Real-time open channel clogging diagnosis system and method based on voiceprint and flow velocity fusion

Abstract

The invention relates to an open channel clogging real-time diagnosis system and method based on voiceprint and flow velocity fusion, and the method comprises the steps: a plurality of monitoring nodes are arranged along an open channel, the monitoring nodes are arranged at intervals according to the length of the channel, the terrain and the conditions of easy-to-deposit sections, and each monitoring node at least comprises an underwater sound sensor installed underwater on the side wall of the channel, the ultrasonic Doppler flow velocity sensor is installed at the position about 0.3 m underwater on the side wall of the channel, and the pressure type water level gauge is installed at the bottom of the channel and used for measuring the water level. According to the invention, various sensors are used for continuously monitoring the operation state of the open channel for 7 * 24 hours, and a multi-source information fusion acquisition mode is adopted, so that abnormal signals can be detected at the initial stage of clogging, intelligent diagnosis can be carried out, the clogging position can be positioned, early warning can be given out, and the inspection is more timely than manual inspection. And timely early warning and rapid positioning are helpful for the management department to take measures as soon as possible, so that small problems are prevented from being evolved into large dangerous cases, and safe operation of the channel is guaranteed.

Description

TECHNICAL FIELD

[0001] The present application relates to a real-time diagnosis system and method for open channel blockage based on the fusion of voiceprint and flow rate, a water conservancy monitoring system and method, and a system and method for detecting the flow of a water conveying open channel. BACKGROUND

[0002] Open channels (open channels) are widely used in irrigation, drainage and water conservancy scheduling, and their safe operation and efficiency are directly related to agricultural production and flood control and drainage capacity. However, problems such as sediment deposition and debris blockage often occur during long-term operation of open channels, resulting in reduced water passage section, poor water flow, and even channel overflow or flow interruption in severe cases. Traditional open channel blockage detection mainly relies on manual inspection and regular measurement, which is not only time-consuming and labor-intensive, but also difficult to detect problems in a timely manner. In recent years, some automatic monitoring technologies have been applied to open channel monitoring, such as installing pressure water level gauges, ultrasonic water level gauges or radar water level gauges in the channel to monitor water level changes, thereby indirectly determining changes in channel water capacity. However, these methods can only detect overall water level or flow anomalies and cannot accurately identify the location and extent of local blockage.

[0003] In the field of drainage pipes, there are methods of using sonar (ultrasonic) technology to detect underwater deposition. For example, an ultrasonic probe is used to scan the cross-section of a pipe filled with water to obtain the deposition profile below the water surface. This type of sonar detection technology can measure the thickness of the deposition without emptying the water body, and has become an important detection method before pipe dredging. However, sonar detection usually requires the probe to be placed in the water and moved along the pipe, which is difficult to implement for large open channels, and cannot provide continuous real-time monitoring.

[0004] Another related technology is voiceprint recognition (sound recognition), which converts sound signals into electrical signals and extracts features for pattern recognition. In the industrial and security fields, voiceprint recognition has been used for equipment fault diagnosis and abnormal sound detection. For example, in water conservancy engineering, a microphone array is placed along the channel to collect water flow noise and analyze its spectral characteristics to identify abnormal sounds in the channel (such as turbulent flow noise caused by blockage). Similarly, in urban drainage pipe networks, there have been studies using voiceprint technology to detect pipe leaks and blockages: by installing high-sensitivity pickups in manholes, sound signals are transmitted remotely to a server for pattern recognition, enabling automatic detection and alarm of pipe blockage, rupture and other abnormalities. Voiceprint recognition technology has the advantages of real-time and non-contact, but it is sensitive to environmental noise and needs to be combined with other sensor information to improve diagnostic accuracy.

[0005] In addition, Doppler flow velocity measurement technology is widely used in hydrological monitoring. Acoustic Doppler Velocimetry (ADV) is based on the principle of Doppler shift, which detects the movement speed of suspended particles in water through ultrasonic waves to calculate the flow velocity. In open channel flow monitoring, Doppler flow velocity sensors are often combined with water level meters to calculate flow using the velocity-area method. For example, Doppler open channel flow meters measure the frequency shift of scattered acoustic waves by particles in the flow to obtain flow velocity, and combine with built-in water level meters or water level meters arranged in the channel to calculate flow. When the channel is partially blocked, the reduction of flow cross-section will cause abnormal flow velocity distribution (such as local flow velocity increase or vortex), which can be captured by Doppler flow velocity sensors. However, relying solely on flow velocity monitoring also has limitations: for example, changes in channel bed morphology or fluctuations in incoming water can cause flow velocity changes, which need to be combined with other information to accurately determine whether it is caused by blockage.

