Method and device for monitoring leakage of water network pipes of coal-fired power plants

Abstract

This application relates to the field of safety monitoring technology for coal-fired power plants, and specifically to a method and device for monitoring leaks in water network pipelines of coal-fired power plants. The method includes: using a hydraulic transient analysis model as a diagnostic capability benchmark and a PINN-LSTM time-series network as a capability carrier; through the collaborative design of feature alignment and physical constraints, the carrier network achieves leak location accuracy and leak quantity estimation capability comparable to the benchmark model under transient excitation, relying solely on steady-state operating data. This application focuses on the specific scenario of frequent fluctuations in operating conditions under the background of deep peak shaving in coal-fired power plants, achieving high-precision, all-time, and low-cost monitoring through a pure software upgrade path. This fills the technical gap in the transfer of diagnostic capabilities between heterogeneous monitoring modes, fundamentally solving the engineering problem of the inability to simultaneously achieve small sample sizes, multiple operating conditions, and high precision.

Description

Technical Field

[0001] This application relates to the field of safety monitoring technology for coal-fired power plants, and in particular to a method and device for monitoring leaks in water network pipelines of coal-fired power plants. Background Technology

[0002] Water pipelines in coal-fired power plants are subjected to high temperatures, high pressures, alternating thermal stress, and corrosion from acidic and alkaline media for extended periods. Leaks are highly concealed, sudden, and cause severe secondary disasters. Furthermore, due to the large plant area and complex pipeline network, precise leak location is often difficult. Statistics show that water system leaks account for over 15% of unplanned power plant outages, with micro-leakage (leakage volume < 1% of total flow) lasting an average of less than 40 minutes, resulting in an extremely narrow maintenance response window. Therefore, achieving real-time, high-precision, and early-warning leak monitoring is a core requirement for intelligent operation and maintenance of coal-fired power plants.

[0003] Existing leak detection technologies are mainly divided into three categories, each with its own structural defects: The first category consists of soft measurement methods based on steady-state operating data, including flow balance methods, pressure gradient methods, and model identification methods. These methods can operate continuously using existing monitoring systems, but they are extremely sensitive to fluctuations in operating conditions. Taking a 600MW unit as an example, during deep peak shaving, the pressure and flow fluctuations in the water system can reach ±30% of the steady-state value, and the false alarm rate of traditional threshold alarms is as high as 15%~20%. The detection rate for micro-leakage (<1% flow) is less than 40%, and the location error is generally greater than ±10m, which cannot meet the requirements for precise handling.

[0004] The second category comprises diagnostic methods based on transient excitation, including negative pressure wave methods, hydraulic transient analysis, and distributed fiber optic sensing technology. However, these methods rely on specific transient events for triggering and cannot provide continuous monitoring. The transient characteristics of slowly developing micro-leakage are smoothed by system damping, making them prone to being missed. If a sudden leak occurs within the steady-state operating window, diagnosis requires waiting for the next transient excitation. Distributed fiber optic temperature / vibration measurement technology can achieve sub-meter level positioning, but the cost of retrofitting a single unit is high, and its long-term stability under high temperature (>120℃) and strong electromagnetic environments has not been verified, making it unsuitable for low-cost upgrades of existing power plants.

[0005] The third category comprises data-driven methods based on deep learning, such as the hybrid 1D-CNN (Convolutional Neural Network)-LSTM (Long Short-Term Memory) model used for early detection of boiler leaks, demonstrating the potential of temporal networks in early warning. However, the purely data-driven approach has two major bottlenecks: first, sparse samples, where the scarcity of labeled samples makes it difficult for supervised models to converge; and second, a lack of physical consistency, where network outputs rely solely on statistical correlation, easily producing predictions that violate fundamental physical laws such as the conservation of mass and momentum, making it difficult to establish trust in engineering settings.

[0006] In summary, existing technologies cannot simultaneously achieve both continuity and accuracy. Furthermore, pure data-driven methods are hampered by sample sparsity and physical consistency, while distributed fiber optic methods are limited by cost and adaptability to operating conditions. These issues urgently need to be addressed. Summary of the Invention

[0007] This application provides a method and device for monitoring leaks in water network pipelines of coal-fired power plants, in order to solve the problems that existing technologies cannot balance continuity and accuracy, and that pure data-driven methods are hampered by sample sparsity and physical consistency, while distributed optical fiber methods are hampered by cost and adaptability to operating conditions.

[0008] The first aspect of this application provides a leakage monitoring device for water network pipelines in a coal-fired power plant, comprising the following steps: collecting transient calibration data and steady-state operation data corresponding to the water network pipelines of the target coal-fired power plant; determining a corresponding benchmark diagnostic rule sample set based on the transient calibration data and a benchmark source model in a pre-constructed heterogeneous network architecture; and generating, offline, feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set through the benchmark source model; inputting the steady-state operation data of the pipeline into the receiving network of the heterogeneous network architecture to output leakage diagnosis results; and based on the feature representations and diagnostic conclusions... Based on the leak diagnosis results, the differences in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set are minimized, so that the receiving network meets the preset capability transfer requirements. Based on the pre-constructed composite loss function and the benchmark diagnostic rule sample set, the receiving network is trained, and the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected. The current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results, wherein the monitoring results include the leakage probability index, the location coordinates of the leakage point, and the leakage development trend.

[0009] Based on the above technical means, the embodiments of this application use a hydraulic transient analysis model as the diagnostic capability benchmark and a PINN (Physics-Informed Neural Networks)-LSTM temporal network as the capability carrier. Through the collaborative design of feature alignment and physical constraints, the carrier network can obtain leakage location accuracy and leakage estimation capability comparable to the benchmark model under transient excitation, relying only on steady-state operating data. This fundamentally solves the engineering problem of not being able to simultaneously achieve small sample size, multiple operating conditions, and high accuracy.

[0010] Optionally, in one embodiment of this application, after inputting the current steady-state data stream into the trained receiving network to output the corresponding monitoring results, the method further includes: performing preset visualization processing on the monitoring results to obtain corresponding visualization information; generating corresponding multi-level early warning signals based on the monitoring results; and based on the multi-level early warning signals, pushing the visualization information and the monitoring results to a preset water balance control system through a preset standardized API interface for terminal display.

[0011] Based on the above technical means, the embodiments of this application present the monitoring results intuitively through visualization processing and multi-level early warning, and push them to the water balance control system through a standardized API interface to achieve data exchange and terminal display, thereby improving the efficiency of operation and maintenance response, facilitating rapid linkage and handling, and enhancing the practicality and integration of the system.

[0012] Optionally, in one embodiment of this application, the step of collecting transient calibration data and steady-state operation data corresponding to the water network pipeline of the target coal-fired power plant, and determining the corresponding benchmark diagnostic rule sample set based on the transient calibration data and the benchmark source model in the pre-constructed heterogeneous network architecture, and generating the feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set offline through the benchmark source model, includes: deploying pressure sensors that meet the preset model requirements at multiple key nodes in the target coal-fired power plant, acquiring the unit operation information corresponding to the water network pipeline of the target coal-fired power plant, and determining the system corresponding to the water network pipeline of the target coal-fired power plant based on the unit operation information. The system employs a unified acquisition mode. When the system acquisition mode is in steady-state monitoring mode, it acquires steady-state operating data of the pipeline corresponding to the target coal-fired power plant's water network pipeline at a first preset frequency. The steady-state operating data includes pressure, flow rate, temperature, pump and valve status, and unit load. When the system acquisition mode is in transient calibration mode, it actively triggers a step operation of the water pump to capture the deviation between the pressure wave propagation time difference and the actual / theoretical hydraulic filling time during the hydraulic transient process. Combined with on-site manual calibration, it generates a leak point location-leakage amount label sample that meets the preset accuracy requirements. The benchmark diagnostic rule sample set is then determined based on the leak point location-leakage amount label sample.

[0013] Based on the above technical means, the embodiments of this application adapt the sensor selection to the complex operating conditions of the power plant, and take into account the massive steady-state unlabeled data and high-precision transient labeled samples through the dual-mode acquisition architecture. This not only provides data support for network fine-tuning, but also enables the construction of accurate benchmark diagnostic rules, effectively breaking through the small sample dilemma and ensuring the reliability of diagnostic capability transfer.

[0014] Optionally, in one embodiment of this application, the step of generating the feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set offline through the benchmark source model includes: performing time-frequency domain transformation on the transient waveforms of leakage point pressure / flow in the benchmark diagnostic rule sample set to generate the corresponding leakage response feature map; acquiring historical steady-state data corresponding to the target coal-fired power plant water network pipeline, and performing sliding segmentation on the historical steady-state data based on a preset time window to obtain the corresponding segmentation result; calculating the corresponding theoretical hydraulic filling time based on the segmentation result and preset pipeline inherent parameters, and calculating the theoretical hydraulic filling time and the actual hydraulic filling time based on the theoretical hydraulic filling time. The residual of the filling time estimate is used to determine the corresponding hydraulic filling time deviation sequence; the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data of the target coal-fired power plant's water network pipeline are obtained; based on the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data, the equivalent hydraulic impedance matrix of the target coal-fired power plant's water network pipeline is solved, and the corresponding equivalent hydraulic impedance mode is determined according to the equivalent hydraulic impedance matrix; based on the leakage response feature spectrum, the hydraulic filling time deviation sequence, and the equivalent hydraulic impedance mode, the feature representation corresponding to the benchmark diagnostic rule sample set is obtained, and the diagnostic conclusion is generated according to the feature representation.

