Tap water leakage monitoring system based on low-power-consumption terminal

Abstract

The invention discloses a tap water leakage monitoring system based on a low-power-consumption terminal, and the system comprises an original monitoring data collection module which is used for collecting the data of terminal equipment, and constructing an original monitoring data set; the low-dimensional acoustic representation vector extraction module is used for executing feature coding to generate a low-dimensional acoustic representation vector; the region context priori vector construction module is used for generating a region context priori vector; the water leakage state judgment module is used for generating a water leakage state label and a judgment confidence value by using a TabPFN model; the confidence verification and far-end cooperation module is used for executing far-end cooperation processing and generating a water leakage state judgment result; the data reporting and scheduling module is used for executing data reporting and maintaining an original sampling period; and the far-end monitoring and positioning module is used for writing terminal information into a database and performing suspected water leakage point position identification processing. The accuracy and the fault-tolerant capability of water leakage monitoring of the urban water supply network are improved.

Description

Technical Field

[0001] This invention relates to the field of urban water supply network monitoring, and in particular to a tap water leakage monitoring system based on a low-power terminal. Background Technology

[0002] In current urban water supply systems, leaks not only lead to significant water waste but also severely impact supply efficiency and operating costs. To address this, the industry commonly employs acoustic sensors to detect leaks in water supply networks, combined with communication modules for remote data transmission, to achieve leak monitoring. However, traditional methods generally suffer from high terminal power consumption, low data acquisition frequency, high false alarm rates, and poor leak location accuracy, making them unsuitable for large-scale deployment and long-term operation.

[0003] In existing technologies, terminals often rely on fixed thresholds to trigger data uploads, lacking the ability to dynamically assess environmental conditions and communication status. This results in rigid data reporting strategies, increasing energy consumption from ineffective communication and potentially missing important event information. Furthermore, many detection systems fail to combine regional water pressure characteristics and historical leakage distribution for contextual analysis, leading to limited accuracy in leak detection, especially in interference scenarios where false alarms are frequent. While some systems have introduced remote assisted judgment mechanisms, the overall process lacks a structured collaborative mechanism, making it difficult to effectively link the judgment results with subsequent location tasks.

[0004] Therefore, how to provide a tap water leakage monitoring system based on a low-power terminal is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a tap water leakage monitoring system based on a low-power terminal. It integrates acoustic feature compression extraction, context prior modeling, and a multi-state classification reasoning method based on the TabPFN model. It improves the accuracy of small sample learning through a dynamic path activation mechanism and introduces a remote collaborative judgment and auxiliary inspection mechanism to enhance the ability to eliminate false alarms in low-confidence judgment scenarios. Finally, it combines an event-level data reporting strategy and a GIS positioning mechanism to achieve high-precision, low-power leakage identification and location. It has the advantages of low deployment energy consumption, low false judgment rate, and strong adaptability.

[0006] A tap water leakage monitoring system based on a low-power terminal according to an embodiment of the present invention includes: The raw monitoring data acquisition module is used to collect sampling data from acoustic sensors and communication status, sampling time, and environmental parameters of terminal devices to construct a raw monitoring data set. The low-dimensional acoustic representation vector extraction module is used to perform structured feature encoding processing on the original monitoring data set and extract low-dimensional acoustic representation vectors that match the input structure of the TabPFN model. The regional context prior vector construction module is used to extract regional labels, water pressure parameters and historical leakage labels based on the terminal installation location, and generate regional context prior vectors and feature offset values. The leakage status judgment module is used to input the low-dimensional acoustic representation vector, the region context prior vector and the feature offset value into the TabPFN model to generate leakage status labels and judgment confidence values. The confidence verification and remote collaboration module is used to trigger the remote collaboration process when the confidence value is lower than the threshold. The remote platform loads the backup model to generate the final leakage status judgment result and outputs the auxiliary inspection identifier based on the auxiliary inspection module. The data reporting and scheduling module is used to perform event-level data reporting and maintain the original sampling cycle based on the leakage status judgment result, communication status and auxiliary inspection identifier; The remote monitoring and positioning module is used to write terminal leakage information into the remote monitoring database and identify the location of suspected leakage points by combining the GIS interface and structured event records.

[0007] Optionally, modules can be integrated using the following methods: Collect sampling data from acoustic sensors deployed at nodes of the water supply network, and simultaneously acquire the communication status, sampling time, and environmental parameters of the terminal devices to construct the original monitoring data set; The original monitoring dataset is subjected to structured feature encoding to extract low-dimensional acoustic representation vectors; Based on the terminal installation location corresponding to the low-dimensional acoustic representation vector, a regional context prior vector is constructed and input into the TabPFN model to dynamically activate the pre-trained task path embedded within it, generate prior path identifiers, perform multi-state classification inference, and generate leakage state labels and judgment confidence values. Based on the judgment confidence value, an abnormal confidence value verification process is performed. If it is lower than the set threshold, the remote collaborative processing process is triggered. The low-dimensional acoustic representation vector, regional context prior vector and prior path identifier are uploaded to the remote platform, the backup model is loaded for re-judgment processing, and the final leakage status judgment result and auxiliary inspection identifier are generated. Based on the water leakage status judgment result, communication status and auxiliary inspection identifier, the event-level data reporting scheduling strategy is executed. If the judgment result is a leakage status and the communication status meets the minimum transmission power condition, the control terminal performs an acoustic data back transmission operation and completes data reporting with the current status label and auxiliary inspection identifier. If the judgment result is a non-leakage status and is identified as a false alarm, the terminal maintains the original sampling period configuration and records the local status log. The terminal leakage information is written into the remote monitoring database to store event information and perform suspected leak location identification processing.

