain the anomaly score of the monitoring node.

[0103] In this embodiment, step five specifically includes:

[0104] Obtain abnormal score records of the target monitoring node in the same historical time period, wherein the same time period includes operating cycles with the same season, consistent meteorological conditions, and consistent heat load levels.

[0105] The abnormal score records are organized according to time sequence and monitoring node number to construct a historical abnormal score distribution dataset;

[0106] Statistical analysis was performed on the historical abnormal score distribution dataset to calculate the abnormal score values ​​corresponding to multiple quantiles;

[0107] A multi-level alarm threshold is set based on the abnormal scores of the multiple quantiles. The multi-level alarm threshold includes a first alarm threshold and a second alarm threshold. When the total abnormal score of the monitoring node in the current operating cycle is greater than or equal to the first alarm threshold, the alarm level is the third level; when the total abnormal score of the monitoring node in the current operating cycle is less than the first alarm threshold but greater than the second alarm threshold, the alarm level is the second level; when the total abnormal score of the monitoring node in the current operating cycle is less than or equal to the second alarm threshold, the alarm level is the first level.

[0108] In this embodiment, step six specifically includes:

[0109] When an alarm is triggered, obtain the abnormal scores of all monitoring nodes in the current period, and map the abnormal scores to the corresponding monitoring node positions in the pipeline topology diagram;

[0110] Based on the spatial connection relationship and anomaly score of the monitoring nodes in the pipeline topology diagram, the monitoring nodes are traversed in descending order of anomaly score, and the first monitoring node whose anomaly score exceeds the first anomaly threshold within a preset number of consecutive sampling periods is taken as the anomaly source node.

[0111] Starting from the abnormal source node, adjacent monitoring nodes are sequentially searched along the upstream and downstream directions of the pipeline topology. Monitoring nodes with abnormal scores greater than the second abnormal threshold are added to the abnormal influence range set until the abnormal scores of adjacent nodes are lower than the second abnormal threshold or there are no further adjacent nodes. The first abnormal threshold is greater than the second abnormal threshold.

[0112] Based on each monitoring node in the set of abnormal source nodes and abnormal influence ranges, and combined with the pipeline network topology connection relationship, an abnormal evidence chain is generated.

[0113] Although the present invention can identify the monitoring node that triggers the alarm in step five, this monitoring node is not necessarily the source node of the anomaly. In order to accurately determine the root cause of the anomaly and its scope of influence, it is necessary to perform a global analysis of the anomaly scores of all monitoring nodes in the current period after the alarm is triggered, and combine the upstream and downstream relationships of the pipeline topology to locate the starting node as the source node of the anomaly, thereby ensuring the accuracy of anomaly location and subsequent control.

[0114] In this embodiment, the specific rules for generating the control instruction set are as follows:

[0115] The maximum temperature deviation value of the monitoring nodes in the abnormal evidence chain is defined as follows: The maximum pressure deviation is The maximum flow deviation value is And set the temperature regulation threshold. Pressure regulation threshold and flow regulation threshold ;

[0116] When the alarm level is Level 1, and the abnormal evidence chain satisfies:

[0117] and and ;

[0118] Furthermore, there is no abnormal impact range distributed across different branch numbers, and the output maintains the standard operating control commands with the original operating parameters;

[0119] When the alarm level is level two, and the abnormal evidence chain satisfies:

[0120] ;

[0121] If any of the conditions in the above conditions are met, and there is no abnormal influence range distributed across different branch numbers, output a single-parameter regulating control command, where: Adjust the valve opening as needed. Adjust the pump set frequency as needed. Adjust the outlet water temperature of the heat source;

[0122] When the alarm level is level 3, and the abnormal evidence chain satisfies:

[0123] ;

[0124] If any two or more of the conditions are met, and the abnormal influence range is distributed across different branch numbers, output a multi-parameter joint regulation control command, simultaneously adjusting the valve opening, pump frequency, and heat source outlet water temperature.

[0125] refer to Figure 2 A big data-based heating network management data monitoring and analysis system includes the following modules:

[0126] The data acquisition module is used to collect heating network operation data, meteorological data, and network topology information, and generate a network topology map.