[0015] Based on the above technical means, the embodiments of this application construct an explicit representation layer of diagnostic rules, transforming implicit physical logic into learnable features, thereby accurately characterizing the leakage pattern, capturing the early fingerprints of micro-leaks, and quantifying the health status of pipelines, significantly improving the learning efficiency and diagnostic interpretability of the receiving network, and providing a reliable basis for the transfer of diagnostic capabilities.

[0016] Optionally, in one embodiment of this application, the step of inputting the pipeline steady-state operation data into the receiving network of the heterogeneous network architecture to output leakage diagnosis results, and minimizing the difference in diagnosis conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnosis rule sample set based on the feature representation, the diagnosis conclusion, and the leakage diagnosis results, so that the receiving network meets the preset capability transfer requirements, includes: constructing the benchmark source model based on the preset hydraulic transient analysis physical diagnosis model, and constructing a corresponding temporal feature extraction module based on an LSTM network with residual connections and multi-head self-attention mechanism, and constructing a corresponding physical embedding module and a diagnostic feature alignment layer based on the preset PINN physical information neural network and feature alignment adapter, respectively; according to the A target PINN-LSTM fusion temporal network is constructed using a physical embedding module, the diagnostic feature alignment layer, and the temporal feature extraction module. Based on the PINN-LSTM fusion temporal network, the receiving network is determined. The pipeline steady-state operation data is input into the temporal feature extraction module of the receiving network to extract the temporal features corresponding to the pipeline steady-state operation data. The temporal features are then input into the physical embedding module to generate corresponding physical constraints. The physical constraints are input into the diagnostic feature alignment layer to minimize the difference between the output diagnostic conclusions of the receiving network and the output diagnostic conclusions of the benchmark source model on the transient calibration data, as well as the difference in intermediate layer feature maps. This allows the benchmark diagnostic rules corresponding to the benchmark diagnostic rule sample set to be structurally transferred from the benchmark source model to the receiving network.

[0017] Based on the above technical means, this application embodiment transfers the high-precision capability of the transient physical diagnostic model to the steady-state time series monitoring model, and constructs a collaborative architecture of "high-precision physical benchmark - lightweight time series network" (i.e., benchmark source - receiving network heterogeneous architecture). This not only preserves the high precision and strong consistency of the physical model, but also enables steady-state continuous monitoring through lightweight PINN-LSTM. Furthermore, the physical embedding and feature alignment layers ensure reliable transfer of diagnostic knowledge. Transient-level accuracy can be achieved using only steady-state data, reducing the false alarm rate and improving stability.

[0018] Optionally, in one embodiment of this application, the receiving network is trained based on a pre-constructed composite loss function and the benchmark diagnostic rule sample set, and the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected. The current steady-state data stream is then input into the trained receiving network to output corresponding monitoring results. The monitoring results include a leakage probability index, leakage point location coordinates, and leakage rate trend. This includes: obtaining operating condition information corresponding to the target coal-fired power plant, determining multiple weight coefficients based on the operating condition information, and weighting and fusing a pre-defined supervised loss function, feature alignment loss function, and physical residual loss function based on the multiple weight coefficients. The composite loss function is constructed; the receiving network is trained based on the transient calibration data, the benchmark diagnostic rule sample set, and the composite loss function, combined with a preset two-stage progressive migration strategy; the current steady-state data stream corresponding to the target coal-fired power plant's water network pipeline is collected and input into the trained receiving network to generate the leakage probability index and the location coordinates of the leakage point; based on the correlation between the hydraulic filling time deviation sequence and the unit load, and combined with a preset time series prediction algorithm, the leakage development range and multi-level early warning information for the target time period are output, and the leakage development trend is determined based on the leakage development range and the multi-level early warning information.

[0019] Based on the above technical means, the embodiments of this application construct a composite loss function that takes into account accuracy, physical consistency and feature alignment, and use a two-stage progressive transfer strategy to train the receiving network. The trained receiving network is used to identify, locate and predict risk trends, thereby achieving continuous monitoring and early warning at all times, which has certain industry promotion value.

[0020] A second aspect of this application provides a leakage monitoring device for water network pipelines in a coal-fired power plant, comprising: a sample set construction module, used to collect transient calibration data and steady-state operation data corresponding to the target coal-fired power plant's water network pipelines, and based on the transient calibration data and a pre-constructed benchmark source model in a heterogeneous network architecture, determine a corresponding benchmark diagnostic rule sample set, and offline generate feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set through the benchmark source model; and a transfer module, used to input the pipeline steady-state operation data into the receiving network of the heterogeneous network architecture to output leakage diagnosis results, and based on the feature representations and diagnostic conclusions, determine a corresponding benchmark diagnostic rule sample set. The network and the benchmark source model are used to minimize the difference in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set, so that the receiving network meets the preset capability transfer requirements; the leakage monitoring module is used to train the receiving network based on the pre-constructed composite loss function and the benchmark diagnostic rule sample set, and to collect the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant, and input the current steady-state data stream into the trained receiving network to output the corresponding monitoring results, wherein the monitoring results include the leakage probability index, the location coordinates of the leakage point, and the leakage development trend.

[0021] Optionally, in one embodiment of this application, it further includes: a visualization module, used to perform preset visualization processing on the monitoring results after inputting the current steady-state data stream into the trained receiving network to output the corresponding monitoring results, so as to obtain corresponding visualization information; and a push module, used to generate corresponding multi-level early warning signals according to the monitoring results, and based on the multi-level early warning signals, push the visualization information and the monitoring results to a preset water balance control system through a preset standardized API interface for terminal display.

[0022] Optionally, in one embodiment of this application, the sample set construction module includes: a first acquisition unit, configured to deploy pressure sensors that meet preset model requirements at multiple key nodes in the target coal-fired power plant, acquire unit operation information corresponding to the water network pipeline of the target coal-fired power plant, and determine the system acquisition mode corresponding to the water network pipeline of the target coal-fired power plant based on the unit operation information; a steady-state monitoring unit, configured to acquire pipeline steady-state operation data corresponding to the water network pipeline of the target coal-fired power plant at a first preset frequency when the system acquisition mode is steady-state monitoring mode, wherein the pipeline steady-state operation data includes pressure, flow rate, temperature, pump and valve status, and unit load; and a transient calibration unit, configured to actively trigger a step operation of the water pump when the system acquisition mode is transient calibration mode, to capture the pressure wave propagation time difference and the deviation between the actual / theoretical hydraulic filling time in the hydraulic transient process, and generate a leak point location-leakage amount label sample that meets preset accuracy requirements in combination with on-site manual calibration, so as to determine the benchmark diagnostic rule sample set based on the leak point location-leakage amount label sample.

[0023] Optionally, in one embodiment of this application, the sample set construction module further includes: a first generation unit, configured to perform time-frequency domain transformation on the transient waveforms of leakage point pressure / flow in the benchmark diagnostic rule sample set to generate a corresponding leakage response feature map; a segmentation unit, configured to acquire historical steady-state data corresponding to the target coal-fired power plant water network pipeline, and perform sliding segmentation on the historical steady-state data based on a preset time window to obtain a corresponding segmentation result; and a calculation unit, configured to calculate the corresponding theoretical hydraulic filling time based on the segmentation result and preset pipeline inherent parameters, and determine the theoretical hydraulic filling time based on the residual between the theoretical hydraulic filling time and the estimated value of the actual hydraulic filling time. The system includes: a corresponding hydraulic filling time deviation sequence; a solution unit, used to acquire the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data corresponding to the target coal-fired power plant water network pipeline; and a solution unit, used to solve the equivalent hydraulic impedance matrix of the target coal-fired power plant water network pipeline based on the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data, so as to determine the corresponding equivalent hydraulic impedance mode according to the equivalent hydraulic impedance matrix; and a second generation unit, used to obtain the feature representation corresponding to the benchmark diagnostic rule sample set based on the leakage response feature spectrum, the hydraulic filling time deviation sequence, and the equivalent hydraulic impedance mode, and generate the diagnostic conclusion according to the feature representation.