[0008] Optionally, the sampling data of the acoustic sensor includes vibration waveform data, spectral envelope characteristics and energy distribution data within a unit time window; the communication status includes signal strength, uplink channel status and remaining transmission power; and the environmental parameters include background noise level and external vibration indicators.

[0009] Optionally, the extraction of the low-dimensional acoustic representation vector includes: Receive the raw monitoring data set, perform frame-by-frame processing based on a set time window, and organize it into a multi-channel feature matrix; Based on the multi-channel feature matrix, the initial acoustic feature structure is constructed by arranging the sound pressure amplitude, frequency response, and energy density in that order. Based on the initial acoustic feature structure, a spectrum channel selection operation is performed. By setting an energy threshold and a fixed channel interval parameter, frequency sub-intervals containing abnormal vibration features are selected to construct a spectrum screening feature set. Sparse feature transformation is performed on the spectral screening feature set to compress intra-frame repetitive amplitudes and periodic channel responses, generating a sparse mapping structure representation; The sparse mapping structure representation is input into the dimension adjustment module, which performs vectorization mapping and dimension recombination based on the required dimension structure input from the TabPFN model to generate a low-dimensional acoustic representation vector.

[0010] Optionally, the generation of the leakage status label and the determination confidence value includes: Obtain the terminal installation location bound to the low-dimensional acoustic representation vector, retrieve the geographical unit to which the location belongs, and extract the corresponding area label, water pressure parameter and historical leakage label; The location index encoding process is performed on the area label to obtain the area location encoding vector. The water pressure parameter is numerically normalized to obtain the standardized water pressure vector. The event frequency statistics process within the period is performed on the historical leakage label to obtain the historical event statistics vector. The area context prior vector is generated by combining the three sub-vectors, and the feature offset value is calculated based on the starting index position of the three sub-vectors. The region context prior vector is input into the TabPFN model. Based on the feature offset value, the unique corresponding pre-trained task path is selected and activated from the set of pre-trained task paths embedded in the TabPFN model, and a prior path identifier is generated. The TabPFN model performs forward inference processing, and outputs the corresponding leakage status label based on the input structure consisting of low-dimensional acoustic representation vector, regional context prior vector and prior path identifier, while generating a decision confidence value.

[0011] Optionally, the generation of the leakage status judgment result and auxiliary inspection identifier includes: Receive the judgment confidence value and compare it with the preset judgment threshold. If it is less than the judgment threshold, trigger the remote collaborative processing flow. The low-dimensional acoustic representation vector, regional context prior vector, and prior path identifier corresponding to the current terminal are encapsulated into a remote processing request data structure and uploaded to the remote platform. After receiving the remote processing request data structure, the remote platform loads the corresponding backup model, executes the re-judgment process, and generates the re-judgment result as the final leakage status judgment result. After completing the re-judgment process, the remote platform starts the auxiliary inspection module to perform false alarm elimination operation on the leakage status judgment result and generate the false alarm elimination judgment basis; Based on the criteria for eliminating false alarms, an auxiliary inspection identifier is generated through a judgment logic algorithm that integrates multi-source heterogeneous information.

[0012] Optionally, the remote platform performs a re-determination process including: The backup model is invoked. Specifically, the backup model is a deep classification model built on a multi-class Transformer structure. It is deployed on a remote platform using a centralized training method and has end-to-end classification capability for multi-input field structures. Perform a re-judgment process, which includes: Receive the remote processing request data structure and combine the three data items into a further judgment input vector; The internal input adaptation module of the backup model is called to convert the re-judgment input vector into an embedded representation structure that conforms to the model input structure, and then input it into the backbone network of the backup model to perform the forward inference process. During forward inference, the backup model performs calculations based on the input vector structure and model training weights, and outputs the corresponding multi-class probability distribution. In a multi-class probability distribution, extract the leakage status label corresponding to the maximum probability value, and extract the maximum probability value as the judgment confidence value. Based on the leakage status label and the judgment confidence value, and combined with the corresponding input vector field index mapping relationship and the current model path identification information, a re-judgment result is constructed.

[0013] Optionally, the event-level data reporting scheduling strategy includes: The system receives the final water leakage status judgment result, terminal communication status and auxiliary inspection identifier. Based on the preset scheduling rule set, it determines whether the data reporting trigger condition is met. The data reporting trigger condition requires that the water leakage status judgment result is a leakage status, the remaining transmission power of the communication status is not lower than the minimum transmission power threshold, and the auxiliary inspection identifier indicates that there are clear abnormal characteristics in the current status. If the data reporting trigger condition is met, the control terminal starts the reporting execution module, encapsulates the low-dimensional acoustic representation vector, leakage status label and auxiliary inspection identifier within the current sampling period, and generates a data reporting structure for real-time reporting; If the data reporting triggering conditions are not met, the terminal will maintain the original sampling cycle configuration, periodically execute the sampling, encoding and local judgment process, and record the leakage status judgment result, auxiliary inspection mark and communication status of the current cycle. If the terminal detects the same leakage status judgment result in multiple consecutive cycles and the auxiliary inspection indicator continues to have an abnormal mark, it will automatically re-evaluate the reporting trigger condition. If the evaluation result is that the condition is met, the reporting process will be restarted.

