Intelligent water meter fault monitoring method and system based on Internet of Things

Abstract

The invention discloses an intelligent water meter fault monitoring method and system based on the Internet of Things, and relates to the technical field of water meter fault monitoring. Obtaining regional water affair data uploaded by the sub-regional nodes, carrying out feature extraction to obtain lightweight tags, and carrying out integrity verification; if the lightweight label is normal, decomposing the lightweight label to obtain label data, and performing variation verification on the label data; if the label data is normal, decoding the label data to obtain water affair data, and inputting the water affair data into a time sequence model to determine a fluctuation value; if the fluctuation value is abnormal, fault detection is carried out on the user node, a fault event is determined according to a fault detection result and the fluctuation value, alarm and watermark label variation verification are carried out according to the type of the fault event to further confirm the accuracy of data, and meanwhile, low computing power consumption is kept; the time sequence model analyzes the water consumption curve and combines the fault detection result and the fluctuation value, the system can quickly judge the fault type and give an alarm, and the water meter fault diagnosis efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of water meter fault monitoring, and in particular relates to an intelligent water meter fault monitoring method and system based on the Internet of Things. Background Art

[0002] Breakthroughs in electronics technology have significantly improved measurement stability with photoelectric coding and ultrasonic sensing. In recent years, the rise of the Internet of Things and edge computing has driven the widespread adoption of smart water meters. These meters integrate NB-IoT communications and AI-powered fault diagnosis, enabling real-time data analysis and remote control. However, current research focuses on issues such as the security of multi-source data fusion and adaptability to complex environments.

[0003] Patent number: CN119782759A, discloses an ultrasonic water meter fault prediction system based on machine learning. The ultrasonic water meter fault prediction system based on machine learning can monitor the operating status of the ultrasonic water meter in real time, predict potential faults in advance, and improve the timeliness of fault diagnosis; through the feature extraction module, fault-related features are extracted from the collected multi-parameter data, thereby improving the accuracy of fault prediction; the SVM model is simultaneously used for fault prediction, which has strong generalization ability, is suitable for processing high-dimensional data, and improves the robustness of the model.

[0004] Traditional methods often rely on high-computing power equipment to perform complex data analysis, and the fault location is inaccurate, the response time is long, and the labor costs are high, resulting in low efficiency of water meter fault detection and waste of computing power resources in traditional water systems. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that traditional methods often rely on high-computing power equipment to perform complex data analysis, and the fault location is inaccurate, the response time is long, and the labor cost is high, resulting in low efficiency of water meter fault detection and waste of computing power resources in traditional water systems. A smart water meter fault monitoring method and system based on the Internet of Things is proposed.

[0006] In a first aspect of the present invention, a method for monitoring faults of a smart water meter based on the Internet of Things is first proposed. The method comprises:

[0007] Obtain regional water affairs data uploaded by sub-regional nodes, extract features from the regional water affairs data to obtain lightweight labels, and verify the integrity of the lightweight labels;

[0008] If the lightweight tag is normal, the lightweight tag is decomposed according to the preset rules to obtain multiple tag data, and the tag data is mutated and verified according to the watermark tag;

[0009] If the tag data is normal, the tag data is decoded to obtain water service data, which is then input into a time series model for water consumption analysis to obtain a water consumption curve, and the fluctuation value is determined based on the water consumption curve; the water service data includes: water consumption and water pressure of the user node;

[0010] If the fluctuation value is abnormal, a fault detection is performed on the user node to obtain a fault detection result, a fault event is determined based on the fault detection result and the fluctuation value, and an alarm is issued based on the type of the fault event; the fault event types include: external force damage events, leakage events and erosion events.

[0011] Optionally, before decomposing the lightweight label according to a preset rule to obtain multiple label data, the following steps are performed:

[0012] The user node converts the water service data into binary form to obtain a binary code, and uploads the binary code to the sub-region node; the water service data includes the user node's water consumption, water pressure, and water meter ID;

[0013] The sub-area node receives the binary code uploaded by each user node, generates lightweight labels and preset rules for the user nodes in the sub-area, divides the lightweight labels according to the preset rules to obtain multiple watermark labels, splices the watermark labels with the binary code to obtain label data, matrix-connects the label data to obtain regional water affairs data, and uploads the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-area node.

[0014] Optionally, water service data can be input into a time series model to perform water consumption analysis to obtain a water consumption curve, including:

[0015] Obtain historical water service data, fault annotation data, and environmental configuration data, and input the historical water use data, the fault annotation data, and the environmental configuration data into a preset model for training; the environmental configuration data includes holiday data and temperature data;

[0016] The output of the preset model is used as a model update parameter, and the model parameters of the preset model are updated according to the model update parameters to obtain a time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is a water consumption curve in a future unit time, and the water consumption curve is used to calculate the slope to obtain a fluctuation value.

[0017] Optionally, a fault detection result is obtained by performing fault detection on the user node, wherein a low-resolution near-infrared camera and infrared fill light are installed above the smart water meter, including:

[0018] Obtain an initial image of the user node, upload the initial image to the sub-region node, so that the sub-region node extracts features from the initial image using the improved YOLOv8 model to obtain a target image, and upload the target image to the target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value;

[0019] Determine the pressure of the user node according to the water service data, and if the similarity value is greater than the similarity threshold and the pressure is within the pressure range, determine that the fault result is a corrosion event;

[0020] If the similarity value is greater than the similarity threshold and the pressure is not within the pressure range, the fault result is determined to be a leakage event;

[0021] If the similarity value is less than or equal to the similarity threshold, and the pressure is within the pressure range, the fault result is determined to be an external force damage event;

[0022] An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model;

[0023] The working principle of the improved GhostConv layer:

[0024] The initial feature map is used as the input of the improved GhostConv layer, and the initial feature map is subjected to feature extraction through the CBL module to obtain a first feature tensor. The initial feature map and the first feature tensor are respectively convolved through a 1×1 convolution layer to obtain a second feature tensor and a third feature tensor. The second feature tensor and the third feature tensor are concatenated to obtain a fourth feature tensor. The fourth feature tensor is subjected to channel dimension normalization through a BN layer to obtain a fifth feature tensor. The fifth feature tensor is activated through a Leaky ReLU activation function to obtain a sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