[0024] Optionally, in one embodiment of this application, the migration module includes: a construction unit, configured to construct the baseline source model based on a preset hydraulic transient analysis physical diagnostic model, and construct a corresponding temporal feature extraction module based on an LSTM network with residual connections and multi-head self-attention mechanism, and construct a corresponding physical embedding module and a diagnostic feature alignment layer based on a preset PINN physical information neural network and feature alignment adapter, respectively; and a first determining unit, configured to construct a target PINN-LSTM fusion temporal network based on the physical embedding module, the diagnostic feature alignment layer and the temporal feature extraction module, so as to construct a target PINN-LSTM fusion temporal network based on the PINN-LSTM fusion... The system comprises a time-series network to determine the receiving network; an extraction unit to input the pipeline steady-state operation data into the time-series feature extraction module of the receiving network to extract the time-series features corresponding to the pipeline steady-state operation data, and input the time-series features into the physical embedding module to generate corresponding physical constraints; and a minimization unit to input the physical constraints into the diagnostic feature alignment layer to minimize the difference between the output diagnostic conclusions of the receiving network and the output diagnostic conclusions of the benchmark source model on the transient calibration data and the difference in intermediate layer feature maps, so that the benchmark diagnostic rules corresponding to the benchmark diagnostic rule sample set are structurally transferred from the benchmark source model to the receiving network.

[0025] Optionally, in one embodiment of this application, the leakage monitoring module includes: a second acquisition unit, configured to acquire operating condition information corresponding to the target coal-fired power plant, determine multiple weight coefficients based on the operating condition information, and perform weighted fusion of a preset supervised loss function, a feature alignment loss function, and a physical residual loss function based on the multiple weight coefficients to construct the composite loss function; a training unit, configured to train the receiving network based on the transient calibration data, the benchmark diagnostic rule sample set, and the composite loss function, combined with a preset two-stage progressive migration strategy; a collection unit, configured to collect the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant, and input the current steady-state data stream into the trained receiving network to generate the leakage probability index and the location coordinates of the leakage point; and a second determination unit, configured to output the leakage development range and multi-level early warning information for the target time period based on the correlation between the hydraulic filling time deviation sequence and the unit load, combined with a preset time series prediction algorithm, and determine the leakage development trend based on the leakage development range and the multi-level early warning information.

[0026] A third aspect of this application provides an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for monitoring leakage in water pipelines of coal-fired power plants as described in the above embodiments.

[0027] A fourth aspect of this application provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described method for monitoring leaks in water pipelines of coal-fired power plants.

[0028] A fifth aspect of this application provides a computer program product, including a computer program that is executed to implement the above-described method for monitoring leaks in water pipelines of coal-fired power plants.

[0029] Therefore, the embodiments of this application have the following beneficial effects: The embodiments of this application can collect transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipeline, and determine the corresponding benchmark diagnostic rule sample set based on the transient calibration data and the benchmark source model in the pre-constructed heterogeneous network architecture. Furthermore, the feature representation and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set are generated offline using the benchmark source model. The steady-state operation data of the pipeline is input into the receiving network of the heterogeneous network architecture to output leakage diagnosis results. Based on the feature representation, diagnostic conclusions, and leakage diagnosis results, the differences in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set are minimized, ensuring that the receiving network meets preset capability transfer requirements. Based on the pre-constructed composite loss function and the benchmark diagnostic rule sample set, the receiving network is trained, and the current steady-state data stream corresponding to the target coal-fired power plant's water network pipeline is collected. The current steady-state data stream is input into the trained receiving network to output corresponding monitoring results, including a leakage probability index, leakage point location coordinates, and leakage rate trend. This application uses a hydraulic transient analysis model as the diagnostic capability benchmark and a PINN-LSTM time-series network as the capability carrier. Through the collaborative design of feature alignment and physical constraints, the carrier network achieves leakage location accuracy and leakage estimation capability comparable to the benchmark model under transient excitation, relying solely on steady-state operating data. This fundamentally solves the engineering challenge of simultaneously achieving small sample sizes, multiple operating conditions, and high accuracy. Thus, it addresses the limitations of existing technologies in balancing continuity and accuracy, as well as the constraints of pure data-driven methods on sample sparsity and physical consistency, and the limitations of distributed fiber optic methods on cost and operating condition adaptability.

[0030] Additional aspects and advantages of this application will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this application. Attached Figure Description

[0031] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart illustrating a method for monitoring leaks in water pipelines of a coal-fired power plant according to an embodiment of this application. Figure 2 This is an example diagram of a leakage monitoring device for water network pipelines in a coal-fired power plant according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application.

[0032] Among them, 10-Leakage monitoring device for water network pipelines in coal-fired power plants; 100-Sample set construction module, 200-Migration module, 300-Leakage monitoring module; 301-Memory, 302-Processor, 303-Communication interface. Detailed Implementation

[0033] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0034] The following describes a method and apparatus for monitoring leakage in water network pipelines of coal-fired power plants, based on embodiments of this application, with reference to the accompanying drawings. To address the problems mentioned in the background, this application provides a method for monitoring leaks in water network pipelines of coal-fired power plants. In this method, transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipelines are collected. Based on the transient calibration data and a pre-constructed benchmark source model in a heterogeneous network architecture, a corresponding benchmark diagnostic rule sample set is determined. The feature representation and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set are generated offline using the benchmark source model. The steady-state operation data of the pipelines are input into the receiving network of the heterogeneous network architecture to output leak diagnosis results. Based on the feature representation, diagnostic conclusions, and leak diagnosis results, the differences in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set are minimized, ensuring that the receiving network meets preset capability transfer requirements. The receiving network is trained based on a pre-constructed composite loss function and the benchmark diagnostic rule sample set. The current steady-state data stream corresponding to the target coal-fired power plant's water network pipelines is collected and input into the trained receiving network to output corresponding monitoring results. These monitoring results include a leak probability index, leak point location coordinates, and leakage rate trend. This application uses a hydraulic transient analysis model as the diagnostic capability benchmark and a PINN-LSTM time-series network as the capability carrier. Through the collaborative design of feature alignment and physical constraints, the carrier network achieves leakage location accuracy and leakage estimation capability comparable to the benchmark model under transient excitation, relying solely on steady-state operating data. This fundamentally solves the engineering challenge of simultaneously achieving small sample sizes, multiple operating conditions, and high accuracy. Thus, it addresses the limitations of existing technologies in balancing continuity and accuracy, as well as the constraints of pure data-driven methods on sample sparsity and physical consistency, and the limitations of distributed fiber optic methods on cost and operating condition adaptability.

[0035] Specifically, Figure 1 This is a flowchart illustrating a method for monitoring leaks in water pipelines of a coal-fired power plant, as provided in an embodiment of this application.

[0036] like Figure 1 As shown, the method for monitoring leaks in the water network pipelines of this coal-fired power plant includes the following steps: In step S101, transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipeline are collected. Based on the transient calibration data and the benchmark source model in the pre-built heterogeneous network architecture, the corresponding benchmark diagnostic rule sample set is determined. Through the benchmark source model, the feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set are generated offline.

[0037] This application embodiment can first deploy a multimodal sensor network to synchronously collect pipeline steady-state operation data (i.e., steady-state time-series data) and transient calibration data. The transient calibration data can be obtained by actively triggering the hydraulic transient process, and a high-precision leak point location-leakage amount labeled sample (i.e., transient label sample or transient calibration sample) can be generated based on the hydraulic transient analysis physical diagnostic model. The sample is then used to form a benchmark diagnostic rule (i.e., benchmark diagnostic rule sample set). Secondly, the embodiments of this application can construct a heterogeneous network architecture for diagnostic capability transfer. This architecture includes a reference source model and a receiving network. The reference source model is a hydraulic transient analysis physical diagnostic model, which provides its feature representation and output diagnostic conclusions on transient calibration samples offline.

[0038] Therefore, the embodiments of this application synchronously collect steady-state and transient data through a multimodal sensor network, construct high-precision benchmark diagnostic rules, and build a benchmark source-receiving network heterogeneous architecture. This not only ensures diagnostic accuracy and reliability based on the physical model, but also lays the data and model foundation for the transfer of diagnostic capabilities, breaking through the limitations of small samples and realizing high-precision leakage monitoring under steady-state data.

[0039] Optionally, in one embodiment of this application, transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipeline are collected. Based on the transient calibration data and a pre-built benchmark source model in a heterogeneous network architecture, a corresponding benchmark diagnostic rule sample set is determined. Furthermore, through the benchmark source model, feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set are generated offline. This includes: deploying pressure sensors that meet preset model requirements at multiple key nodes in the target coal-fired power plant; acquiring unit operation information corresponding to the target coal-fired power plant's water network pipeline; and determining the system corresponding to the target coal-fired power plant's water network pipeline based on the unit operation information. The system employs a unified acquisition mode. When the system acquisition mode is in steady-state monitoring mode, it acquires steady-state operating data of the pipeline corresponding to the water network pipeline of the target coal-fired power plant at a first preset frequency. The steady-state operating data of the pipeline includes pressure, flow rate, temperature, pump and valve status, and unit load. When the system acquisition mode is in transient calibration mode, it actively triggers the step operation of the water pump to capture the deviation between the pressure wave propagation time difference and the actual / theoretical hydraulic filling time during the hydraulic transient process. Combined with on-site manual calibration, it generates a leak point location-leakage amount label sample that meets the preset accuracy requirements. The benchmark diagnostic rule sample set is then determined based on the leak point location-leakage amount label sample.