[0006] In summary, existing open channel blockage detection technologies have their own advantages and disadvantages: manual inspection and traditional instrument monitoring have poor real-time performance or cannot locate the problem; sonar detection can measure accumulation but is difficult to monitor continuously; voiceprint recognition can sense abnormal sounds but is easily disturbed by noise; Doppler flow velocity monitoring can reflect the state of water flow but needs to exclude other interference factors. Therefore, it is necessary to combine various sensing technologies and intelligent algorithms to realize real-time and accurate diagnosis of open channel blockage. SUMMARY

[0007] In order to overcome the problems of the prior art, the present application proposes an open channel blockage real-time diagnosis system and method based on the fusion of voiceprint and flow velocity. The system and method realize real-time monitoring and intelligent diagnosis of open channel blockage through multi-sensor information fusion and artificial intelligence algorithms, can timely find local blockage and locate its position, and provide scientific basis for safe operation of the channel and dredging decision.

[0008] The purpose of the present application is achieved in that: a real-time open channel siltation diagnosis system based on the fusion of voiceprint and flow rate, comprising: a plurality of monitoring nodes are arranged along the open channel, the monitoring nodes are arranged at intervals or selectively according to the length of the channel, the terrain and the siltation section, each monitoring node at least comprises: an underwater acoustic sensor for collecting underwater noise signals, the underwater acoustic sensor is installed underwater on the side wall of the channel, for obtaining sound characteristics related to the interaction of water flow and siltation; an ultrasonic Doppler flow rate sensor, the ultrasonic Doppler flow rate sensor is installed about 0.3 meters underwater on the side wall of the channel, the beam is directed to the center of the channel to measure the flow rate distribution of the section; a pressure type water level gauge for measuring water level, the pressure type water level gauge is installed at the bottom of the channel, each node is also provided with a solar power supply module and a communication module, each monitoring node is connected with a cloud server by wire or wireless, the cloud server is connected with at least one client, the cloud server is provided with a data receiving and storage unit, a signal processing unit, an AI diagnosis model unit and a warning issuing unit.

[0009] A real-time open channel siltation diagnosis method based on the fusion of voiceprint and flow rate using the above diagnosis system, the method comprises the following steps:

[0010] Step 1, data acquisition: the underwater acoustic sensor collects underwater sound signals in real time, the Doppler flow rate sensor measures the water flow rate distribution in real time, the pressure sensor measures the water level or the change of water bottom pressure, the data of each sensor is sent to the cloud server in real time through the wireless communication transmission module, and the data collected by each monitoring node is uploaded to the cloud server every 5 seconds;

[0011] Step 2, feature extraction: the cloud server respectively pre-processes and extracts features from the received sound signals and flow rate signals; for sound signals, the frequency spectrum features are extracted by using fast Fourier transform or wavelet transform, including sound pressure level distribution of different frequency bands and main frequency components; for flow rate signals, the flow rate values and their time change rates of each measuring point, as well as the shape features of the flow rate profile, are extracted, and the pressure sensor data extracts the abnormal fluctuation features of water level or pressure;

[0012] Step 3, fusion diagnosis: the extracted sound frequency spectrum features and flow rate features are input into the trained AI diagnosis model, and the model is used to evaluate and diagnose the current siltation blockage state of the open channel; the AI diagnosis model is constructed by using machine learning or deep learning algorithm, and support vector machine, random forest, neural network or convolutional neural network; the model realizes automatic identification of siltation by learning the differences between sound and flow rate features in normal state and siltation state in historical data;

[0013] Step 4, early warning and positioning: Once the AI diagnosis model determines that there is a blockage in a certain monitoring node area, the diagnosis system immediately generates an early warning message. The early warning message includes the time of the occurrence of the blockage, the estimated location, and the severity level. The early warning message is sent to relevant management personnel through SMS, mobile applications, or the monitoring center interface to remind them to take timely dredging measures. At the same time, these monitoring data and diagnosis results are stored as historical data in the database for subsequent analysis and model optimization and training.

[0014] The advantages and benefits of the present application are: the present application uses various sensors to monitor the operation state of open channels 7x24 hours without interruption, and adopts a multi-source information fusion collection method, which can detect abnormal signals at the early stage of blockage, perform intelligent diagnosis, and locate the blockage position, thereby issuing an early warning, which is more timely than manual inspection. Timely warning and rapid positioning help the management department to take measures early to prevent small problems from becoming big risks and ensure the safe operation of the channel.

[0015] Multi-source information fusion, accurate and reliable diagnosis: The data of multiple sensors such as sound and flow rate (pressure if necessary) are fused, and AI model is used for comprehensive analysis to improve the accuracy and robustness of diagnosis. A single sensor may have false positives or false negatives, but the present application can effectively eliminate environmental noise, flow fluctuations, and other disturbances through information complementation to ensure the reliability of the diagnosis results.

[0016] Positioning the blockage location: By arranging multiple monitoring nodes along the channel, the location of the blockage can be determined. When a node detects an abnormal blockage, the node location can be notified to the management personnel as the approximate location of the blockage occurrence, facilitating rapid processing. Compared with traditional methods that can only give overall abnormalities, the present application can provide more specific fault location information.