[0127] The spatiotemporal baseline modeling module is used to construct a spatiotemporal dynamic baseline model and output the expected trajectory.

[0128] The deviation decomposition module is used to calculate the difference between the measured value of the monitoring node and the expected operating trajectory and to perform periodic and seasonal decomposition to obtain the net residual sequence.

[0129] An improved USAD detection module is used to input the expected running trajectory and the net residual sequence into an improved USAD model with a baseline-conditional adversarial reconstruction mechanism to obtain anomaly scores;

[0130] The threshold alarm module is used to set multi-level alarm thresholds and trigger alarms based on historical anomaly score distribution;

[0131] The anomaly localization module is used to locate the source node and the scope of influence of anomalies based on the pipeline network topology map and anomaly score, and generate anomaly evidence chain;

[0132] The control instruction generation module is used to generate control instruction sets based on alarm levels and abnormal evidence chains.

[0133] Example 1:

[0134] To verify the feasibility of this invention in practice, it was applied to the urban heating network system managed by a central heating company in a northern city. This city has a long winter heating season, covering an area of ​​approximately 2.8 million square meters, with a total pipeline length of about 130 kilometers. It has 82 key monitoring nodes, encompassing various operating parameters such as temperature, pressure, flow rate, valve opening, pump frequency, and heat source outlet water temperature. For a long time, this heating system has been prone to problems such as localized insufficient heating, hydraulic imbalance in the pipeline network, and overloaded operation of the heat source under extreme weather conditions. Traditional monitoring relies on manual inspections and static threshold alarms, which suffer from drawbacks such as delayed response, high false alarm rates, and difficulty in locating the source of the fault. This invention addresses these problems by being deployed and implemented throughout a complete heating season. It utilizes two years of historical operating data as training data for baseline modeling, combined with real-time collected meteorological information and operating data for dynamic monitoring and control.

[0135] In this embodiment, the system first connects to the SCADA system and meteorological data interface via a data acquisition module to obtain real-time network node operation data every 5 minutes, while simultaneously importing pipeline topology information stored in the GIS system. The system automatically constructs a topology map including pipe lengths, diameters, flow directions, and hydraulic adjacency relationships between nodes, and calculates the coupling degree between nodes based on historical data, providing spatial structural features for subsequent modeling. Before a strong cold wave arrives, the system generates the expected operating trajectories of all network nodes based on historical operational data under similar meteorological conditions, and calculates the deviation between measured values ​​and trajectories in real time, obtaining the net residual sequence through periodic and seasonal decomposition.

[0136] The improved USAD detection module inputs the expected operating trajectory and net residual sequence into a dual autoencoder structure with a baseline-conditional adversarial reconstruction mechanism. In the early morning before an extreme low-temperature event, the system detected a group of nodes in section C of the heating branch in Xicheng District with significantly increased abnormal scores. Among them, the abnormal scores of nodes C17 and C18 exceeded the first abnormal threshold in multiple consecutive sampling periods, while the abnormal scores of downstream nodes C19 and C20 were also higher than the second abnormal threshold. The system automatically identified C17 as the abnormal source node and generated an abnormal evidence chain including a sudden drop in node temperature of 2.8℃, a pressure drop of 0.06MPa, and a decrease in downstream flow of 3.5m³ / h.

[0137] The threshold alarm module determined the event to be a Level 3 alarm. The control command generation module immediately issued a multi-parameter joint adjustment control command to the dispatch center, including increasing the opening of the C15 branch valve by 6%, increasing the frequency of the D pump group by 0.7Hz, and increasing the heat source outlet water temperature by 2.5℃. After the control was executed, the abnormal score dropped to the normal range within half an hour, and the heating temperature recovered to an average of 52.4℃, significantly improving the heating experience for downstream residents.

[0138] To compare the effects, a similar cold wave event from the same period in a previous year was selected as a reference. At that time, the method of this invention was not used; only manual inspection and static alarms were relied upon. Table 1 below shows the differences in abnormal response indicators before and after the application of this invention.