[0040] It should be noted that the embodiments of this application can deploy pressure sensors (sampling frequency ≥500Hz in transient calibration mode and 1Hz in steady-state monitoring mode) at key nodes of the circulating water, ash removal, desulfurization and industrial water systems of coal-fired power plants. The sensor selection is adapted to the requirements of hydraulic transient analysis, with a dynamic response time ≤2ms and a natural frequency ≥50kHz. Ultrasonic flow meters and temperature sensors are also used, and the sensor selection is adapted to high temperature and high pressure and strong interference conditions (protection level ≥IP6 / 7, operating temperature -40℃~120℃).

[0041] In the embodiments of this application, the system operates on a dual-mode acquisition architecture, as described below: 1. Steady-state monitoring mode: During normal operation of the unit, steady-state time-series data such as pressure, flow, temperature, pump and valve status, and unit load are continuously collected at a frequency of 10Hz to form a massive unlabeled dataset (one time section is collected per second for a single unit, with a daily increase of ≥86,400 data entries), which is used to support the self-supervised fine-tuning of the network.

[0042] 2. Transient Calibration Mode: During preset windows such as unit start-up and shutdown, and periodic maintenance, the pump is actively triggered to perform step operation, capturing the pressure wave propagation time difference and the deviation between the actual / theoretical hydraulic filling time during the hydraulic transient process. Combined with on-site manual calibration, a high-precision "leak point location - leakage amount" tagged sample is generated. The sample generated in this mode has a positioning accuracy of ≤±2m and a leakage amount estimation error of ≤5%, which constitutes the "benchmark diagnostic rule" for subsequent diagnostic capability transfer. In engineering practice, each unit only needs to complete 10-20 sets of transient calibration sample collection to meet the transfer requirements, completely breaking through the small sample dilemma.

[0043] Therefore, the embodiments of this application adapt to the complex operating conditions of power plants by selecting appropriate sensors, and take into account both massive steady-state unlabeled data and high-precision transient labeled samples through a dual-mode acquisition architecture. This not only provides data support for network fine-tuning, but also enables the construction of accurate benchmark diagnostic rules, effectively breaking through the small sample dilemma and ensuring the reliability of diagnostic capability transfer.

[0044] Optionally, in one embodiment of this application, feature representations and diagnostic conclusions corresponding to a benchmark diagnostic rule sample set are generated offline using a benchmark source model. This includes: performing time-frequency domain transformation on the transient waveforms of pressure / flow at the leakage point in the benchmark diagnostic rule sample set to generate corresponding leakage response feature maps; acquiring historical steady-state data corresponding to the water network pipeline of the target coal-fired power plant, and performing sliding segmentation on the historical steady-state data based on a preset time window to obtain corresponding segmentation results; calculating the corresponding theoretical hydraulic filling time based on the segmentation results and preset pipeline inherent parameters, and calculating the theoretical hydraulic filling time and actual... The residual of the estimated hydraulic filling time is used to determine the corresponding hydraulic filling time deviation sequence; pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data of the target coal-fired power plant's water network pipeline are obtained; based on the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data, the equivalent hydraulic impedance matrix of the target coal-fired power plant's water network pipeline is solved, and the corresponding equivalent hydraulic impedance mode is determined according to the equivalent hydraulic impedance matrix; based on the leakage response feature spectrum, hydraulic filling time deviation sequence, and equivalent hydraulic impedance mode, the feature representation corresponding to the benchmark diagnostic rule sample set is obtained, and a diagnostic conclusion is generated based on the feature representation.

[0045] Understandably, unlike traditional strategies that directly use the original waveform or physical parameters as input features, this application embodiment can construct an explicit representation layer for diagnostic rules to transform the implicit diagnostic logic of the hydraulic transient analysis model into explicit features that can be effectively learned by the receiving network. The specific process is as follows: (1) Leakage response feature map: The pressure / flow transient waveforms at the precise location of the leak point are transformed in the time and frequency domain (such as wavelet packet decomposition and Hilbert-Huang transform) to construct a standardized feature basis function library of the leak pattern. This feature not only describes the waveform shape, but also marks the corresponding leak location and leakage amount, forming a pairing rule unit of "waveform-diagnosis conclusion".

[0046] (2) Hydraulic filling time deviation sequence: The historical steady-state data is divided into 10-minute windows (i.e. time windows). The theoretical hydraulic filling time is calculated based on the inherent parameters of the pipeline (pipe diameter, material, length). The residual between the actual hydraulic filling time estimate and the actual hydraulic filling time estimate constitutes the deviation time sequence. This sequence is the only physical fingerprint of micro-leakage in steady-state operation. Its slow drift trend has more early diagnostic value than instantaneous pressure fluctuation.

[0047] (3) Equivalent hydraulic impedance mode: Based on pressure wave velocity, pipe segment geometric characteristics and transient pressure fluctuation data, the equivalent hydraulic impedance matrix of the pipeline system is solved, and the abstract transient response is transformed into a quantifiable indicator of pipe segment health status.

[0048] Furthermore, the embodiments of this application can use the above three types of features to constitute an explicit representation of the "benchmark diagnostic rules", which is not only the basis for the benchmark source model to output diagnostic conclusions, but also the diagnostic knowledge object that the receiving network needs to focus on learning.

[0049] Therefore, the embodiments of this application construct an explicit representation layer of diagnostic rules, transforming implicit physical logic into learnable features, thereby accurately characterizing the leakage pattern, capturing early fingerprints of micro-leaks, and quantifying the health status of pipelines, significantly improving the learning efficiency and interpretability of the receiving network, and providing a reliable basis for the transfer of diagnostic capabilities.

[0050] In step S102, the pipeline steady-state operation data is input into the receiving network of the heterogeneous network architecture to output the leakage diagnosis result. Based on the feature representation, diagnosis conclusion and leakage diagnosis result, the difference between the diagnosis conclusion of the receiving network and the benchmark source model on the benchmark diagnosis rule sample set and the difference between the intermediate layer features are minimized so that the receiving network meets the preset capability transfer requirements.

[0051] Furthermore, embodiments of this application can construct a heterogeneous network architecture of "baseline source-receiving network" (i.e., heterogeneous network architecture for diagnostic capability transfer) based on a baseline source model and a receiving network. The receiving network is a PINN-LSTM fusion model, with steady-state operating data as input and leakage diagnosis results as output. Secondly, embodiments of this application can minimize the differences in output diagnostic conclusions and intermediate layer features between the receiving network and the baseline source model on transient samples through a diagnostic feature alignment layer, so as to transfer the positioning accuracy and leakage estimation capability of the baseline source model to the receiving network.

[0052] Therefore, the embodiments of this application construct a heterogeneous architecture of "baseline source-receiving network" and adopt PINN-LSTM to fuse the receiving network, and combine it with a diagnostic feature alignment layer, so that the high-precision diagnostic capability of the baseline source model can be transferred to the receiving network, so that high-precision leakage diagnosis results can be output using only steady-state data, thereby improving the practicality of diagnosis.

[0053] Optionally, in one embodiment of this application, steady-state pipeline operation data is input into a heterogeneous network architecture receiving network to output leakage diagnosis results. Based on feature representation, diagnostic conclusions, and leakage diagnosis results, the differences in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set are minimized, so that the receiving network meets preset capability transfer requirements. This includes: constructing a benchmark source model based on a preset hydraulic transient analysis physical diagnostic model, constructing a corresponding temporal feature extraction module based on an LSTM network with residual connections and multi-head self-attention mechanism, and constructing corresponding physical embedding modules and diagnostic feature pairs based on preset PINN physical information neural networks and feature alignment adapters, respectively. The process involves three layers: a target PINN-LSTM fusion temporal network is constructed based on the physical embedding module, a diagnostic feature alignment layer, and a temporal feature extraction module. A receiving network is then determined based on this PINN-LSTM fusion temporal network. Steady-state pipeline operation data is input into the temporal feature extraction module of the receiving network to extract the corresponding temporal features. These temporal features are then input into the physical embedding module to generate corresponding physical constraints. The physical constraints are then input into the diagnostic feature alignment layer to minimize the differences in diagnostic conclusions between the receiving network output and the benchmark source model on transient calibration data, as well as the differences in intermediate layer feature maps. This allows the benchmark diagnostic rules corresponding to the benchmark diagnostic rule sample set to be structurally transferred from the benchmark source model to the receiving network.

[0054] It should be noted that the embodiments of this application use the hydraulic transient analysis physical diagnostic model as the benchmark source model. This model is based on the one-dimensional compressible fluid transient flow equation (i.e., the partial differential equations of pipeline hydraulic control, including the continuity equation and the momentum equation) and the empirical mapping relationship of "hydraulic filling time deviation - leakage amount". It achieves leakage location and quantification by solving the transient inverse problem. The core advantages of the benchmark source model are high diagnostic accuracy and strong physical consistency. Its core limitation is that it depends on transient excitation and cannot be continuously monitored. In the embodiments of this application, the benchmark source model does not need to run online in real time. It only needs to provide its feature representation and output diagnostic conclusions on transient calibration samples offline, as a "standard example" for network learning.