[0014] Optionally, the remote monitoring includes: Construct a structured event information data table, setting fields including terminal number, event timestamp, leakage status judgment result, judgment confidence value, auxiliary inspection identifier and prior path identifier. After receiving data reported by the terminal, write it into the remote monitoring database. The GIS interface data display module is invoked to obtain the terminal's geographical location information based on the terminal number and event timestamp fields and mark it in the geographic information layer. The historical judgment trajectory of the terminal is drawn, and the leakage status judgment result of the current period is dynamically superimposed to generate a terminal status time sequence display diagram. Based on the geographical location associated with the terminal number, the spatial coordinate information of that location in the GIS system is extracted and combined with the synchronous event records of other terminals in the same area in the structured event information data table to form a multi-point status spatial input set. The multi-point state space input set is input into the positioning mechanism module to perform suspected leak point location identification processing and generate the coordinates of the suspected leak point location.

[0015] The beneficial effects of this invention are: First, by designing a low-dimensional acoustic representation vector extraction mechanism, this invention significantly reduces the requirements for terminal-side computing resources and energy consumption, adapts to deployment environments of various types of low-power sensor terminals, and achieves efficient feature compression encoding while ensuring the integrity of water leakage feature expression, thereby improving the practicality and stability of the system in large-scale pipeline network deployment scenarios.

[0016] Secondly, a regional context prior vector is constructed by combining the terminal installation location, and a path activation classification reasoning method based on the TabPFN model is introduced. Under the condition of limited training samples, the discrimination and generalization ability of water leakage status identification is effectively enhanced, and the stable reasoning performance of the model in multi-class complex interference environments is improved.

[0017] Furthermore, this invention proposes a confidence-driven remote collaborative processing mechanism and an auxiliary inspection module linkage strategy. When the confidence level of the terminal-side model is insufficient, the remote re-judgment and false alarm investigation process is automatically triggered, which effectively reduces false alarms and missed alarms caused by environmental interference and improves the overall intelligence, accuracy and fault tolerance of the water leakage monitoring system.

[0018] Finally, by combining event-level data reporting and scheduling strategies with GIS positioning mechanisms, precise data reporting control and spatial labeling of leakage locations were achieved, providing the operation and maintenance department with more accurate, real-time, and visual decision-making support, and significantly improving the leakage control capabilities of the urban water supply system. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0020] Figure 1 This is an overall flowchart of a water leakage monitoring system based on a low-power terminal proposed in this invention. Figure 2 This is a schematic diagram of the collaborative processing structure for low-dimensional acoustic representation vector extraction and region context prior vector construction in this invention; Figure 3 This is a flowchart illustrating the linkage between the remote collaborative processing mechanism triggered by the judgment confidence value and the auxiliary inspection module in this invention. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0022] refer to Figure 1-3 A tap water leakage monitoring system based on a low-power terminal, comprising: The raw monitoring data acquisition module is used to collect sampling data from acoustic sensors deployed at water supply network nodes, and simultaneously acquire the communication status, sampling time and environmental parameters of terminal devices to construct a raw monitoring data set. The sampling data from the acoustic sensors includes vibration waveform data, spectral envelope characteristics and energy distribution data within a unit time window. The communication status includes signal strength, uplink channel status and remaining transmission power. The environmental parameters include background noise level and external vibration indicators. The low-dimensional acoustic representation vector extraction module is used to perform structured feature encoding processing on the original monitoring data set. It extracts low-dimensional acoustic representation vectors that match the input structure of the TabPFN model through operations such as acoustic spectrum channel selection, sparse feature transformation and dimension adjustment. The regional context prior vector construction module is used to extract the regional label, water pressure parameter and historical leakage label associated with the terminal installation location corresponding to the low-dimensional acoustic representation vector. The module performs location index encoding on the regional label, numerical normalization on the water pressure parameter, and event frequency statistics on the historical leakage label within the period. The three types of processing results are combined in a preset splicing order to generate the regional context prior vector and the feature offset value is calculated. The leakage status judgment module is used to input the low-dimensional acoustic representation vector, the region context prior vector and the feature offset value into the TabPFN model, activate the unique pre-trained task path inside the model and generate the prior path identifier, perform multi-state classification inference, and generate leakage status label and judgment confidence value. The confidence verification and remote collaboration module is used to trigger the remote collaboration process when the confidence value is lower than a preset threshold. It encapsulates the low-dimensional acoustic representation vector, the regional context prior vector, and the prior path identifier into a remote processing request data structure and uploads it to the remote platform. The remote platform loads the backup model and performs re-judgment processing to generate the final leakage status judgment result. It also calls the auxiliary inspection module to perform false alarm elimination operation based on the terminal's multi-time period sampling playback data, terminal displacement records, and feedback information from other terminals in the area, and generates an auxiliary inspection identifier. The data reporting and scheduling module is used to execute the event-level data reporting and scheduling strategy based on the water leakage status judgment result, communication status and auxiliary inspection identifier. When the preset conditions are met, the data reporting process is triggered, and acoustic data and status information are encapsulated and uploaded to the remote monitoring database. When the conditions are not met, the terminal maintains the original sampling cycle configuration and records the local status log. The remote monitoring and positioning module is used to write terminal leakage information into a structured event information data table, and combine it with the GIS interface to mark the terminal's geographical location, historical judgment trajectory and current status; generate positioning mechanism input parameters based on the terminal's geographical location, construct a multi-point status space input set in combination with structured event records, perform suspected leakage point location identification processing, and output the coordinates of the suspected leakage point location.