[0139] Table 1 Comparison of Abnormal Response Data Before and After Implementation

[0140] index traditional methods Embodiments of the present invention Anomaly detection delay (min) 75 5 Fault location accuracy 67% 96% Recovery time (min) 195 35 The average room temperature decreased by (°C) during this period. 3.1 0.6

[0141] As can be seen from the comparison results in Table 1 above, the embodiments of the present invention have significant advantages over traditional methods in terms of abnormal monitoring and control of heating network pipes. The abnormal detection delay is reduced from 75 minutes to 5 minutes, achieving near real-time fault detection; the fault location accuracy is improved from 67% to 96%, significantly reducing the risk of false alarms and missed alarms; the fault recovery time is reduced from 195 minutes to 35 minutes, demonstrating the high efficiency of abnormal location and control command generation; in terms of operational impact, the average room temperature drop during the period was only 0.6℃, significantly better than the 3.1℃ drop of the traditional method, effectively ensuring user comfort, and further proving the comprehensive improvement effect of the present invention in terms of the operational safety, intelligence level, and stability of heating network pipes.

[0142] This embodiment significantly improves the real-time monitoring and accurate diagnosis capabilities of the heating network operation status by introducing a spatiotemporal dynamic baseline model and a baseline conditional adversarial reconstruction mechanism. It can quickly detect and accurately locate the source of anomalies and their impact range. Combined with the control instruction set generated by multi-level alarm thresholds and anomaly evidence chains, it realizes intelligent and automated adjustment of operating parameters, effectively shortens the fault handling cycle, reduces operational fluctuations, and ensures the stability of the heating system and the comfort of users' heating.

[0143] This embodiment deeply integrates heating network operation data, meteorological data, and network topology information to construct a spatiotemporal dynamic baseline model. Combined with an improved USAD model's baseline conditional adversarial reconstruction mechanism, it achieves global, real-time, and precise monitoring of the operating status. The system can automatically identify the characteristic differences between normal operation and deviation modes, triggering alarms at the initial stage of abnormal signals and quickly locating the anomaly source and its impact range using network topology relationships. Simultaneously, it generates targeted control commands based on anomaly evidence chain analysis, enabling intelligent adjustments to valve opening, pump frequency, and heat source outlet water temperature, thereby reducing the need for manual intervention while ensuring heating stability. This invention not only automates and automates the monitoring process but also significantly improves anomaly response speed and control accuracy, effectively coping with complex and changing operating environments. It is of great significance for improving the safety, reliability, and energy efficiency of urban heating systems.