[0025] Optionally, improve the working principle of the MPCA layer, including:

[0026] Obtain an original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain a G feature map, perform a height-wise average pooling operation on the original feature map to obtain a Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain an X feature map;

[0027] Concatenating the Y feature map and the X feature map to obtain a first feature map, convolving the first feature map and the G feature map respectively through a 1×1 convolution layer to obtain a second feature map and a third feature map, and performing a split operation on the first feature map and the second feature map respectively to obtain a feature group; the feature group includes a first feature group and a second feature group, and the first feature group and the second feature group each include multiple sub-features;

[0028] Performing an element-by-element multiplication operation on the third feature map and the third feature map to obtain a fourth feature map, cross-performing an element-by-element multiplication operation on the first feature group and the second feature group to obtain a fifth feature map and a sixth feature map, and performing an element-by-element penalty operation on the original feature map, the fourth feature map, the fifth feature map, and the sixth feature map to obtain a final feature map.

[0029] In a second aspect of the present invention, a smart water meter fault monitoring system based on the Internet of Things is proposed, comprising: an integrity verification module, a variation verification module, a fluctuation value determination module, and a fault warning module:

[0030] The integrity verification module is used to obtain the regional water affairs data uploaded by the sub-regional nodes, perform feature extraction on the regional water affairs data to obtain lightweight labels, and perform integrity verification on the lightweight labels;

[0031] The variation verification module is used to decompose the lightweight tag according to preset rules to obtain multiple tag data if the lightweight tag is normal, and perform variation verification on the tag data according to the watermark tag;

[0032] The fluctuation value determination module is configured to decode the tag data to obtain water service data if the tag data is normal, input the water service data into a time series model to perform water consumption analysis to obtain a water consumption curve, and determine the fluctuation value based on the water consumption curve; the water service data includes: water consumption and water pressure of the user node;

[0033] The fault warning module is used to perform fault detection on the user node to obtain a fault detection result if the fluctuation value is abnormal, determine a fault event based on the fault detection result and the fluctuation value, and issue an alarm based on the type of the fault event; the fault event types include: external force damage events, leakage events, and erosion events.

[0034] Optionally, the system further includes: a data encoding module and a data lightweight module:

[0035] The data encoding module is used for the user node to convert the water service data into binary form to obtain binary code, and upload the binary code to the sub-region node; the water service data includes the water consumption, water pressure and water meter ID of the user node;

[0036] The data lightweight module is used for the sub-area node to receive the binary code uploaded by each user node, generate lightweight labels and preset rules for the user nodes in the sub-area, segment the lightweight labels according to the preset rules to obtain multiple watermark labels, splice the watermark labels with the binary code to obtain label data, matrix-connect the label data to obtain regional water affairs data, and upload the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-area node.

[0037] Optionally, the fluctuation value determination module includes: a training data acquisition module and a model update module:

[0038] The training data acquisition module is used to acquire historical water service data, fault marking data and environmental configuration data, and input the historical water use data, the fault marking data and the environmental configuration data into a preset model for training; the environmental configuration data includes holiday data and temperature data;

[0039] The model update module is used to use the output of the preset model as a model update parameter, and update the model parameters of the preset model according to the model update parameter to obtain a time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is a water consumption curve in a future unit time, and the water consumption curve is used to calculate the slope to obtain a fluctuation value.

[0040] Optionally, the fault warning module includes: an image acquisition module, a first judgment module, a second judgment module and a third judgment module:

[0041] The image acquisition module is used to acquire an initial image of the user node, upload the initial image to the sub-region node, so that the sub-region node extracts features of the initial image using the improved YOLOv8 model to obtain a target image, and upload the target image to the target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value;

[0042] The first judgment module is configured to determine the pressure of the user node based on the water service data, and if the similarity value is greater than the similarity threshold and the pressure is within the pressure range, determine that the fault result is a corrosion event;

[0043] The second judgment module is configured to determine that the fault result is a leakage event if the similarity value is greater than the similarity threshold and the pressure is not within the pressure range;

[0044] The third judgment module is configured to determine that the fault result is an external force damage event if the similarity value is less than or equal to the similarity threshold and the pressure is within the pressure range;

[0045] An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model;

[0046] The working principle of the improved GhostConv layer:

[0047] The initial feature map is used as the input of the improved GhostConv layer, and the initial feature map is subjected to feature extraction through the CBL module to obtain a first feature tensor. The initial feature map and the first feature tensor are respectively convolved through a 1×1 convolution layer to obtain a second feature tensor and a third feature tensor. The second feature tensor and the third feature tensor are concatenated to obtain a fourth feature tensor. The fourth feature tensor is subjected to channel dimension normalization through a BN layer to obtain a fifth feature tensor. The fifth feature tensor is activated through a Leaky ReLU activation function to obtain a sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

[0048] Optionally, improve the working principle of the MPCA layer, including:

[0049] Obtain an original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain a G feature map, perform a height-wise average pooling operation on the original feature map to obtain a Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain an X feature map;

[0050] Concatenating the Y feature map and the X feature map to obtain a first feature map, convolving the first feature map and the G feature map respectively through a 1×1 convolution layer to obtain a second feature map and a third feature map, and performing a split operation on the first feature map and the second feature map respectively to obtain a feature group; the feature group includes a first feature group and a second feature group, and the first feature group and the second feature group each include multiple sub-features;

[0051] Performing an element-by-element multiplication operation on the third feature map and the third feature map to obtain a fourth feature map, cross-performing an element-by-element multiplication operation on the first feature group and the second feature group to obtain a fifth feature map and a sixth feature map, and performing an element-by-element penalty operation on the original feature map, the fourth feature map, the fifth feature map, and the sixth feature map to obtain a final feature map.