[0023] In this embodiment, the modules are interconnected using the following method: The sampling data of acoustic wave sensors deployed at the nodes of the water supply network are collected, and the communication status, sampling time and environmental parameters of the terminal equipment are acquired simultaneously to construct the original monitoring data set. The sampling data of the acoustic wave sensors include vibration waveform data, spectral envelope characteristics and energy distribution data within a unit time window. The communication status includes signal strength, uplink channel status and remaining transmission power. The environmental parameters include background noise level and external vibration indicators. The original monitoring dataset is subjected to structured feature encoding processing. Low-dimensional acoustic representation vectors are extracted through spectral channel selection, sparse feature transformation and dimension adjustment operations. The low-dimensional acoustic representation vectors are matched with the input structure of the TabPFN model. Based on the terminal installation location corresponding to the low-dimensional acoustic representation vector, the region label, water pressure parameter and historical leakage label associated with the location are extracted to construct the region context prior vector. The region context prior vector is input into the TabPFN model to dynamically activate the pre-trained task path embedded in it and generate the prior path identifier. Multi-state classification reasoning is performed to generate leakage state label and judgment confidence value. Based on the judgment confidence value, an abnormal confidence value verification process is performed. If the judgment confidence value is lower than the set threshold, the remote collaborative processing process is triggered. The low-dimensional acoustic representation vector, the regional context prior vector, and the prior path identifier are uploaded to the remote platform. The remote platform loads the backup model for re-judgment processing, generates the final water leakage status judgment result, and calls the auxiliary inspection module to perform false alarm elimination operation. The auxiliary inspection module includes terminal multi-time period sampling playback data, terminal displacement records, and feedback information from other terminals in the area, and outputs auxiliary inspection identifier. Based on the water leakage status judgment result, communication status and auxiliary inspection identifier, the event-level data reporting scheduling strategy is executed. If the judgment result is a leakage status and the communication status meets the minimum transmission power condition, the control terminal performs an acoustic data back transmission operation and completes data reporting with the current status label and auxiliary inspection identifier. If the judgment result is a non-leakage status and is identified as a false alarm, the terminal maintains the original sampling period configuration and records the local status log. The terminal leakage information is written into the remote monitoring database. The terminal leakage information includes the leakage status judgment result, judgment confidence value, auxiliary inspection identifier and prior path identifier. The database stores the event information reported by each terminal in a structured manner, and combines the terminal's geographical location, historical judgment trajectory and current status with the GIS interface. The database also generates positioning mechanism input parameters based on the terminal's geographical location and performs suspected leakage point location identification processing.

[0024] In this embodiment, the sampling data of the acoustic sensor includes vibration waveform data, spectral envelope characteristics and energy distribution data within a unit time window; the communication status includes signal strength, uplink channel status and remaining transmission power; and the environmental parameters include background noise level and external vibration indicators.

[0025] In this embodiment, the extraction of the low-dimensional acoustic representation vector includes: It receives vibration waveform data, spectral envelope characteristics and energy distribution data from the raw monitoring data set, and performs frame processing based on a set time window, organizing each frame of signal data into a multi-channel feature matrix. Based on the multi-channel feature matrix, the initial acoustic feature structure is constructed by arranging the sound pressure amplitude, frequency response, and energy density in that order. Based on the energy distribution of frequency response components in the initial acoustic feature structure, a spectrum channel selection operation is performed. The spectrum channel selection operation filters out frequency sub-intervals containing abnormal vibration features by setting an energy threshold and a fixed channel interval parameter, retains the acoustic channel features of the corresponding intervals, removes background noise frequency band information, and constructs a spectrum screening feature set. A sparse feature transformation operation is performed on the acoustic spectrum screening feature set. The sparse feature transformation operation adopts a position merging and difference compression strategy to compress intra-frame repetitive amplitudes and periodic channel responses, generating a sparse mapping structure representation to remove low-rate-of-change channel components. The sparse mapping structure representation is input into the dimension adjustment module, which performs vectorization mapping and dimension reorganization according to the required dimension structure of the TabPFN model input, generating a low-dimensional acoustic representation vector that satisfies the fixed number of channels and feature length, which is used for subsequent construction of regional context prior vectors and TabPFN model inference processing.