[0055] Specifically, embodiments of this application can design a lightweight PINN-LSTM fusion timing network (i.e., a PINN-LSTM fusion model) and use it as a receiving network. This network includes the following modules: (1) Temporal feature extraction module: an LSTM network with residual connections and multi-head self-attention mechanism is adopted, and the input is steady-state time series data (pressure, flow, unit load, etc.); among which, the self-attention mechanism can adaptively focus on special periods such as sudden load changes and pump valve switching that are easy to cover up leakage characteristics, dynamically adjust the temporal weights, and reduce false alarms due to operating condition disturbances; residual connections ensure the gradient stability of deep network training.

[0056] (2) Physical Embedding Module: The embodiments of this application can construct a Physical Constraint Neural Network (PINN) submodule, which encodes the partial differential equation of pipeline hydraulic control and the mapping relationship of "hydraulic filling time-leakage" into soft constraints of network output; compared with the technical means of using physical loss to constrain the error between network output and the true value of physical response in the prior art, the physical constraints of the embodiments of this application are used to constrain the error between the hidden layer representation of the network and the physical benchmark derived from hydraulic transient analysis.

[0057] It should be noted that the partial differential equations for pipeline hydraulic control in this embodiment include continuity equations and momentum equations. The derivative of the physical embedding module output with respect to the input physical quantity is calculated by automatic differentiation technology and compared with the theoretical derivative to form physical residual loss. The mapping relationship of "hydraulic filling time-leakage" is encoded as the nonlinear transformation basis of the output layer of the physical embedding module.

[0058] (3) Diagnostic feature alignment layer: As the core innovative component of this application embodiment, the alignment layer sets feature alignment adapters in the output layer and the intermediate hidden layer of the receiving network respectively. By minimizing the difference between the output diagnosis conclusions of the receiving network and the benchmark source model on transient samples (KL divergence) and the difference in the feature maps of the intermediate layer (mean square error), the benchmark diagnosis rules are structurally transferred from the benchmark source to the receiving network. This layer enables the receiving network to approximate the high-precision diagnosis results that the benchmark source model can obtain under transient excitation when only steady-state time series data are input.

[0059] Therefore, the embodiments of this application transfer the high-precision capability of the transient physical diagnostic model to the steady-state time-series monitoring model, and construct a collaborative architecture of "high-precision physical benchmark - lightweight time-series network" (i.e., benchmark source - receiving network heterogeneous architecture). This not only preserves the high precision and strong consistency of the physical model, but also enables steady-state continuous monitoring through lightweight PINN-LSTM. Furthermore, the physical embedding and feature alignment layers ensure reliable transfer of diagnostic knowledge. Transient-level accuracy can be achieved using only steady-state data, reducing the false alarm rate and improving stability.

[0060] In step S103, based on the pre-constructed composite loss function and the benchmark diagnostic rule sample set, the receiving network is trained, and the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected. The current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results. The monitoring results include the leakage probability index, the location coordinates of the leakage point, and the development trend of the leakage amount.

[0061] Subsequently, embodiments of this application can utilize a composite loss function consisting of supervised loss, feature alignment loss, and physical residual loss, and employ a two-stage progressive transfer strategy to complete the training of the receiving network, and deploy the trained receiving network; furthermore, embodiments of this application can access the real-time steady-state data stream (i.e., the current steady-state data stream) into the trained receiving network to output the leakage probability index, the location coordinates of the leakage point, and the development trend of the leakage amount.

[0062] Therefore, the embodiments of this application can integrate physical information neural networks and temporal deep networks to achieve high-precision leak monitoring at all times through cross-modal diagnostic knowledge transfer. It is well applicable to online identification, accurate location and risk prediction of leaks in complex water network pipelines such as circulating water, ash removal, desulfurization and industrial water systems in coal-fired power plants.

[0063] Optionally, in one embodiment of this application, a receiving network is trained based on a pre-constructed composite loss function and a benchmark diagnostic rule sample set. Current steady-state data streams corresponding to the water network pipelines of the target coal-fired power plant are collected, and these data streams are input into the trained receiving network to output corresponding monitoring results. The monitoring results include a leakage probability index, leakage point location coordinates, and leakage rate trend. This includes: obtaining operating condition information corresponding to the target coal-fired power plant, determining multiple weight coefficients based on the operating condition information, and applying these weight coefficients to a pre-defined supervised loss function, feature alignment loss function, and physical residual loss function. Weighted fusion is used to construct a composite loss function. Based on transient calibration data, a benchmark diagnostic rule sample set, and the composite loss function, and combined with a preset two-stage progressive migration strategy, the receiving network is trained. The current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected and input into the trained receiving network to generate a leakage probability index and leakage point location coordinates. Based on the correlation between the hydraulic filling time deviation sequence and the unit load, and combined with a preset time series prediction algorithm, the leakage development range and multi-level early warning information for the target time period are output, and the leakage development trend is determined based on the leakage development range and multi-level early warning information.

[0064] As one possible approach, embodiments of this application can balance task accuracy, physical consistency, and feature alignment by constructing a composite loss function, which mainly includes the following parts: 1. Monitoring loss: On a small number of transient tag samples, calculate the mean square error between the predicted values ​​of leakage location and leakage amount output by the receiving network and the actual calibration values ​​to ensure that the migration process does not deviate from the real physical world.

[0065] 2. Feature Alignment Loss: On transient label samples, calculate the KL divergence between the diagnostic conclusion of the inheriting network output layer and the output layer of the benchmark source model, as well as the cosine similarity loss of the feature maps of the intermediate hidden layers; this loss term is the core carrier of diagnostic capability transfer, ensuring that the inheriting network inherits the diagnostic logic of the benchmark source.

[0066] 3. Physical residual loss: On a large number of unlabeled steady-state samples, the residuals between the pressure / flow predictions output by the PINN module and the pipeline hydraulic control equations, as well as the fitting error between the hydraulic filling time deviation estimate and the theoretical mapping relationship, are calculated. This loss term ensures that the receiving network maintains physical consistency on unsupervised data and avoids "diagnostic capability drift" during the migration process.

[0067] Therefore, the embodiments of this application can obtain the operating condition information corresponding to the target coal-fired power plant, determine the weight coefficients corresponding to different loss functions, and perform weighted fusion of the supervision loss function, feature alignment loss function and physical residual loss function according to multiple weight coefficients to construct a composite loss function. The weight coefficients can be dynamically adjusted according to the power plant operating conditions.

[0068] Secondly, the embodiments of this application can employ a two-stage progressive migration strategy to train the receiving network, the specific process of which is as follows: 1. First Stage: Offline Injection (Diagnostic Rule Injection). On the transient calibration sample set, the receiving network is pre-trained using a combination of supervised loss and feature alignment loss. The goal of this stage is to inject the expert diagnostic capabilities of the benchmark source model into the parameters of the receiving network, enabling the receiving network to initially establish a nonlinear mapping from steady-state temporal features to high-precision leak location.

[0069] 2. Second Stage: Online Adaptation (Consolidation of Diagnostic Rules). On a massive amount of unlabeled steady-state samples, self-supervised fine-tuning is performed, primarily using physical residual loss and feature alignment loss. This stage has two objectives: first, to strengthen the self-consistency between the network output and hydraulic laws through physical constraints; and second, to continuously align the network with the baseline source feature space through feature alignment loss, preventing network drift. This two-stage strategy ensures the integrity and stability of diagnostic capability transfer.

[0070] After the network deployment is completed and training is complete, this embodiment of the application can access the real-time steady-state monitoring data stream to achieve continuous monitoring throughout the day and output the following three levels of diagnostic results (i.e., monitoring results): (1) Leakage probability index: a continuous value of 0-1, reflecting the possibility of leakage in each monitored pipe section; according to actual measurement, the micro-leakage detection rate of the embodiment of this application for leakage amount ≥0.5% of the total flow is 91%, which is significantly improved compared with the traditional steady state method (detection rate <40%), and the early warning advance is increased by more than 40 minutes on average compared with the threshold method.

[0071] (2) Leakage point location coordinates: The model directly outputs the distance between the leakage point and the upstream node. In this embodiment, the positioning accuracy of the reference source model can be inherited through diagnostic capability transfer. The positioning error in the steady-state monitoring scenario is ≤ ±5m, which is close to the accuracy of the reference source model in the transient calibration scenario (±2m), while the positioning error of the traditional steady-state method is generally above ±10m.

[0072] (3) Dynamic estimation and trend prediction of leakage: Based on the correlation between the hydraulic filling time deviation sequence and the unit load, this application embodiment can output the leakage development range and three-level early warning (attention / alarm / emergency) for the next hour (i.e. the target time period) and the multi-level early warning signal (i.e., multi-level early warning signal) based on the time deviation sequence of hydraulic filling and the unit load. This function enables the operation and maintenance mode to shift from "post-event handling" to "pre-event intervention".

[0073] Therefore, the embodiments of this application construct a composite loss function that takes into account accuracy, physical consistency and feature alignment, and use a two-stage progressive transfer strategy to train the receiving network. The trained receiving network is then used to identify, locate and predict risk trends, thereby enabling continuous monitoring and early warning at all times, which has certain industry promotion value.