[0052] Beneficial effects of the present invention:

[0053] The present invention proposes a smart water meter fault monitoring method based on the Internet of Things. The method obtains regional water service data uploaded by sub-region nodes, extracts features, obtains lightweight labels, and performs integrity verification. If the lightweight label is normal, the lightweight label is decomposed to obtain label data, and the label data is verified for variation. If the label data is normal, the label data is decoded to obtain water service data, and the water service data is input into a time series model to determine the fluctuation value. If the fluctuation value is abnormal, the user node is detected for faults, and the fault event is determined according to the fault detection result and the fluctuation value. An alarm is issued according to the type of the fault event. The variation verification of the watermark label further confirms the accuracy of the data while maintaining low computing power consumption. The time series model analyzes the water use curve and combines the fault detection result and the fluctuation value. The system can quickly determine the fault type and issue an alarm, thereby improving the efficiency and accuracy of water meter fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] The present invention will be further described below with reference to the accompanying drawings.

[0055] Figure 1 A flowchart of a method for monitoring faults of an intelligent water meter based on the Internet of Things provided by an embodiment of the present invention;

[0056] Figure 2 A framework diagram of an IoT-based smart water meter fault monitoring system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention, in conjunction with the accompanying drawings. Obviously, the described embodiments represent only a portion of the embodiments of the present invention, not all of them. The term "and / or" herein simply describes an association relationship between associated objects, indicating that three possible relationships exist. For example, "A" and "B" can represent: A exists alone, A and B exist simultaneously, or B exists alone. Furthermore, references to "first," "second," and so on in the present invention are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined as "first" or "second" may explicitly or implicitly include at least one of these features. Furthermore, the technical solutions of the various embodiments may be combined, but only if they are achievable by a person of ordinary skill in the art. If a combination of technical solutions contradicts or is unachievable, such combination shall be deemed non-existent and outside the scope of protection claimed by the present invention.

[0058] Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative work shall fall within the scope of protection of the present invention.

[0059] The embodiment of the present invention provides a method for monitoring faults of smart water meters based on the Internet of Things. Figure 1 , Figure 1 A flowchart of a method for monitoring faults in a smart water meter based on the Internet of Things is provided in an embodiment of the present invention. The method comprises the following steps:

[0060] S101, obtaining regional water affairs data uploaded by the sub-regional node, performing feature extraction on the regional water affairs data to obtain lightweight labels, and performing integrity verification on the lightweight labels;

[0061] S102: If the lightweight tag is normal, decompose the lightweight tag according to a preset rule to obtain multiple tag data, and perform mutation verification on the tag data according to the watermark tag;

[0062] S103: If the tag data is normal, decode the tag data to obtain water service data, input the water service data into a time series model to perform water consumption analysis to obtain a water consumption curve, and determine the fluctuation value based on the water consumption curve;

[0063] S104: If the fluctuation value is abnormal, a fault detection is performed on the user node to obtain a fault detection result, a fault event is determined based on the fault detection result and the fluctuation value, and an alarm is issued based on the type of the fault event;

[0064] Water service data includes: water consumption and water pressure at user nodes; fault event types include: external force damage events, leakage events and erosion events.

[0065] An IoT-based smart water meter fault monitoring method provided by an embodiment of the present invention ensures correct data transmission through integrity verification of lightweight tags, reduces the complexity of data processing, and further confirms the accuracy of the data through variation verification of watermark tags while maintaining low computing power consumption. The timing model analyzes the water consumption curve and combines the fault detection results and fluctuation values. The system can quickly determine the fault type and issue an alarm, thereby improving the efficiency and accuracy of water meter fault diagnosis.

[0066] In one implementation, the target node extracts features from the regional water data uploaded by the sub-regional node to obtain a lightweight tag, and then verifies the integrity of the lightweight tag. This process ensures that the data is not tampered with or damaged during transmission, thus guaranteeing the integrity and reliability of the data. Integrity verification is usually implemented through checksums, CRC checks, or other hash algorithms. For example, when a sub-regional node generates a lightweight tag, it calculates a check value and appends it to the tag. After receiving the data, the target node recalculates the check value and compares it with the appended check value. If the two are consistent, the lightweight tag is complete and correct.

[0067] In one implementation, if the lightweight tag passes integrity verification, the target node will decompose the lightweight tag into multiple tag data according to preset rules and perform mutation verification on these tag data based on the watermark tag. Mutation verification is used to detect whether unforeseen changes have occurred to the tag data during transmission, such as random noise or malicious tampering. The verification process usually involves comparing the correlation between the watermark tag and the tag data. For example, the watermark tag is generated based on the specific characteristics of the tag data, and the target node determines whether the data is normal by verifying the degree of match between the watermark tag and the tag data. If the match is lower than the preset threshold, it is considered that the tag data has mutated.

[0068] In one implementation, if the label data passes the mutation verification, the target node decodes the label data, restores the original water service data, and inputs this data into the time series model for water consumption analysis to obtain a water consumption curve. This process realizes the conversion from binary data to actual water consumption, and models and analyzes water consumption behavior through the time series model. The water consumption curve reflects the changes in the user's water consumption at different time points (and in the future). For example, by analyzing the water consumption curve, it is possible to discover the user's water consumption habits in a specific time period, such as peak water consumption in the morning and evening. The time series model can identify abnormal fluctuations in the water consumption curve, such as an increase in water consumption within a cycle or a sharp increase in water consumption in a short period of time.

[0069] In one implementation, the fluctuation value is positively correlated with the rate of change of the slope of the water usage curve. When the rate of change of the slope increases within a preset time period, it indicates that the fluctuation value is increasing. If the fluctuation value is abnormal, the target node performs fault detection on the target water meter (water meter appearance inspection and water pressure inspection, with a low-resolution near-infrared (NIR) camera and infrared fill light installed on the water meter), obtains the fault detection result, and determines the type of fault event based on the fault detection result and the fluctuation value. For example: If the fluctuation value shows a sharp increase in water consumption and abnormal water pressure, the abnormality may be caused by a water supply pipe leak; if the fluctuation value is abnormal and the water pressure meter appearance is abnormal, the water meter may have been damaged by external forces; if the fluctuation value is abnormal, but the appearance and water pressure are normal, the water meter internal sensor may be abnormal.