[0026] In this embodiment, the generation of the leakage status label and the determination confidence value includes: Obtain the terminal installation location bound to the low-dimensional acoustic representation vector, retrieve the geographical unit to which the location belongs in the remote monitoring database through the terminal installation location, and extract the area label, water pressure parameters and historical leakage labels corresponding to the geographical unit to form a set of regional basic data bound to the terminal location; The location index encoding process is performed on the regional labels in the regional basic data set to obtain the regional location encoding vector. The water pressure parameter is numerically normalized to obtain the standardized water pressure vector. The historical leakage label is subjected to event frequency statistics within the period to obtain the historical event statistics vector. The regional location encoding vector, the standardized water pressure vector and the historical event statistics vector are combined in a preset splicing order to generate the regional context prior vector. The feature offset value is calculated based on the starting index position of the three sub-vectors in the combined structure to represent the positioning information of various feature fields in the input vector. The region context prior vector is input into the TabPFN model. Based on the feature offset value in the region context prior vector, a unique corresponding pre-trained task path is selected and activated from the set of pre-trained task paths embedded in the TabPFN model, and a prior path identifier is generated. The prior path identifier is used to record the path number and path structure on which the current inference process depends. The TabPFN model performs forward inference processing. Based on the input structure consisting of low-dimensional acoustic representation vectors, regional context prior vectors, and prior path identifiers, it outputs corresponding leakage status labels. The leakage status labels represent the leakage status category to which the current terminal sampled data belongs. At the same time, it generates a judgment confidence value. The judgment confidence value is a real number representing the confidence level of the inference result and is used for subsequent abnormal confidence verification processing.

[0027] In this embodiment, the generation of the leakage status judgment result and auxiliary inspection mark includes: Receive the decision confidence value generated by the TabPFN model inference process, and compare the decision confidence value with the preset decision threshold. If the decision confidence value is less than the decision threshold, trigger the remote collaborative processing flow and record the current inference task number. The low-dimensional acoustic representation vector, regional context prior vector, and prior path identifier corresponding to the current terminal are encapsulated into a remote processing request data structure, which is then uploaded to the remote platform through the communication interface to complete the remote collaborative processing data submission operation. After receiving the remote processing request data structure, the remote platform loads the corresponding backup model, calls the backup model to perform the re-judgment process, generates the re-judgment result based on the uploaded low-dimensional acoustic representation vector, regional context prior vector and prior path identifier, and uses the re-judgment result as the final leakage status judgment result. After completing the re-judgment process, the remote platform starts the auxiliary inspection module to perform false alarm elimination operation on the water leakage status judgment result. The auxiliary inspection module calls the terminal multi-time period sampling playback data of the current terminal in the most recent three consecutive sampling cycles, the displacement record data of the current terminal, and the feedback information data of other terminals located in the same area as the current terminal, and performs joint analysis on the three types of data to generate the false alarm elimination judgment basis. Based on the false alarm elimination judgment criteria output by the auxiliary inspection module, the remote platform generates an auxiliary inspection identifier through a judgment logic algorithm that fuses multi-source heterogeneous information. The auxiliary inspection identifier is used to characterize whether there is a risk of false alarm in the final water leakage status judgment result, and serves as the input condition for the subsequent data reporting scheduling strategy.

[0028] In this embodiment, the remote platform performs a re-judgment process, including: The backup model is invoked. Specifically, the backup model is a deep classification model built on a multi-class Transformer structure. Unlike the TabPFN model, the backup model is deployed on a remote platform using a centralized training method. It has end-to-end classification capability for multi-input field structures and is used to perform the re-judgment process. Perform a re-judgment process, which includes: The system receives the low-dimensional acoustic representation vector, the region context prior vector, and the prior path identifier from the remote processing request data structure, and combines the three data items into a re-judgment input vector according to the preset field concatenation order. The internal input adaptation module of the backup model is called to convert the re-judgment input vector into an embedded representation structure that conforms to the model input structure, and then input it into the backbone network of the backup model to perform the forward inference process. During forward inference, the backup model performs calculations based on the input vector structure and model training weights, and outputs the corresponding multi-class probability distribution. In a multi-class probability distribution, extract the leakage status label corresponding to the maximum probability value, and extract the maximum probability value as the judgment confidence value. Based on the leakage status label and the judgment confidence value, and combined with the corresponding input vector field index mapping relationship and the current model path identification information, a re-judgment result is constructed. The re-judgment result is used to replace the initial judgment result and serves as the input content for subsequent calls to the auxiliary inspection module.

[0029] In this embodiment, the event-level data reporting scheduling strategy includes: The system receives the final water leakage status judgment result, terminal communication status and auxiliary inspection identifier. Based on the preset scheduling rule set, it determines whether the data reporting trigger condition is met. The data reporting trigger condition requires that the water leakage status judgment result is a leakage status, the remaining transmission power of the communication status is not lower than the minimum transmission power threshold, and the auxiliary inspection identifier indicates that there are clear abnormal characteristics in the current status. If the data reporting trigger condition is met, the control terminal starts the reporting execution module, encapsulates the low-dimensional acoustic representation vector, water leakage status label and auxiliary inspection identifier within the current sampling period, generates a data reporting structure, and reports to the remote monitoring database in real time through the NB-IoT communication link, and records the reporting timestamp, terminal number and trigger type. If the data reporting triggering conditions are not met, the terminal will maintain the original sampling cycle configuration, periodically execute the sampling, encoding and local judgment process, and record the leakage status judgment result, auxiliary inspection identifier and communication status of the current cycle in the terminal's local log structure for the purpose of continuity judgment and trend comparison in subsequent cycles. If the terminal detects the same leakage status judgment result in multiple consecutive cycles and the auxiliary inspection indicator continues to have abnormal markings, it will automatically re-evaluate the reporting trigger conditions. If the evaluation result is that the conditions are met, the reporting process will be restarted to compensate for the risk of missed reporting caused by insufficient judgment in a single cycle.