[0074] Optionally, in one embodiment of this application, after inputting the current steady-state data stream into the trained receiving network to output the corresponding monitoring results, the method further includes: performing preset visualization processing on the monitoring results to obtain corresponding visualization information; generating corresponding multi-level early warning signals based on the monitoring results; and based on the multi-level early warning signals, pushing the visualization information and monitoring results to a preset water balance control system through a preset standardized API interface for terminal display.

[0075] Subsequently, the embodiments of this application can perform preset visualization processing on the monitoring results. The embodiments of this application can support automatic plotting of the coordinates of the leak point through the built-in GIS pipeline layout map, and perform three-level early warning and graded push (central control screen / operation and maintenance terminal / mobile APP) operation.

[0076] The embodiments of this application can provide a standardized API interface to push the leakage status vector (leakage probability, location, leakage amount, and development trend) to the water balance control system in real time, thereby realizing a monitoring-control closed loop.

[0077] Therefore, the embodiments of this application present monitoring results intuitively through visualization processing and multi-level early warning, and push them to the water balance control system through a standardized API interface to achieve data exchange and terminal display, thereby improving operation and maintenance response efficiency, facilitating rapid linkage and handling, and enhancing the system's practicality and integration.

[0078] Furthermore, this application can also construct a corresponding coal-fired power plant water network pipeline leakage monitoring system based on the execution logic of the aforementioned coal-fired power plant water network pipeline leakage monitoring method. The coal-fired power plant water network pipeline leakage monitoring system of this application mainly includes a perception layer, a transient calibration engine, a fusion modeling layer, a decision output layer, and a terminal display layer, as described below: 1. Sensing Layer: Used to collect and preprocess steady-state and transient operating data of the pipeline, including high-frequency dynamic pressure sensors, ultrasonic flow meters and temperature sensors; among them, the dynamic response time of the pressure sensor is ≤2ms, the natural frequency is ≥50kHz, the sampling frequency of the transient calibration mode is ≥500Hz, and the sampling frequency of the steady-state monitoring mode is 1Hz; the edge computing unit is responsible for data cleaning and standardization.

[0079] 2. Transient Calibration Engine: Utilizes the hydraulic transient analysis physical diagnostic model to generate a baseline diagnostic rule sample set offline. This engine is only enabled during the model initialization phase or power plant maintenance window and does not participate in real-time monitoring. 3. Fusion Modeling Layer: Deploys the PINN-LSTM fusion model and diagnostic capability transfer training framework, with built-in physical constraint calculation unit, diagnostic feature alignment unit and time series feature extraction unit to support online incremental updates of model parameters and adapt to the operating condition drift caused by long-term operation of power plants; 4. Decision Output Layer: Visualizes the monitoring results and generates three-level early warning signals; This application embodiment has a built-in GIS pipeline layout map, supports automatic plotting of leak point coordinates, and can push three-level early warnings in stages (central control screen / maintenance terminal / mobile APP). 5. Terminal display layer: integrated into the power plant's existing SIS / MIS system, and pushes leakage status vectors (leakage probability, location, leakage amount, and development trend) to the water balance control system through a standardized API interface to achieve a closed loop of monitoring and control.

[0080] In summary, the embodiments of this application pioneer a new paradigm for the transfer of diagnostic capabilities from heterogeneous monitoring models, demonstrating significant novelty and inventiveness. Existing leak monitoring technologies are all "enhancements within a single monitoring model," i.e., optimization within the physical model or the data model, without involving a cross-paradigm capability inheritance architecture for transferring the expert diagnostic capabilities of transient physical diagnostic models to steady-state time-series sensing models. The embodiments of this application, for the first time, construct a heterogeneous capability transfer channel of "physical diagnostic benchmark source - time-series sensing receiving network," enabling deep coupling between two technical approaches that have long been considered "complementary but not compatible." This paradigm innovation is not a simple combination of existing technologies, but rather opens up a new technological path in the field of leak monitoring.

[0081] Secondly, this application's embodiments completely overcome the small sample problem, significantly improving the feasibility of engineering implementation. It is understandable that the scarcity of leak samples from coal-fired power plants is a fundamental obstacle to the implementation of artificial intelligence technology. This application's embodiments, through diagnostic capability transfer, migrate the high-precision diagnostic capabilities of the hydraulic transient analysis model on 10-20 sets of transient calibration samples to a time-series network capable of processing massive amounts of steady-state data. This allows the latter to achieve diagnostic accuracy close to that of a baseline source model without requiring a large number of historical leak samples. This technique fundamentally avoids dependence on scarce leak samples, improving engineering applicability by two orders of magnitude compared to purely data-driven solutions (which require hundreds of tagged samples). Simultaneously, this application's embodiments reuse existing sensor and server resources, with a single unit upgrade investment of less than 300,000 yuan, offering a significant cost advantage compared to distributed fiber optic solutions (>2 million yuan).

[0082] Furthermore, the multi-condition adaptive and micro-leakage early detection capabilities of the embodiments in this application significantly surpass existing technologies. Specifically, the embodiments in this application introduce a multi-head self-attention mechanism into LSTM and combine it with physical constraints and feature alignment. The receiving network can adaptively identify operating condition disturbances such as sudden load changes and pump / valve switching, with a false alarm rate of ≤3%, far lower than the traditional steady-state method (15%-20%). At the same time, based on the tracking capability of the hydraulic filling time deviation sequence, the embodiments in this application achieve a micro-leakage detection rate of 91% for leakage amounts ≥0.5% of the total flow, which is significantly improved compared to the Chongqing University 1D-CNN-LSTM pure data-driven scheme (detection rate of approximately 75%). The early warning lead time is increased by an average of more than 40 minutes compared to the traditional threshold method, providing maintenance personnel with a valuable response window.

[0083] Subsequently, this application's embodiments construct a closed-loop system for the entire lifecycle of "transient calibration - steady-state monitoring," filling a gap in technical standards. This application's embodiments, through a high-precision transient calibration strategy combined with continuous steady-state monitoring, transform calibration results into continuous operation and maintenance capabilities. Both units share a hydraulic transient analysis engine, achieving long-term benefits from a single calibration. This collaborative architecture not only improves the overall efficiency of the monitoring system but also defines a novel equipment operation and maintenance model of "transient calibration - steady-state monitoring" at the technical standard level, possessing industry-wide promotion value.

[0084] Furthermore, the leakage status vector (probability, location, leakage amount, trend) output by the embodiments of this application can be pushed to the water balance control system in real time via a standardized API, enabling the control layer to upgrade from passively responding to leakage alarms to actively predicting leakage trends and dynamically optimizing water intake strategies. Simulation calculations show that this closed-loop system can reduce the average fresh water replenishment volume for a single leakage incident by 40%-60% and reduce the risk of unplanned shutdowns by more than 35%, demonstrating significant economic and safety benefits.

[0085] According to the leakage monitoring method for water network pipelines in coal-fired power plants proposed in this application, transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipelines are collected. Based on the transient calibration data and a pre-constructed benchmark source model in a heterogeneous network architecture, a corresponding benchmark diagnostic rule sample set is determined. The feature representation and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set are generated offline using the benchmark source model. The steady-state operation data of the pipelines are input into the receiving network of the heterogeneous network architecture to output leakage diagnostic results. Based on the feature representation, diagnostic conclusions, and leakage diagnostic results, the difference in diagnostic conclusions and intermediate layer features between the receiving network and the benchmark source model on the benchmark diagnostic rule sample set is minimized, ensuring that the receiving network meets preset capability transfer requirements. Based on a pre-constructed composite loss function and the benchmark diagnostic rule sample set, the receiving network is trained. The current steady-state data stream corresponding to the target coal-fired power plant's water network pipelines is collected and input into the trained receiving network to output corresponding monitoring results. The monitoring results include a leakage probability index, leakage point location coordinates, and leakage rate trend. This application uses a hydraulic transient analysis model as the diagnostic capability benchmark and a PINN-LSTM time series network as the capability carrier. Through the collaborative design of feature alignment and physical constraints, the carrier network can achieve leakage location accuracy and leakage estimation capability comparable to the benchmark model under transient excitation, relying only on steady-state operating data. This fundamentally solves the engineering problem of not being able to simultaneously achieve small sample size, multiple operating conditions, and high accuracy.

[0086] Secondly, the leakage monitoring device for water network pipelines in coal-fired power plants according to the embodiments of this application is described with reference to the accompanying drawings.

[0087] Figure 2 This is a block diagram of a leakage monitoring device for water network pipelines in a coal-fired power plant, according to an embodiment of this application.

[0088] like Figure 2 As shown, the leakage monitoring device 10 for the water network pipeline of the coal-fired power plant includes: a sample set construction module 100, a migration module 200, and a leakage monitoring module 300.

[0089] The sample set construction module 100 is used to collect transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipeline. Based on the transient calibration data and the benchmark source model in the pre-built heterogeneous network architecture, it determines the corresponding benchmark diagnostic rule sample set. Through the benchmark source model, it generates the feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set offline.