[0144] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A big data-based heating network management data monitoring and analysis method, characterized in that, The method comprises the following steps: Step one: collecting heating network pipe operation data, meteorological data and pipe network topology information, and constructing a pipe network topology graph; Step two: constructing a space-time dynamic baseline model for each monitoring node in the pipe network topology graph, and generating a natural running track of the monitoring node in the current and predicted period; Step three: calculating the difference between the measured value of the monitoring node and the natural running track, and performing periodic and seasonal decomposition to obtain a net residual sequence; Step four: inputting the natural running track and the net residual sequence into an improved USAD model with baseline conditioning adversarial reconstruction mechanism, obtaining reconstruction error through a double auto-encoder structure, and generating an anomaly score based on the reconstruction error; Step five: setting multi-level alarm thresholds based on the anomaly score distribution of the same period in history, and determining the alarm level according to the total anomaly score of the monitoring node in the current operation period; Step six: when the alarm is triggered, analyzing the space-time distribution of the anomaly score based on the pipe network topology graph, locating the abnormal source node and the influence range, and generating an abnormal evidence chain; Step seven: generating a control instruction set according to the alarm level and the abnormal evidence chain, the control instruction set including valve opening adjustment, pump group frequency adjustment and heat source outlet water temperature adjustment; The step two is specifically: Determine the topological relationship of the target monitoring node and the upstream and downstream adjacent monitoring nodes in the pipe network topology graph; Obtain the historical heating network pipe operation data of the target monitoring node and the adjacent monitoring nodes, and perform time alignment processing, eliminate missing data and fill in missing values; Extract time series features from the processed historical heating network pipe operation data, and extract connection relationship, pipe length, pipe diameter and flow direction information as spatial structure features from the topological relationship; Fuse the time series features and the spatial structure features, which includes time dimension coding of the time series features, topological relationship coding of the spatial structure features, and splicing of the coded features in a unified feature space to generate a comprehensive feature set; Divide the comprehensive feature set into a training data set and a validation data set according to time sequence and monitoring node distribution, establish a mapping relationship between the monitoring node running state and the comprehensive feature set by using an iterative optimization based fitting algorithm, and obtain a space-time dynamic baseline model; Input the current heating network pipe operation data and meteorological data into the space-time dynamic baseline model, and output the natural running track in the current period and the preset prediction period; The step four is specifically: Align the natural running track and the net residual sequence according to the monitoring node number and time sequence, obtain a baseline condition vector by feature coding processing of the natural running track, obtain a residual signal vector by feature coding processing of the net residual sequence, and splice the baseline condition vector and the residual signal vector in the feature dimension to form a conditioned input feature set; Input the conditioned input feature set into the improved USAD model with baseline conditioning adversarial reconstruction mechanism, the baseline conditioning adversarial reconstruction mechanism including an encoding stage and a decoding stage; The encoding stage simultaneously receives the baseline condition vector and the residual signal vector in a shared weight manner, so that the alignment relationship between the normal operation mode and the deviation mode is retained in the latent space of the encoding result; The decoding stage introduces the baseline condition vector into the decoding layer and fuses it with the latent space feature, so that the reconstruction result conforms to the normal operation trajectory; The improved USAD model includes a first autoencoder and a second autoencoder, the first autoencoder performs self-reconstruction based on the conditioned input feature set, the second autoencoder performs cross-reconstruction based on the output of the first autoencoder, and the residual adjustment branch of the baseline condition vector in the cross-reconstruction path enhances the abnormal amplification effect; The reconstruction errors of the self-reconstruction path and the cross-reconstruction path are calculated respectively, the reconstruction error of the cross-reconstruction path is amplified according to the preset weight, and the amplified reconstruction error of the cross-reconstruction path and the reconstruction error of the self-reconstruction path are fused by weighting to obtain the abnormal score of the monitoring node.

2. The method according to claim 1, wherein, The heat supply network pipe operation data includes temperature data, pressure data, flow data, valve opening data, pump group frequency data and heat source outlet water temperature of each monitoring node in the pipe network; According to the geographical information, pipe connection relationship and equipment arrangement information of the pipe network, the connection relationship between the monitoring nodes is determined, the monitoring nodes directly connected with each other are defined as hydraulic adjacent nodes, and the fluid transmission path direction, pipe length and pipe diameter information between the adjacent monitoring nodes are recorded in the connection edge of the topological graph; By analyzing the correlation of temperature, pressure and flow changes of adjacent monitoring nodes in historical operation data, the coupling degree information reflecting the correlation degree of the operation states of the monitoring nodes is obtained, and the coupling degree information is associated and labeled with the connection edge to form a pipe network topological graph.

3. The method according to claim 1, wherein, The step three is specifically: Calculate the difference between the measured value of the monitoring node in the current period and the corresponding normal operation trajectory to obtain a deviation sequence; A time series data set is constructed according to the time sequence of the deviation sequence, and the average value and fluctuation range of each time period are calculated based on a sliding time window; According to the average value, a periodic fitting sequence is generated according to the repeated change rule within adjacent time windows; according to the average value, a seasonal fitting sequence is generated according to the change rule that the average value presents a continuous rising or falling trend under different seasons and climate conditions, and the fluctuation range is kept within a preset fluctuation threshold; The periodic fitting sequence and the seasonal fitting sequence are respectively operated according to the time index of the deviation sequence, and the periodic fitting sequence and the seasonal fitting sequence are removed from the deviation sequence to obtain a net residual sequence.