[0070] In one implementation, alarms are generated based on the type of fault event. This process ensures the system can respond promptly to abnormal situations, minimizing losses and impacts. For example, if a leak is detected, the system immediately issues an alarm, notifying maintenance personnel to address the situation, preventing water waste and potential water damage. If corrosion is detected, the system also issues an alarm, prompting maintenance personnel to inspect the water meter for corrosion and promptly replace damaged components. This alarm mechanism improves system reliability and responsiveness, enhancing the efficiency and safety of water management.

[0071] In one embodiment, before decomposing the lightweight tag according to a preset rule to obtain multiple tag data, the following steps are included:

[0072] The user node converts the water service data into binary form to obtain binary code, and uploads the binary code to the sub-region node; the water service data includes the user node's water consumption, water pressure, and water meter ID;

[0073] The sub-region node receives the binary code uploaded by each user node, generates lightweight labels and preset rules for the user nodes in the sub-region, divides the lightweight labels according to the preset rules to obtain multiple watermark labels, splices the watermark labels with the binary code to obtain label data, connects the label data in a matrix to obtain regional water affairs data, and uploads the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-region node.

[0074] In one implementation, each sub-area corresponds to a lightweight tag (the lightweight tag is also binary-coded), and a preset rule is determined based on the number of user nodes in the sub-area, and the preset rule is used to segment the lightweight tags; the target area contains a target node, and the target node is used to upload data to the cloud system and perform data analysis. The target area contains multiple sub-areas, each sub-area has a sub-area node, each sub-area contains multiple users, and each user corresponds to a user node; the above-mentioned nodes are smart water meters, and the computing capabilities of each water meter are different (target node > sub-area node > user node).

[0075] In one implementation, by converting water service data (such as water pressure, water consumption, and water meter ID) into binary form, user nodes can upload data to sub-region nodes in a compact format. This binary encoding method significantly reduces the amount of data transmission and improves transmission efficiency. For example, if a data packet containing water consumption (integer) and water meter ID (string) is directly transmitted in text form, it may occupy a lot of bandwidth; after conversion to binary, the amount of data is greatly reduced, which is suitable for efficient transmission under low-bandwidth, high-latency network conditions in the Internet of Things environment; Binary: Water consumption (integer, unit: liter) can be directly converted into binary form, for example: 10 liters of water is binary 1010; Water pressure (floating point number, unit: Pascal) can be binary encoded using the IEEE754 standard. For example, the floating point number 3.14 is represented in binary as 01000000010010010000111111011011 in IEEE754 single-precision (32-bit) format.

[0076] In one implementation, a sub-region node generates a lightweight label for each user node, splits it into multiple watermark labels, and then concatenates these watermark labels with a binary code. This binding method uniquely identifies each user node's data during transmission, and the watermark labels enable rapid verification of data integrity and authenticity. For example, if a data segment is tampered with during transmission, the receiver can detect the anomaly by verifying the watermark label without requiring complex verification of the entire data segment, thereby reducing computational costs.

[0077] In one implementation, sub-region nodes connect the label data of each user node in a matrix to form a regional water data matrix. This matrix organization facilitates batch processing and analysis by the target node. For example, a target node can receive matrix data from multiple sub-regions at once and quickly calculate information such as total water consumption and average water consumption in each region through matrix operations, improving the efficiency and flexibility of data processing. Matrix data also facilitates subsequent data analysis and mining, such as using machine learning algorithms to process matrix data and predict water usage trends.

[0078] In one implementation, a hierarchical architecture (user nodes, sub-region nodes, and target nodes) and on-demand allocation of computing resources enable efficient system operation. User nodes are responsible for data collection and simple encoding, sub-region nodes are responsible for lightweight label generation and data splicing, and target nodes are responsible for data aggregation and analysis. This architecture fully utilizes the computing power of different nodes, avoiding waste of computing resources. For example, user nodes with weaker computing power only need to perform simple binary encoding, while more powerful sub-region nodes and target nodes handle complex label processing and data analysis, ensuring overall system performance.

[0079] In one embodiment, water service data is input into a time series model to perform water consumption analysis to obtain a water consumption curve, including:

[0080] Obtain historical water service data, fault annotation data, and environmental configuration data, and input these data into a preset model for training; the environmental configuration data includes holiday data and temperature data;

[0081] The output of the preset model is used as the model update parameter, and the model parameters of the preset model are updated according to the model update parameter to obtain the time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is the water consumption curve in the future unit time, and the water consumption curve is used to calculate the slope to obtain the fluctuation value.

[0082] In one implementation, historical water service data: user water consumption data and water pressure collected from smart water meters, arranged in time series, including data at different time granularities such as hours, days, and weeks; fault annotation data: water consumption data containing known fault events (external damage, leakage, erosion, etc.), these data should be marked with the fault type and occurrence time, and used for the model to learn fault characteristics; environmental configuration data: holiday data and temperature data, these data can help the model understand the external factors of water consumption changes; the preset models can be: 1. ARIMA (autoregressive integrated moving average model), which is suitable for stationary interval sequence data and can capture the autocorrelation of data; 2. LSTM (long short-term memory network), a deep learning model, which is particularly suitable for processing long sequence data and can learn complex time series patterns; 3. Hybrid model, which combines the advantages of multiple models, such as using ARIMA for short-term prediction and LSTM for long-term trend analysis.

[0083] In one implementation, model parameters include hyperparameters, feature engineering parameters, and model configuration parameters. 1. Hyperparameters: such as learning rate, regularization parameter, batch size, etc. These parameters affect the training process and performance of the model. 2. Feature engineering parameters: such as time window size, sliding step size, etc. These parameters affect the shape of the input data and the predictive ability of the model. 3. Model configuration parameters: such as the parameters of the autoregressive term, difference term, and sliding average term in the ARIMA model; the number of neurons and layers in the LSTM.

[0084] In one implementation, the time series model is trained by integrating historical water use data, fault annotation data, and environmental configuration data to accurately predict future water use curves per unit time. This predictive capability helps water utilities plan resource allocation in advance, optimize water supply scheduling, and ensure the efficient operation of the water supply system. The time series model can dynamically update model parameters based on the latest data, including hyperparameters, feature engineering parameters, and model configuration parameters. This dynamic optimization capability enables the model to continuously adapt to changes in user water use behavior (periodically updated, with an annual update cycle, for example, once a year), maintain high-precision prediction and anomaly detection performance, and ensure the long-term stable operation of the system.