[0090] The migration module 200 is used to input pipeline steady-state operation data into the receiving network of the heterogeneous network architecture to output leakage diagnosis results. Based on feature representation, diagnosis conclusion and leakage diagnosis results, it minimizes the difference in diagnosis conclusion and intermediate layer feature difference between the receiving network and the benchmark source model on the benchmark diagnosis rule sample set, so that the receiving network meets the preset capability migration requirements.

[0091] The leakage monitoring module 300 is used to train the receiving network based on a pre-built composite loss function and a sample set of benchmark diagnostic rules, and to collect the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant. The current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results, including the leakage probability index, the location coordinates of the leakage point, and the leakage rate trend.

[0092] Optionally, in one embodiment of this application, the leakage monitoring device 10 for water network pipelines in coal-fired power plants further includes a visualization module and a push module.

[0093] The visualization module is used to perform preset visualization processing on the monitoring results after the current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results, so as to obtain the corresponding visualization information.

[0094] The push module is used to generate corresponding multi-level early warning signals based on the monitoring results, and based on the multi-level early warning signals, push the visualized information and monitoring results to the preset water balance control system through the preset standardized API interface for terminal display.

[0095] Optionally, in one embodiment of this application, the sample set construction module 100 includes: a first acquisition unit, a steady-state monitoring unit, and a transient calibration unit.

[0096] The first acquisition unit is used to deploy pressure sensors that meet the preset model requirements at multiple key nodes in the target coal-fired power plant, acquire the unit operation information corresponding to the water network pipeline of the target coal-fired power plant, and determine the system acquisition mode corresponding to the water network pipeline of the target coal-fired power plant based on the unit operation information.

[0097] The steady-state monitoring unit is used to collect steady-state operation data of the target coal-fired power plant's water network pipeline at a first preset frequency when the system acquisition mode is steady-state monitoring mode. The steady-state operation data of the pipeline includes pressure, flow rate, temperature, pump and valve status, and unit load.

[0098] The transient calibration unit is used to actively trigger the water pump step operation when the system acquisition mode is transient calibration mode, so as to capture the pressure wave propagation time difference and the deviation of the actual / theoretical hydraulic filling time in the hydraulic transient process. Combined with on-site manual calibration, it generates a leak point location-leakage amount label sample that meets the preset accuracy requirements, so as to determine the benchmark diagnostic rule sample set based on the leak point location-leakage amount label sample.

[0099] Optionally, in one embodiment of this application, the sample set construction module 100 further includes: a first generation unit, a segmentation unit, a calculation unit, a solution unit, and a second generation unit.

[0100] The first generation unit is used to perform time-frequency domain transformation on the transient waveforms of leakage point pressure / flow in the benchmark diagnostic rule sample set to generate the corresponding leakage response feature map.

[0101] The segmentation unit is used to acquire historical steady-state data corresponding to the water network pipeline of the target coal-fired power plant, and to perform sliding segmentation on the historical steady-state data based on a preset time window to obtain the corresponding segmentation results.

[0102] The calculation unit is used to calculate the corresponding theoretical hydraulic filling time based on the segmentation results and preset pipeline inherent parameters, and to determine the corresponding hydraulic filling time deviation sequence based on the residual between the theoretical hydraulic filling time and the estimated value of the actual hydraulic filling time.

[0103] The solution unit is used to obtain the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data of the target coal-fired power plant's water network pipeline. Based on the pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data, it solves the equivalent hydraulic impedance matrix of the target coal-fired power plant's water network pipeline, so as to determine the corresponding equivalent hydraulic impedance mode according to the equivalent hydraulic impedance matrix.

[0104] The second generation unit is used to obtain the feature representation corresponding to the benchmark diagnostic rule sample set based on the leakage response feature spectrum, hydraulic filling time deviation sequence and equivalent hydraulic impedance mode, and generate diagnostic conclusions based on the feature representation.

[0105] Optionally, in one embodiment of this application, the migration module 200 includes: a construction unit, a first determination unit, an extraction unit, and a minimization unit.

[0106] The construction unit is used to build a baseline source model based on a pre-built hydraulic transient analysis physical diagnostic model, and to build a corresponding temporal feature extraction module based on an LSTM network with residual connections and multi-head self-attention mechanism. It also builds a corresponding physical embedding module and a diagnostic feature alignment layer based on a pre-built PINN physical information neural network and a feature alignment adapter.

[0107] The first determining unit is used to construct a target PINN-LSTM fusion temporal network based on the physical embedding module, the diagnostic feature alignment layer, and the temporal feature extraction module, so as to determine the successor network based on the PINN-LSTM fusion temporal network.

[0108] The extraction unit is used to input the pipeline steady-state operation data into the time series feature extraction module of the receiving network to extract the time series features corresponding to the pipeline steady-state operation data, and input the time series features into the physical embedding module to generate the corresponding physical constraints.

[0109] The minimization unit is used to input physical constraints into the diagnostic feature alignment layer to minimize the differences in diagnostic conclusions between the output of the receiving network and the output of the benchmark source model on transient calibration data, as well as the differences in intermediate layer feature maps. This allows the benchmark diagnostic rules corresponding to the benchmark diagnostic rule sample set to be structurally transferred from the benchmark source model to the receiving network.

[0110] Optionally, in one embodiment of this application, the leakage monitoring module 300 includes: a second acquisition unit, a training unit, a collection unit, and a second determination unit.

[0111] The second acquisition unit is used to acquire the operating condition information of the target coal-fired power plant, determine multiple weight coefficients based on the operating condition information, and perform weighted fusion of the preset supervision loss function, feature alignment loss function and physical residual loss function based on the multiple weight coefficients to construct a composite loss function.

[0112] The training unit is used to train the receiving network based on transient calibration data, a benchmark diagnostic rule sample set, and a composite loss function, combined with a preset two-stage progressive transfer strategy.

[0113] The acquisition unit is used to acquire the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant, and input the current steady-state data stream into the trained receiving network to generate the leakage probability index and the location coordinates of the leakage point.

[0114] The second determining unit is used to output the leakage development range and multi-level early warning information for the target time period based on the correlation between the hydraulic filling time deviation sequence and the unit load, and in combination with the preset time series prediction algorithm, and to determine the leakage development trend based on the leakage development range and multi-level early warning information.

[0115] It should be noted that the foregoing explanation of the embodiment of the method for monitoring leakage in water network pipelines of coal-fired power plants also applies to the leakage monitoring device for water network pipelines of coal-fired power plants in this embodiment, and will not be repeated here.

[0116] The leakage monitoring device for water network pipelines in coal-fired power plants proposed in this application includes a sample set construction module 100, used to collect transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipelines, and determine the corresponding benchmark diagnostic rule sample set based on the transient calibration data and a pre-constructed benchmark source model in a heterogeneous network architecture. Furthermore, it generates feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set offline through the benchmark source model. A transfer module 200 is used to input the steady-state operation data of the pipelines into the receiving network of the heterogeneous network architecture to output leakage diagnosis results, and based on the feature representations... The application uses a hydraulic transient analysis model as the diagnostic capability benchmark and a PINN-LSTM temporal network as the capability transfer carrier. Through the collaborative design of feature alignment and physical constraints, the application enables the network to achieve leakage location accuracy and leakage estimation capability comparable to the benchmark model under transient excitation, relying only on steady-state operating data. This fundamentally solves the engineering problem of simultaneously achieving high accuracy with small sample sizes, multiple operating conditions, and high precision. The leakage monitoring module 300 is used to train the network based on a pre-constructed composite loss function and benchmark diagnostic rule sample set, and to collect the current steady-state data stream corresponding to the target coal-fired power plant's water network pipeline. The current steady-state data stream is input into the trained network to output corresponding monitoring results, including leakage probability index, leakage point location coordinates, and leakage trend.

[0117] Figure 3 A schematic diagram of the structure of an electronic device provided in an embodiment of this application. The electronic device may include: The memory 301, the processor 302, and the computer program stored on the memory 301 and capable of running on the processor 302.

[0118] When the processor 302 executes the program, it implements the method for monitoring leakage in the water network pipelines of coal-fired power plants provided in the above embodiments.

[0119] Furthermore, electronic devices also include: Communication interface 303 is used for communication between memory 301 and processor 302.

[0120] The memory 301 is used to store computer programs that can run on the processor 302.

[0121] The memory 301 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk storage device.

[0122] If the memory 301, processor 302, and communication interface 303 are implemented independently, then the communication interface 303, memory 301, and processor 302 can be interconnected via a bus to complete communication between them. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 3 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0123] Optionally, in a specific implementation, if the memory 301, processor 302, and communication interface 303 are integrated on a single chip, then the memory 301, processor 302, and communication interface 303 can communicate with each other through an internal interface.

[0124] Processor 302 may be a central processing unit (CPU), an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of this application.

[0125] This application also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the above-described method for monitoring leaks in water pipelines of coal-fired power plants.