[0030] In this embodiment, the remote monitoring includes: Construct a structured event information data table, setting fields including terminal number, event timestamp, leakage status judgment result, judgment confidence value, auxiliary inspection identifier and prior path identifier. After receiving data reported by the terminal, parse the above field contents and write them into the remote monitoring database. Create an indexed linked list for the event records of the same terminal in chronological order. The GIS interface data display module is called to obtain the terminal's geographical location information based on the terminal number and event timestamp fields, and the current terminal location is marked in the geographic information layer. At the same time, the historical judgment trajectory of the terminal is drawn based on the historical records in the structured event information data table, and the leakage status judgment results of the current period are dynamically superimposed in the layer to generate a terminal status time sequence display diagram. Based on the geographical location associated with the terminal number, the spatial coordinate information of that location in the GIS system is extracted, and this spatial coordinate is used as part of the input parameters of the positioning mechanism. It is then combined with the synchronous event records of other terminals in the same area in the structured event information data table to form a multi-point status spatial input set. The multi-point state space input set is input into the positioning mechanism module to perform suspected leak point location identification processing. The identification process calculates spatial aggregation index and time consistency index based on the distribution of leak status judgment results reported by each terminal in the time series, the distribution of auxiliary inspection marks and communication status record changes, and generates the location coordinates of suspected leak points, which are then used by the subsequent platform alarm mechanism and manual maintenance path planning module.

[0031] Example 1: To verify the feasibility of this invention in practice, it was applied to a long-term leakage monitoring scenario in urban water supply networks. In this scenario, the water supply network is characterized by dense pipeline distribution, complex environmental noise, and limited terminal power supply. Traditional monitoring methods relying on manual inspections or high-frequency data transmission struggle to balance identification accuracy and energy consumption, often resulting in high false alarm rates, high maintenance costs, and insufficient equipment battery life. This embodiment addresses these issues by deploying a low-power acoustic monitoring terminal.

[0032] In this scenario, low-power terminals are installed at multiple nodes in the water supply network. Each terminal continuously collects vibration waveform data, spectral envelope features, and energy distribution data from acoustic sensors, and simultaneously collects communication status and environmental parameters, forming a raw monitoring data set. The terminal performs structured feature encoding processing on the raw monitoring data set locally. Through spectral channel selection, sparse feature transformation, and dimensionality adjustment operations, it extracts a low-dimensional acoustic representation vector that matches the input structure of the TabPFN model. This processing method avoids uploading the complete high-dimensional acoustic features, reducing the computational load and data transmission scale on the terminal side.

[0033] After acoustic feature processing, the terminal extracts the corresponding area label, water pressure parameter, and historical leakage label based on its installation location. It then performs location index encoding on the area label, numerical normalization on the water pressure parameter, and periodic event frequency statistics on the historical leakage label. These three results are combined to generate an area context prior vector, and a feature offset value is calculated. The low-dimensional acoustic representation vector, the area context prior vector, and the feature offset value are input into the TabPFN model. The model completes multi-state classification inference without requiring a large number of training samples, outputting a leakage state label and a decision confidence value. This process completes the initial judgment on the terminal side, effectively reducing irrelevant data entering the communication link.

[0034] During operation, when the confidence level is within a stable range, the terminal directly executes the event-level data reporting scheduling strategy based on the leakage status judgment result. When the confidence level is lower than a preset threshold, the system automatically triggers a remote collaborative processing flow, encapsulating the low-dimensional acoustic representation vector, regional context prior vector, and prior path identifier into a remote processing request data structure and uploading it to the remote platform. The remote platform loads the backup model and performs re-judgment processing, and combines the auxiliary inspection module to comprehensively analyze multi-time period sampling playback data, terminal displacement records, and feedback information from other terminals in the area to generate an auxiliary inspection identifier to correct the initial judgment result. Finally, the confirmed leakage status judgment result, the confidence level, and the auxiliary inspection identifier are written into the remote monitoring database and spatially labeled through the GIS interface for subsequent leakage point location analysis.

[0035] During continuous operation, the system performance was statistically analyzed by comparing it with traditional methods based on fixed threshold acoustic detection and timed full reporting, and the following experimental data comparison results were obtained.