[0085] In one embodiment, a fault detection result is obtained by performing fault detection on a user node. A low-resolution near-infrared camera and infrared fill light are installed above the smart water meter, including:

[0086] Obtain the initial image of the user node, upload the initial image to the sub-region node, so that the sub-region node extracts features from the initial image through the improved YOLOv8 model to obtain the target image, and upload the target image to the target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value;

[0087] Determine the pressure of the user node based on the water service data. If the similarity value is greater than the similarity threshold and the pressure is within the pressure range, the fault result is determined to be a corrosion event.

[0088] If the similarity value is greater than the similarity threshold and the pressure is not within the pressure range, the fault result is determined to be a leakage event;

[0089] If the similarity value is less than or equal to the similarity threshold, and the pressure is within the pressure range, the fault result is determined to be an external force damage event;

[0090] An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model;

[0091] Improved working principle of GhostConv layer:

[0092] The initial feature map is used as the input of the improved GhostConv layer. The initial feature map is extracted by the CBL module to obtain the first feature tensor. The initial feature map and the first feature tensor are convolved by a 1×1 convolution layer to obtain the second feature tensor and the third feature tensor respectively. The second feature tensor and the third feature tensor are concatenated to obtain the fourth feature tensor. The fourth feature tensor is normalized by the BN layer to obtain the fifth feature tensor. The fifth feature tensor is activated by the Leaky ReLU activation function to obtain the sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

[0093] In one implementation, the system accurately classifies different fault events by combining image similarity calculations with pressure data. Using an improved YOLOv8 model to extract image features and compare them with standard images for similarity, combined with pressure monitoring data, the system can accurately identify erosion, leakage, or external damage events. This diagnostic method based on image and data fusion not only enhances the system's diagnostic capabilities but also improves the comprehensiveness and reliability of fault diagnosis. The entire process is automated and real-time, significantly shortening fault response time, improving the operational efficiency of the water system, reducing operation and maintenance costs, and ensuring the stability and safety of the water supply system.

[0094] In one implementation, the first feature tensor (basic feature) is extracted through the CBL module, while the initial feature map is retained for 1×1 convolution to generate the second feature tensor (detail feature), which is combined with dual-path feature splicing (third feature tensor) to effectively retain the multi-scale information of the input and enhance the representation ability of complex textures and small targets; the channel dimension of the spliced features is normalized through the BN layer (fourth to fifth feature tensors) to eliminate the distribution differences of cross-branch feature fusion and improve feature consistency; the Leaky ReLU activation function (fifth to sixth feature tensors) introduces controllable negative interval activation to alleviate the gradient vanishing problem and enhance the model's sensitivity to low-contrast targets; an MPCA (multi-path channel attention) layer is added between the eighth and ninth layers to dynamically allocate channel weights to strengthen the expression of important feature channels, suppress redundant information, and enhance Backbone's ability to focus on key targets.

[0095] In one implementation, C3Ghost reconstructs cross-layer connections based on the GhostConv module, retaining the multi-branch interaction advantages of the C3 module, while further compressing the model volume through the lightweight characteristics of GhostConv, reducing memory usage, and accelerating the inference speed; the MPCA layer uses a multi-path channel attention mechanism to adaptively fuse feature information at different levels to enhance the model's robustness to occluded and deformed targets (for example, accurately identifying and removing water pipes in the image, focusing attention on the water meter part); the initial feature map directly participates in the 1×1 convolution branch (generating the second feature tensor), avoiding the loss of details caused by feature splitting in traditional GhostConv, especially improving the edge clarity of small targets (such as electronic components); in the Neck part, the C3Ghost layer uses lightweight cross-layer connections (GhostConv replaces standard Conv) to balance the fusion quality of shallow high-resolution features and deep semantic features, thereby improving the detection consistency of multi-scale targets.

[0096] In one embodiment, the working principle of the MPCA layer is improved, including:

[0097] Get the original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain the G feature map, perform a height-wise average pooling operation on the original feature map to obtain the Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain the X feature map;

[0098] The Y feature map and the X feature map are connected to obtain the first feature map. The first feature map and the G feature map are convolved through a 1×1 convolution layer to obtain the second feature map and the third feature map. The first feature map and the second feature map are split to obtain a feature group. The feature group includes the first feature group and the second feature group. The first feature group and the second feature group both contain multiple sub-features.

[0099] The third feature map is element-wise multiplied with the third feature map to obtain the fourth feature map, the first feature group and the second feature group are cross-operated element-wise multiplied to obtain the fifth feature map and the sixth feature map, and the original feature map, the fourth feature map, the fifth feature map and the sixth feature map are subjected to element-wise penalty operation to obtain the final feature map.

[0100] In one implementation, global average pooling (G Avg Pool) performs mean compression on the input feature map along all spatial locations (H×W) to generate a C×1×1 G feature map, which reflects the global statistical characteristics of the channel dimension. Height-wise average pooling (Y Avg Pool) compresses the feature map along its height (H) to a C×1×W Y feature map, preserving spatial correlation in the width direction (such as the continuity of horizontal defects). Width-wise average pooling (X Avg Pool) compresses the feature map along its width (W) to an C×H×1 X feature map, preserving spatial correlation in the height direction (such as the ductility of vertical defects). Through pooling operations in different dimensions, both global channel information and local spatial details are captured simultaneously, avoiding the information loss caused by single pooling. The G feature map emphasizes channel importance, and the Y / X feature maps encode the vertical and horizontal spatial distribution characteristics of defects, respectively, laying the foundation for subsequent multi-scale feature fusion.