4. The method according to claim 1, wherein, The step five is specifically: Obtain the abnormal score record of the target monitoring node in the same type period, the same type period includes the same season, the same meteorological condition value range and the same heat load level range operation period; The abnormal score record is sorted according to time sequence and monitoring node number to construct a historical abnormal score distribution data set; Statistical analysis is performed on the historical abnormal score distribution data set to calculate the abnormal score values corresponding to multiple quantile points; The multi-level alarm threshold is set according to the abnormal score values of the plurality of sub-points, and the multi-level alarm threshold comprises a first alarm threshold and a second alarm threshold; when the total abnormal score of the monitoring node in the current operation period is greater than or equal to the first alarm threshold, the alarm level is the third level; when the total abnormal score of the monitoring node in the current operation period is less than the first alarm threshold and greater than the second alarm threshold, the alarm level is the second level; and when the total abnormal score of the monitoring node in the current operation period is less than or equal to the second alarm threshold, the alarm level is the first level.

5. The method according to claim 1, wherein, The step six is specifically: When the alarm is triggered, the abnormal score of all monitoring nodes in the current period is obtained, and the abnormal score is mapped to the corresponding monitoring node position in the pipe network topology graph; In the pipe network topology graph, based on the spatial connection relationship and the abnormal score of the monitoring node, the monitoring nodes are traversed in the order of the abnormal score from high to low, the first monitoring node whose abnormal score exceeds the first abnormal threshold in a preset number of continuous sampling periods is taken as an abnormal source node; Starting from the abnormal source node, adjacent monitoring nodes are searched in the upstream and downstream directions of the pipe network topology graph, and the monitoring nodes whose abnormal scores are greater than the second abnormal threshold are added to an abnormal influence range set until the abnormal score of the adjacent node is lower than the second abnormal threshold or there is no further adjacent node, and the first abnormal threshold is greater than the second abnormal threshold; Based on each monitoring node in the abnormal source node and the abnormal influence range set, and combined with the pipe network topology connection relationship, an abnormal evidence chain is generated.

6. The big data-based heating network management data monitoring and analyzing method according to claim 1, characterized in that, The generation rule of the control instruction set is specifically: define a maximum temperature deviation value of the monitoring nodes in the abnormal evidence chain as , a maximum pressure deviation value as , and a maximum flow deviation value as , and set a temperature adjustment threshold value , a pressure adjustment threshold value , and a flow adjustment threshold value ; When the alarm level is the first level, and the abnormal evidence chain satisfies: and and ; And there is no abnormal influence range set distributed in different branch numbers, a standard operation control instruction for keeping the original operation parameters is outputted; When the alarm level is the second level, and the abnormal evidence chain satisfies: ； Any one condition in the above, and there is no abnormal influence range set distribution in different branch number, output single parameter adjustment type control instruction, wherein: Adjust the valve opening when, Adjust the pump group frequency when, Adjust the heat source outlet water temperature; When the alarm level is the third level, and the abnormal evidence chain satisfies: ； Any two or more conditions in the above, and the abnormal influence range set is distributed in different branch numbers, a multi-parameter joint adjustment type control instruction is outputted, and the valve opening, the pump group frequency and the heat source outlet water temperature are adjusted.

7. A big data-based heating network management data monitoring and analysis system, which executes a big data-based heating network management data monitoring and analysis method according to any one of claims 1-6. Comprise the following modules: A data acquisition module is configured to acquire heating network pipe operation data, meteorological data and pipe network topology information and generate a pipe network topology graph; A space-time baseline modeling module is configured to construct a space-time dynamic baseline model and output a should-be operation trajectory; A deviation decomposition module is configured to calculate a difference between a measured value of a monitoring node and the should-be operation trajectory and perform periodic and seasonal decomposition to obtain a net residual sequence; An improved USAD detection module is configured to input the should-be operation trajectory and the net residual sequence into an improved USAD model with a baseline conditioning adversarial reconstruction mechanism to obtain an abnormal score; A threshold alarm module is configured to set a multi-level alarm threshold based on a historical abnormal score distribution and trigger an alarm; An abnormal positioning module is configured to position an abnormal source node and an influence range based on a pipe network topology graph and an abnormal score, and generate an abnormal evidence chain; A control instruction generation module is configured to generate a control instruction set according to an alarm level and an abnormal evidence chain.

Images