[0036] Table 1. Performance Comparison Results of Tap Water Leakage Detection Systems

[0037] As shown in Table 1, the method of this invention improves the accuracy of leak detection by approximately 6 percentage points compared to the traditional method. This is mainly due to the introduction of a regional context prior vector, which allows the model to combine regional water pressure characteristics and historical leak distribution during the judgment process, reducing the interference of environmental noise on acoustic feature determination. Regarding the false alarm rate, the method of this invention is significantly lower than the traditional method. This is thanks to the remote collaborative processing mechanism triggered by the judgment confidence value and the introduction of the auxiliary inspection module, which allows low-confidence judgment results to be re-verified and corrected.

[0038] In terms of data transmission and energy consumption, the method of this invention exhibits more significant advantages. Due to the adoption of an event-level data reporting scheduling strategy, a large amount of non-leaking state data is not frequently uploaded, reducing the average daily data upload volume per terminal to about one-third of that of traditional methods, and correspondingly reducing communication energy consumption. This result demonstrates that the present invention can effectively extend the terminal's battery life and reduce maintenance frequency in long-term operating scenarios.

[0039] Regarding the success rate of correction in low-confidence scenarios, the method of this invention significantly improves upon traditional methods, demonstrating that the collaborative processing mechanism of loading backup models on a remote platform and combining them with multi-source auxiliary inspection information can provide more stable judgment results under complex operating conditions. Simultaneously, by incorporating the location mechanism input parameters generated from the GIS interface and structured event records, the error in identifying suspected leak locations is further reduced, providing a more reliable spatial reference for subsequent maintenance decisions.

[0040] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A tap water leakage monitoring system based on a low-power terminal, characterized in that, include: The raw monitoring data acquisition module is used to collect sampling data from acoustic sensors and communication status, sampling time, and environmental parameters of terminal devices to construct a raw monitoring data set. The low-dimensional acoustic representation vector extraction module is used to perform structured feature encoding processing on the original monitoring data set and extract low-dimensional acoustic representation vectors that match the input structure of the TabPFN model. The regional context prior vector construction module is used to extract regional labels, water pressure parameters and historical leakage labels based on the terminal installation location, and generate regional context prior vectors and feature offset values. The leakage status judgment module is used to input the low-dimensional acoustic representation vector, the region context prior vector and the feature offset value into the TabPFN model to generate leakage status labels and judgment confidence values. The confidence verification and remote collaboration module is used to trigger the remote collaboration process when the confidence value is lower than the threshold. The remote platform loads the backup model to generate the final leakage status judgment result and outputs the auxiliary inspection identifier based on the auxiliary inspection module. The data reporting and scheduling module is used to perform event-level data reporting and maintain the original sampling cycle based on the leakage status judgment result, communication status and auxiliary inspection identifier; The remote monitoring and positioning module is used to write terminal leakage information into the remote monitoring database and identify the location of suspected leakage points by combining the GIS interface and structured event records.

2. The tap water leakage monitoring system based on a low-power terminal according to claim 1, characterized in that, The modules are connected in the following way: Collect sampling data from acoustic sensors deployed at nodes of the water supply network, and simultaneously acquire the communication status, sampling time, and environmental parameters of the terminal devices to construct the original monitoring data set; The original monitoring dataset is subjected to structured feature encoding to extract low-dimensional acoustic representation vectors; Based on the terminal installation location corresponding to the low-dimensional acoustic representation vector, a regional context prior vector is constructed and input into the TabPFN model to dynamically activate the pre-trained task path embedded within it, generate prior path identifiers, perform multi-state classification inference, and generate leakage state labels and judgment confidence values. Based on the judgment confidence value, an abnormal confidence value verification process is performed. If it is lower than the set threshold, the remote collaborative processing process is triggered. The low-dimensional acoustic representation vector, regional context prior vector and prior path identifier are uploaded to the remote platform, the backup model is loaded for re-judgment processing, and the final leakage status judgment result and auxiliary inspection identifier are generated. Based on the water leakage status judgment result, communication status and auxiliary inspection identifier, the event-level data reporting scheduling strategy is executed. If the judgment result is a leakage status and the communication status meets the minimum transmission power condition, the control terminal performs an acoustic data back transmission operation and completes data reporting with the current status label and auxiliary inspection identifier. If the judgment result is a non-leakage status and is identified as a false alarm, the terminal maintains the original sampling period configuration and records the local status log. The terminal leakage information is written into the remote monitoring database to store event information and perform suspected leak location identification processing.

3. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The sampling data of the acoustic sensor includes vibration waveform data, spectral envelope characteristics and energy distribution data within a unit time window; the communication status includes signal strength, uplink channel status and remaining transmission power; and the environmental parameters include background noise level and external vibration indicators.

4. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The extraction of the low-dimensional acoustic representation vector includes: Receive the raw monitoring data set, perform frame-by-frame processing based on a set time window, and organize it into a multi-channel feature matrix; Based on the multi-channel feature matrix, the initial acoustic feature structure is constructed by arranging the sound pressure amplitude, frequency response, and energy density in that order. Based on the initial acoustic feature structure, a spectrum channel selection operation is performed. By setting an energy threshold and a fixed channel interval parameter, frequency sub-intervals containing abnormal vibration features are selected to construct a spectrum screening feature set. Sparse feature transformation is performed on the spectral screening feature set to compress intra-frame repetitive amplitudes and periodic channel responses, generating a sparse mapping structure representation; The sparse mapping structure representation is input into the dimension adjustment module, which performs vectorization mapping and dimension recombination based on the required dimension structure input from the TabPFN model to generate a low-dimensional acoustic representation vector.

5. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The generation of the leakage status label and the determination confidence value includes: Obtain the terminal installation location bound to the low-dimensional acoustic representation vector, retrieve the geographical unit to which the location belongs, and extract the corresponding area label, water pressure parameter and historical leakage label; The location index encoding process is performed on the area label to obtain the area location encoding vector. The water pressure parameter is numerically normalized to obtain the standardized water pressure vector. The event frequency statistics process within the period is performed on the historical leakage label to obtain the historical event statistics vector. The area context prior vector is generated by combining the three sub-vectors, and the feature offset value is calculated based on the starting index position of the three sub-vectors. The region context prior vector is input into the TabPFN model. Based on the feature offset value, the unique corresponding pre-trained task path is selected and activated from the set of pre-trained task paths embedded in the TabPFN model, and a prior path identifier is generated. The TabPFN model performs forward inference processing, and outputs the corresponding leakage status label based on the input structure consisting of low-dimensional acoustic representation vector, regional context prior vector and prior path identifier, while generating a decision confidence value.

6. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The generation of the leakage status assessment result and auxiliary inspection indicators includes: Receive the judgment confidence value and compare it with the preset judgment threshold. If it is less than the judgment threshold, trigger the remote collaborative processing flow. The low-dimensional acoustic representation vector, regional context prior vector, and prior path identifier corresponding to the current terminal are encapsulated into a remote processing request data structure and uploaded to the remote platform. After receiving the remote processing request data structure, the remote platform loads the corresponding backup model, executes the re-judgment process, and generates the re-judgment result as the final leakage status judgment result. After completing the re-judgment process, the remote platform starts the auxiliary inspection module to perform false alarm elimination operation on the leakage status judgment result and generate the false alarm elimination judgment basis; Based on the criteria for eliminating false alarms, an auxiliary inspection identifier is generated through a judgment logic algorithm that integrates multi-source heterogeneous information.

7. A tap water leakage monitoring system based on a low-power terminal according to claim 6, characterized in that, The remote platform performs a re-judgment process, which includes: The backup model is invoked. Specifically, the backup model is a deep classification model built on a multi-class Transformer structure. It is deployed on a remote platform using a centralized training method and has end-to-end classification capability for multi-input field structures. Perform a re-judgment process, which includes: Receive the remote processing request data structure and combine the three data items into a further judgment input vector; The internal input adaptation module of the backup model is called to convert the re-judgment input vector into an embedded representation structure that conforms to the model input structure, and then input it into the backbone network of the backup model to perform the forward inference process. During forward inference, the backup model performs calculations based on the input vector structure and model training weights, and outputs the corresponding multi-class probability distribution. In a multi-class probability distribution, extract the leakage status label corresponding to the maximum probability value, and extract the maximum probability value as the judgment confidence value. Based on the leakage status label and the judgment confidence value, and combined with the corresponding input vector field index mapping relationship and the current model path identification information, a re-judgment result is constructed.

8. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The event-level data reporting scheduling strategy includes: The system receives the final water leakage status judgment result, terminal communication status and auxiliary inspection identifier. Based on the preset scheduling rule set, it determines whether the data reporting trigger condition is met. The data reporting trigger condition requires that the water leakage status judgment result is a leakage status, the remaining transmission power of the communication status is not lower than the minimum transmission power threshold, and the auxiliary inspection identifier indicates that there are clear abnormal characteristics in the current status. If the data reporting trigger condition is met, the control terminal starts the reporting execution module, encapsulates the low-dimensional acoustic representation vector, leakage status label and auxiliary inspection identifier within the current sampling period, and generates a data reporting structure for real-time reporting; If the data reporting triggering conditions are not met, the terminal will maintain the original sampling cycle configuration, periodically execute the sampling, encoding and local judgment process, and record the leakage status judgment result, auxiliary inspection mark and communication status of the current cycle. If the terminal detects the same leakage status judgment result in multiple consecutive cycles and the auxiliary inspection indicator continues to have an abnormal mark, it will automatically re-evaluate the reporting trigger condition. If the evaluation result is that the condition is met, the reporting process will be restarted.

9. A tap water leakage monitoring system based on a low-power terminal according to claim 2, characterized in that, The remote monitoring includes: Construct a structured event information data table, setting fields including terminal number, event timestamp, leakage status judgment result, judgment confidence value, auxiliary inspection identifier and prior path identifier. After receiving data reported by the terminal, write it into the remote monitoring database. The GIS interface data display module is invoked to obtain the terminal's geographical location information based on the terminal number and event timestamp fields and mark it in the geographic information layer. The historical judgment trajectory of the terminal is drawn, and the leakage status judgment result of the current period is dynamically superimposed to generate a terminal status time sequence display diagram. Based on the geographical location associated with the terminal number, the spatial coordinate information of that location in the GIS system is extracted and combined with the synchronous event records of other terminals in the same area in the structured event information data table to form a multi-point status spatial input set. The multi-point state space input set is input into the positioning mechanism module to perform suspected leak point location identification processing and generate the coordinates of the suspected leak point location.

Images