[0101] In one implementation, concatenation (Concat): concatenates the Y feature map (C×1×W) and the X feature map (C×H×1) along the channel dimension to form a first feature map (2C×H×W), fusing spatial information in the height and width directions. 1×1 convolution: performs convolution operations on the concatenated first feature map and G feature map respectively to generate a second feature map (channel dimensionality reduction or reorganization) and a third feature map (channel attention weight). Split operation: splits the first and second feature maps into multiple sub-feature groups to refine the feature granularity (e.g., dividing them into local regions of different scales). Cross-dimensional feature fusion is achieved through concatenation and convolution. The 1×1 convolution layer (with fewer parameters) dynamically adjusts the channel dimension to enhance feature expression capabilities. The split operation decomposes the features into finer-grained sub-features, making it easier for the subsequent attention mechanism to differentially weight different regions, thereby improving sensitivity to minor defects (such as surface scratches).

[0102] In one implementation, element-wise multiplication (Mul) is performed: the third feature map is multiplied by itself (channel attention weights) to generate the fourth feature map, strengthening the weights of significant channels. Feature group cross-multiplication is performed: the first feature group (spatial refinement features) is cross-channel element-wise multiplied with the second feature group (channel adjustment features) to generate the fifth and sixth feature maps, implementing joint spatial-channel attention computation. The self-multiplication operation enhances the activation strength of important channels through nonlinear interactions, while the cross-multiplication forces information exchange between different feature groups, simulating the key-value pair relationship in the self-attention mechanism. This design dynamically allocates attention weights, allowing the model to focus on defect-related areas (such as crack edges) and suppress background noise. For example, in the first and second feature groups, each group contains two sub-features: the first feature group contains sub-features A and B, and the second feature group contains sub-features α and β. Sub-features A and α are multiplied, while sub-features B and β are multiplied.

[0103] In one implementation, the original feature map, the fourth feature map (enhanced channel weights), the fifth feature map (spatial-channel interaction features), and the sixth feature map (cross-group refined features) are weighted and summed element-by-element. The weight range is constrained (e.g., normalized to [0, 1]) using activation functions such as Sigmoid. Element-by-element multiplication balances the contributions of different feature maps, preventing a single feature from dominating the output. The original feature map retains basic information, the fourth feature map enhances channel importance, and the fifth and sixth feature maps supplement spatial details. The final output is a defect-sensitive feature that combines global semantics with local details. This collaborative enhancement mechanism significantly improves the ability to distinguish weakly significant defects (such as low-contrast stains) while maintaining computational efficiency.

[0104] In one implementation, the element-by-element penalty operation refers to dynamically adjusting the weights or performing nonlinear transformations on the corresponding elements of multiple feature maps (such as the original feature map, the fourth, fifth, and sixth feature maps) at the same spatial position and channel, thereby fusing multi-stage features by suppressing redundant features and enhancing key information. Its core is to assign differentiated contribution weights to elements at the same position in different feature maps through element-wise mathematical operations (such as multiplication, weighted summation, or gating mechanisms) or activation functions (such as Sigmoid thresholding), thereby preserving details (the original feature map) while strengthening high-response areas after normalization (the fifth feature map) and activation (the sixth feature map), ultimately improving the effectiveness and robustness of the feature map.

[0105] Based on the same inventive concept, the embodiment of the present invention also provides a smart water meter fault monitoring system based on the Internet of Things. Figure 2 , Figure 2A schematic diagram of the structure of an IoT-based smart water meter fault monitoring system provided by an embodiment of the present invention includes: an integrity verification module, a variation verification module, a fluctuation value determination module, and a fault warning module:

[0106] The integrity verification module is used to obtain the regional water affairs data uploaded by the sub-regional nodes, extract features from the regional water affairs data to obtain lightweight labels, and perform integrity verification on the lightweight labels;

[0107] The variation verification module is used to decompose the lightweight label according to preset rules to obtain multiple label data if the lightweight label is normal, and perform variation verification on the label data according to the watermark label;

[0108] The fluctuation value determination module is used to decode the tag data to obtain water service data if the tag data is normal, input the water service data into the time series model to perform water consumption analysis to obtain a water consumption curve, and determine the fluctuation value based on the water consumption curve; the water service data includes: water consumption and water pressure of the user node;

[0109] The fault warning module is used to perform fault detection on the user node to obtain the fault detection result if the fluctuation value is abnormal, determine the fault event based on the fault detection result and the fluctuation value, and issue an alarm based on the type of fault event; the fault event types include: external force damage event, leakage event and erosion event.

[0110] An IoT-based smart water meter fault monitoring system provided by an embodiment of the present invention ensures correct data transmission through integrity verification of lightweight tags, reduces the complexity of data processing, and further confirms the accuracy of the data through variation verification of watermark tags while maintaining low computing power consumption. The timing model analyzes the water usage curve and combines the fault detection results and fluctuation values. The system can quickly determine the fault type and issue an alarm, thereby improving the efficiency and accuracy of water meter fault diagnosis.

[0111] In one embodiment, the system further includes: a data encoding module and a data lightweight module:

[0112] The data encoding module is used by the user node to convert water service data into binary form to obtain binary code, and then upload the binary code to the sub-region node; the water service data includes the user node's water consumption, water pressure and water meter ID;

[0113] The data lightweight module is used for the sub-region node to receive the binary code uploaded by each user node, generate lightweight labels and preset rules for the user nodes in the sub-region, segment the lightweight labels according to the preset rules to obtain multiple watermark labels, splice the watermark labels with the binary code to obtain label data, matrix-connect the label data to obtain regional water affairs data, and upload the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-region node.

[0114] In one embodiment, the fluctuation value determination module includes: a training data acquisition module and a model update module:

[0115] The training data acquisition module is used to obtain historical water service data, fault annotation data, and environmental configuration data, and input the historical water use data, fault annotation data, and environmental configuration data into the preset model for training; the environmental configuration data includes holiday data and temperature data;

[0116] The model update module is used to use the output of the preset model as the model update parameter, and update the model parameters of the preset model according to the model update parameter to obtain a time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is the water consumption curve in the future unit time, and the water consumption curve is used to calculate the slope to obtain the fluctuation value.