[0126] This application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-described method for monitoring leaks in water pipelines of coal-fired power plants.

[0127] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0128] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0129] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0130] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0131] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0132] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0133] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0134] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of this application.

Claims

1. A method for monitoring leaks in water pipelines of a coal-fired power plant, characterized in that, Includes the following steps: The transient calibration data and steady-state operation data of the water network pipeline of the target coal-fired power plant are collected. Based on the transient calibration data and the benchmark source model in the pre-built heterogeneous network architecture, the corresponding benchmark diagnostic rule sample set is determined. The feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set are generated offline through the benchmark source model. The pipeline steady-state operation data is input into the receiving network of the heterogeneous network architecture to output leakage diagnosis results. Based on the feature representation, the diagnosis conclusion and the leakage diagnosis results, the difference between the diagnosis conclusion of the receiving network and the intermediate layer feature difference between the reference source model on the reference diagnosis rule sample set is minimized, so that the receiving network meets the preset capability transfer requirements. Based on the pre-constructed composite loss function and the benchmark diagnostic rule sample set, the receiving network is trained, and the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected. The current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results, wherein the monitoring results include the leakage probability index, the location coordinates of the leakage point, and the leakage development trend.

2. The method for monitoring leakage in water pipelines of coal-fired power plants according to claim 1, characterized in that, After inputting the current steady-state data stream into the trained receiving network to output the corresponding monitoring results, the method further includes: The monitoring results are subjected to preset visualization processing to obtain corresponding visualized information; Based on the monitoring results, corresponding multi-level early warning signals are generated, and based on the multi-level early warning signals, the visualized information and the monitoring results are pushed to the preset water balance control system through a preset standardized API interface for terminal display.

3. The method for monitoring leakage in water pipelines of coal-fired power plants according to claim 1, characterized in that, The process involves collecting transient calibration data and steady-state operation data corresponding to the water network pipelines of the target coal-fired power plant. Based on the transient calibration data and a pre-constructed benchmark source model in a heterogeneous network architecture, a corresponding benchmark diagnostic rule sample set is determined. Furthermore, through the benchmark source model, feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set are generated offline, including: Pressure sensors that meet the preset model requirements are deployed at multiple key nodes in the target coal-fired power plant, and the unit operation information corresponding to the water network pipeline of the target coal-fired power plant is obtained. The system acquisition mode corresponding to the water network pipeline of the target coal-fired power plant is determined based on the unit operation information. When the system acquisition mode is steady-state monitoring mode, the steady-state operation data of the pipeline corresponding to the water network pipeline of the target coal-fired power plant is acquired through a first preset frequency. The steady-state operation data of the pipeline includes pressure, flow rate, temperature, pump and valve status and unit load. When the system acquisition mode is transient calibration mode, it actively triggers the water pump step operation to capture the pressure wave propagation time difference and the deviation between the actual / theoretical hydraulic filling time in the hydraulic transient process. Combined with on-site manual calibration, it generates a leak point location-leakage amount label sample that meets the preset accuracy requirements, so as to determine the benchmark diagnostic rule sample set based on the leak point location-leakage amount label sample.

4. The method for monitoring leakage in water pipelines of coal-fired power plants according to claim 1, characterized in that, The step of generating feature representations and diagnostic conclusions corresponding to the benchmark diagnostic rule sample set offline using the benchmark source model includes: The transient waveforms of pressure / flow at the leak points in the benchmark diagnostic rule sample set are transformed in the time and frequency domain to generate the corresponding leak response feature map. Historical steady-state data corresponding to the water network pipeline of the target coal-fired power plant is obtained, and the historical steady-state data is subjected to sliding segmentation based on a preset time window to obtain the corresponding segmentation result; The theoretical hydraulic filling time is calculated based on the segmentation results and the preset inherent parameters of the pipeline. Based on the residual between the theoretical hydraulic filling time and the estimated value of the actual hydraulic filling time, the corresponding hydraulic filling time deviation sequence is determined. The pressure wave velocity, pipe segment geometric characteristics, and transient pressure fluctuation data corresponding to the water network pipeline of the target coal-fired power plant are obtained. Based on the pressure wave velocity, the pipe segment geometric characteristics, and the transient pressure fluctuation data, the equivalent hydraulic impedance matrix of the water network pipeline of the target coal-fired power plant is solved, so as to determine the corresponding equivalent hydraulic impedance mode according to the equivalent hydraulic impedance matrix. Based on the leakage response feature map, the hydraulic filling time deviation sequence, and the equivalent hydraulic impedance mode, the feature representation corresponding to the benchmark diagnostic rule sample set is obtained, and the diagnostic conclusion is generated according to the feature representation.

5. The method for monitoring leakage in water pipelines of coal-fired power plants according to claim 4, characterized in that, The process involves inputting the pipeline steady-state operation data into the receiving network of the heterogeneous network architecture to output leakage diagnosis results. Based on the feature representation, the diagnosis conclusion, and the leakage diagnosis results, the process minimizes the difference in diagnosis conclusions and intermediate layer feature differences between the receiving network and the benchmark source model on the benchmark diagnosis rule sample set, ensuring that the receiving network meets preset capability transfer requirements. This includes: Based on the pre-constructed hydraulic transient analysis physical diagnostic model, the baseline source model is constructed, and based on the LSTM network with residual connections and multi-head self-attention mechanism, a corresponding temporal feature extraction module is constructed. Based on the pre-constructed PINN physical information neural network and feature alignment adapter, a corresponding physical embedding module and diagnostic feature alignment layer are constructed respectively. A target PINN-LSTM fusion temporal network is constructed based on the physical embedding module, the diagnostic feature alignment layer, and the temporal feature extraction module, so as to determine the successor network based on the PINN-LSTM fusion temporal network; The pipeline steady-state operation data is input into the time-series feature extraction module of the receiving network to extract the time-series features corresponding to the pipeline steady-state operation data, and the time-series features are input into the physical embedding module to generate the corresponding physical constraints; The physical constraints are input into the diagnostic feature alignment layer to minimize the difference between the output diagnostic conclusions of the receiving network and the output diagnostic conclusions of the benchmark source model on the transient calibration data, as well as the difference in the intermediate layer feature maps, so that the benchmark diagnostic rules corresponding to the benchmark diagnostic rule sample set are structurally transferred from the benchmark source model to the receiving network.

6. The method for monitoring leakage in water pipelines of coal-fired power plants according to claim 5, characterized in that, The receiving network is trained based on a pre-constructed composite loss function and the benchmark diagnostic rule sample set. Current steady-state data streams corresponding to the target coal-fired power plant's water network pipelines are collected, and these data streams are input into the trained receiving network to output corresponding monitoring results. These monitoring results include a leakage probability index, leakage point location coordinates, and leakage rate trend, including: Obtain the operating condition information corresponding to the target coal-fired power plant, determine multiple weight coefficients based on the operating condition information, and perform weighted fusion of the preset supervision loss function, feature alignment loss function and physical residual loss function based on the multiple weight coefficients to construct the composite loss function; The receiving network is trained based on the transient calibration data, the benchmark diagnostic rule sample set, and the composite loss function, combined with a preset two-stage progressive transfer strategy. The current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant is collected, and the current steady-state data stream is input into the trained receiving network to generate the leakage probability index and the location coordinates of the leakage point; Based on the correlation between the hydraulic filling time deviation sequence and the unit load, and combined with the preset time series prediction algorithm, the leakage development range and multi-level early warning information for the target time period are output, and the leakage development trend is determined according to the leakage development range and the multi-level early warning information.

7. A leakage monitoring device for water network pipelines in a coal-fired power plant, characterized in that, include: The sample set construction module is used to collect transient calibration data and steady-state operation data of the target coal-fired power plant's water network pipeline, and based on the transient calibration data and the benchmark source model in the pre-built heterogeneous network architecture, determine the corresponding benchmark diagnostic rule sample set, and generate the feature representation and diagnostic conclusion corresponding to the benchmark diagnostic rule sample set offline through the benchmark source model. The migration module is used to input the pipeline steady-state operation data into the receiving network of the heterogeneous network architecture to output leakage diagnosis results. Based on the feature representation, the diagnosis conclusion and the leakage diagnosis results, it minimizes the difference between the diagnosis conclusion of the receiving network and the intermediate layer feature difference between the reference source model on the reference diagnosis rule sample set, so that the receiving network meets the preset capability migration requirements. The leakage monitoring module is used to train the receiving network based on a pre-constructed composite loss function and the benchmark diagnostic rule sample set, and to collect the current steady-state data stream corresponding to the water network pipeline of the target coal-fired power plant. The current steady-state data stream is input into the trained receiving network to output the corresponding monitoring results, wherein the monitoring results include the leakage probability index, the location coordinates of the leakage point, and the leakage amount development trend.

8. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, the processor executing the program to implement the method for monitoring leaks in water pipelines of a coal-fired power plant as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the method for monitoring leaks in water pipelines of coal-fired power plants as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, The computer program is executed to implement the method for monitoring leaks in water pipelines of coal-fired power plants as described in any one of claims 1-6.

Images