[0117] In one embodiment, the fault warning module includes: an image acquisition module, a first judgment module, a second judgment module, and a third judgment module:

[0118] An image acquisition module is used to acquire an initial image of a user node, upload the initial image to a sub-region node, so that the sub-region node extracts features from the initial image using an improved YOLOv8 model to obtain a target image, and upload the target image to a target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value;

[0119] The first judgment module is used to determine the pressure of the user node based on the water service data. If the similarity value is greater than the similarity threshold and the pressure is within the pressure range, the fault result is determined to be a corrosion event;

[0120] A second judgment module is configured to determine that the fault result is a leakage event if the similarity value is greater than the similarity threshold and the pressure is not within the pressure range;

[0121] A third judgment module is configured to determine that the fault result is an external force damage event if the similarity value is less than or equal to the similarity threshold and the pressure is within the pressure range;

[0122] An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model;

[0123] Improved working principle of GhostConv layer:

[0124] The initial feature map is used as the input of the improved GhostConv layer. The initial feature map is extracted by the CBL module to obtain the first feature tensor. The initial feature map and the first feature tensor are convolved by a 1×1 convolution layer to obtain the second feature tensor and the third feature tensor respectively. The second feature tensor and the third feature tensor are concatenated to obtain the fourth feature tensor. The fourth feature tensor is normalized by the BN layer to obtain the fifth feature tensor. The fifth feature tensor is activated by the Leaky ReLU activation function to obtain the sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

[0125] In one embodiment, the working principle of the MPCA layer is improved, including:

[0126] Get the original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain the G feature map, perform a height-wise average pooling operation on the original feature map to obtain the Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain the X feature map;

[0127] The Y feature map and the X feature map are connected to obtain the first feature map. The first feature map and the G feature map are convolved through a 1×1 convolution layer to obtain the second feature map and the third feature map. The first feature map and the second feature map are split to obtain a feature group. The feature group includes the first feature group and the second feature group. The first feature group and the second feature group both contain multiple sub-features.

[0128] The third feature map is element-wise multiplied with the third feature map to obtain the fourth feature map, the first feature group and the second feature group are cross-operated element-wise multiplied to obtain the fifth feature map and the sixth feature map, and the original feature map, the fourth feature map, the fifth feature map and the sixth feature map are subjected to element-wise penalty operation to obtain the final feature map.

[0129] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and should not be considered to limit the scope of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for monitoring faults of smart water meters based on the Internet of Things, characterized in that: The method comprises: Obtain regional water affairs data uploaded by sub-regional nodes, extract features from the regional water affairs data to obtain lightweight labels, and verify the integrity of the lightweight labels; If the lightweight tag is normal, the lightweight tag is decomposed according to the preset rules to obtain multiple tag data, and the tag data is mutated and verified according to the watermark tag; If the tag data is normal, the tag data is decoded to obtain water service data, which is then input into a time series model for water consumption analysis to obtain a water consumption curve, and the fluctuation value is determined based on the water consumption curve; the water service data includes: water consumption and water pressure of the user node; If the fluctuation value is abnormal, a fault detection is performed on the user node to obtain a fault detection result, a fault event is determined based on the fault detection result and the fluctuation value, and an alarm is issued based on the type of the fault event; the fault event types include: external force damage events, leakage events and erosion events.

2. The method for monitoring faults of smart water meters based on the Internet of Things according to claim 1, characterized in that: Before decomposing lightweight tags into multiple tag data using preset rules, the following steps are performed: The user node converts the water service data into binary form to obtain a binary code, and uploads the binary code to the sub-region node; the water service data includes the water consumption, water pressure and water meter ID of the user node; The sub-area node receives the binary code uploaded by each user node, generates lightweight labels and preset rules for the user nodes in the sub-area, divides the lightweight labels according to the preset rules to obtain multiple watermark labels, splices the watermark labels with the binary code to obtain label data, matrix-connects the label data to obtain regional water affairs data, and uploads the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-area node.

3. The method for monitoring faults of smart water meters based on the Internet of Things according to claim 1, characterized in that: Input water service data into the time series model to analyze water consumption and obtain water consumption curves, including: Obtain historical water service data, fault annotation data, and environmental configuration data, and input the historical water use data, the fault annotation data, and the environmental configuration data into a preset model for training; the environmental configuration data includes holiday data and temperature data; The output of the preset model is used as a model update parameter, and the model parameters of the preset model are updated according to the model update parameters to obtain a time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is a water consumption curve in a future unit time, and the water consumption curve is used to calculate the slope to obtain a fluctuation value.

4. The method for monitoring faults of smart water meters based on the Internet of Things according to claim 1, characterized in that: Perform fault detection on the user node to obtain a fault detection result. A low-resolution near-infrared camera and infrared fill light are installed above the smart water meter, including: Obtain an initial image of the user node, upload the initial image to the sub-region node, so that the sub-region node extracts features from the initial image using the improved YOLOv8 model to obtain a target image, and upload the target image to the target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value; Determine the pressure of the user node according to the water service data, and if the similarity value is greater than the similarity threshold and the pressure is within the pressure range, determine that the fault result is a corrosion event; If the similarity value is greater than the similarity threshold and the pressure is not within the pressure range, the fault result is determined to be a leakage event; If the similarity value is less than or equal to the similarity threshold, and the pressure is within the pressure range, the fault result is determined to be an external force damage event; An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model; The working principle of the improved GhostConv layer: The initial feature map is used as the input of the improved GhostConv layer, and the initial feature map is subjected to feature extraction through the CBL module to obtain a first feature tensor. The initial feature map and the first feature tensor are respectively convolved through a 1×1 convolution layer to obtain a second feature tensor and a third feature tensor. The second feature tensor and the third feature tensor are concatenated to obtain a fourth feature tensor. The fourth feature tensor is subjected to channel dimension normalization through a BN layer to obtain a fifth feature tensor. The fifth feature tensor is activated through a Leaky ReLU activation function to obtain a sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

5. The method for monitoring faults of smart water meters based on the Internet of Things according to claim 4, characterized in that: Improvements to the working principle of the MPCA layer, including: Obtain an original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain a G feature map, perform a height-wise average pooling operation on the original feature map to obtain a Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain an X feature map; Concatenating the Y feature map and the X feature map to obtain a first feature map, convolving the first feature map and the G feature map respectively through a 1×1 convolution layer to obtain a second feature map and a third feature map, and performing a split operation on the first feature map and the second feature map respectively to obtain a feature group; the feature group includes a first feature group and a second feature group, and the first feature group and the second feature group each include multiple sub-features; Performing an element-by-element multiplication operation on the third feature map and the third feature map to obtain a fourth feature map, cross-performing an element-by-element multiplication operation on the first feature group and the second feature group to obtain a fifth feature map and a sixth feature map, and performing an element-by-element penalty operation on the original feature map, the fourth feature map, the fifth feature map, and the sixth feature map to obtain a final feature map.

6. A smart water meter fault monitoring system based on the Internet of Things, characterized in that: The system includes: an integrity verification module, a variation verification module, a fluctuation value determination module and a fault warning module: The integrity verification module is used to obtain the regional water affairs data uploaded by the sub-regional nodes, perform feature extraction on the regional water affairs data to obtain lightweight labels, and perform integrity verification on the lightweight labels; The variation verification module is used to decompose the lightweight tag according to preset rules to obtain multiple tag data if the lightweight tag is normal, and perform variation verification on the tag data according to the watermark tag; The fluctuation value determination module is configured to decode the tag data to obtain water service data if the tag data is normal, input the water service data into a time series model to perform water consumption analysis to obtain a water consumption curve, and determine the fluctuation value based on the water consumption curve; the water service data includes: water consumption and water pressure of the user node; The fault warning module is used to perform fault detection on the user node to obtain a fault detection result if the fluctuation value is abnormal, determine a fault event based on the fault detection result and the fluctuation value, and issue an alarm based on the type of the fault event; the fault event types include: external force damage events, leakage events, and erosion events.

7. The intelligent water meter fault monitoring system based on the Internet of Things according to claim 6 is characterized in that: The system also includes: a data encoding module and a data lightweight module: The data encoding module is used for the user node to convert the water service data into binary form to obtain binary code, and upload the binary code to the sub-region node; the water service data includes the water consumption, water pressure and water meter ID of the user node; The data lightweight module is used for the sub-area node to receive the binary code uploaded by each user node, generate lightweight labels and preset rules for the user nodes in the sub-area, segment the lightweight labels according to the preset rules to obtain multiple watermark labels, splice the watermark labels with the binary code to obtain label data, matrix-connect the label data to obtain regional water affairs data, and upload the regional water affairs data to the target node; the target node is used to receive the data uploaded by each sub-area node.

8. The intelligent water meter fault monitoring system based on the Internet of Things according to claim 6 is characterized in that: The fluctuation value determination module includes: a training data acquisition module and a model update module: The training data acquisition module is used to acquire historical water service data, fault marking data and environmental configuration data, and input the historical water use data, the fault marking data and the environmental configuration data into a preset model for training; the environmental configuration data includes holiday data and temperature data; The model update module is used to use the output of the preset model as a model update parameter, and update the model parameters of the preset model according to the model update parameter to obtain a time series model; the model parameters include hyperparameters, feature engineering parameters and model configuration parameters; the output of the time series model is a water consumption curve in a future unit time, and the water consumption curve is used to calculate the slope to obtain a fluctuation value.

9. The intelligent water meter fault monitoring system based on the Internet of Things according to claim 6, characterized in that: The fault warning module includes: an image acquisition module, a first judgment module, a second judgment module and a third judgment module: The image acquisition module is used to acquire an initial image of the user node, upload the initial image to the sub-region node, so that the sub-region node extracts features of the initial image using the improved YOLOv8 model to obtain a target image, and upload the target image to the target node so that the target node calculates the similarity between the target image and the standard image to obtain a similarity value; The first judgment module is configured to determine the pressure of the user node based on the water service data, and if the similarity value is greater than the similarity threshold and the pressure is within the pressure range, determine that the fault result is a corrosion event; The second judgment module is configured to determine that the fault result is a leakage event if the similarity value is greater than the similarity threshold and the pressure is not within the pressure range; The third judgment module is configured to determine that the fault result is an external force damage event if the similarity value is less than or equal to the similarity threshold and the pressure is within the pressure range; An improved MPCA layer is added between the eighth and ninth layers in the YOLOv8 model, the Conv layer in the YOLOv8 model is replaced with an improved GhostConv layer, and the C2f layer in the YOLOv8 model is replaced with a C3Ghost layer to obtain an improved YOLOv8 model; The working principle of the improved GhostConv layer: The initial feature map is used as the input of the improved GhostConv layer, and the initial feature map is subjected to feature extraction through the CBL module to obtain a first feature tensor. The initial feature map and the first feature tensor are respectively convolved through a 1×1 convolution layer to obtain a second feature tensor and a third feature tensor. The second feature tensor and the third feature tensor are concatenated to obtain a fourth feature tensor. The fourth feature tensor is subjected to channel dimension normalization through a BN layer to obtain a fifth feature tensor. The fifth feature tensor is activated through a Leaky ReLU activation function to obtain a sixth feature tensor. The sixth feature tensor is input into the CBL module to obtain the original feature map.

10. The intelligent water meter fault monitoring system based on the Internet of Things according to claim 9, characterized in that: Improvements to the working principle of the MPCA layer, including: Obtain an original feature map, use the original feature map as the input of the improved MPCA layer, perform a global average pooling operation on the original feature map to obtain a G feature map, perform a height-wise average pooling operation on the original feature map to obtain a Y feature map, and perform a width-wise average pooling operation on the original feature map to obtain an X feature map; Concatenating the Y feature map and the X feature map to obtain a first feature map, convolving the first feature map and the G feature map respectively through a 1×1 convolution layer to obtain a second feature map and a third feature map, and performing a split operation on the first feature map and the second feature map respectively to obtain a feature group; the feature group includes a first feature group and a second feature group, and the first feature group and the second feature group each include multiple sub-features; Performing an element-by-element multiplication operation on the third feature map and the third feature map to obtain a fourth feature map, cross-performing an element-by-element multiplication operation on the first feature group and the second feature group to obtain a fifth feature map and a sixth feature map, and performing an element-by-element penalty operation on the original feature map, the fourth feature map, the fifth feature map, and the sixth feature map to obtain a final feature map.

Images