[0017] Intelligent diagnosis, reducing manual intervention: The system uses trained AI models to automatically analyze data and achieve intelligent diagnosis. There is no need for real-time human monitoring of data, reducing labor costs. At the same time, the AI model can be continuously learned and optimized, and the diagnosis accuracy improves with the accumulation of historical data, reflecting the development direction of smart water conservancy.

[0018] Wide range of applications: The present application is suitable for various types of open channels, including irrigation channels, drainage ditches, urban river channels, etc. As long as there is water flow, blockage monitoring can be achieved by arranging appropriate sensors. In addition, the system is flexible and can be expanded to add sensor types (such as water quality sensors, video cameras, etc.) to enrich the monitoring content.

[0019] The open channel siltation real-time diagnosis method and system proposed by the present application conforms to the development trend of smart water conservancy and Internet of Things technology, and has broad application prospects. First, in large irrigation areas and water diversion projects, the invention can be used to monitor the main channel and branch channel comprehensively, discover siltation hazards in time, and ensure the continuity and uniformity of irrigation water supply. Second, in the field of urban flood control and drainage, the smoothness of open channels and rivers is related to urban safety, and the invention can be used to monitor urban flood drainage channels, city moats, etc., to prevent garbage and silt from causing waterlogging. Third, the technical idea of the invention can also be extended to pipeline water delivery systems, such as siltation and blockage monitoring of long-distance water delivery pipelines, by adjusting the sensor to a form suitable for pipelines.

[0020] With the further development of sensor technology and artificial intelligence, future open channel monitoring systems will be more intelligent and integrated. For example, high-definition cameras and unmanned aerial vehicle inspection can be introduced as a supplement, combined with the data of the system, to realize an "air-space-ground" integrated monitoring system. At the same time, combined with digital twin technology, the channel water flow and siltation conditions can be simulated in real time in a virtual environment, verified with actual measurement data, and the accuracy of prediction can be improved. The present application provides an effective means for the intelligent operation and maintenance of open channels, and is expected to play an important role in water conservancy engineering management, and promote the transformation and upgrading of traditional water conservancy to smart water conservancy. BRIEF DESCRIPTION OF DRAWINGS

[0021] The present application will be further described below in conjunction with the drawings and examples.

[0022] Figure 1 is the principle block diagram of the diagnosis system used by the method described in the embodiment of the present application;

[0023] Figure 2 is a flowchart of the method described in the embodiment of the present application. DETAILED DESCRIPTION

[0024] Example 1:

[0025] This embodiment is an open channel siltation real-time diagnosis system based on voiceprint recognition and Doppler fusion, as shown in Figure 1As shown, it comprises: arranging a plurality of monitoring nodes along the open channel, the monitoring nodes are arranged according to the length of the channel, the terrain and the situation of the easily silted section, each monitoring node at least comprises: an underwater acoustic sensor (such as a waterproof microphone or a hydrophone) for collecting underwater noise signals, the underwater acoustic sensor is installed underwater on the side wall of the channel, for obtaining the sound characteristics related to the interaction of water flow and silt; an ultrasonic Doppler flow sensor, the ultrasonic Doppler flow sensor is installed about 0.3 meters underwater on the side wall of the channel, the beam is directed to the center of the channel to measure the cross-sectional flow velocity distribution; a pressure type water level gauge for measuring water level, the pressure type water level gauge is installed at the bottom of the channel, each node is also provided with a solar power supply module and a communication module, each monitoring node is connected with a cloud server by wire or wireless, the cloud server is connected with at least one client, the cloud server is provided with a data receiving and storage unit, a signal processing unit, an AI diagnosis model unit and a warning issuing unit.

[0026] In the diagnostic system, the monitoring nodes can be arranged according to the length and importance of the open channel, and generally arranged densely at the key sections or easily silted parts of the channel (such as bends and slope changes). For example, a monitoring node is set every certain distance (such as 50 to 200 meters) to realize the coverage monitoring of the whole channel section. When an abnormal siltation is detected by a node, the data of the adjacent nodes can be used for further verification to accurately locate the siltation area.

[0027] The diagnostic system specifically comprises:

[0028] Monitoring node: arranged along the open channel, each node comprises an underwater acoustic sensor, a Doppler flow velocity sensor, and a pressure sensor. These sensors are responsible for real-time collection of water flow sound, flow velocity and pressure data. The underwater acoustic sensor is preferably a waterproof microphone or a hydrophone, which can sensitively capture underwater and water surface sound signals; the Doppler flow velocity sensor can be an ultrasonic Doppler flow velocity profiler (ADCP) or a point ADV, which is used to measure the cross-sectional flow velocity distribution; the pressure sensor is used to measure the water level or water pressure, and a submerged pressure water level gauge can be used.

[0029] Data transmission module: installed in each monitoring node, which transmits the data collected by the sensors to the cloud server in real time through a wireless communication network (such as 4G / 5G, NB-IoT or LoRa, etc.). The module includes a microcontroller, a wireless communication module and a power management unit to ensure long-term stable operation in the field environment.

[0030] Cloud server: deployed on the cloud platform, including data receiving and storage unit, signal processing unit, AI diagnosis model unit and early warning release unit. The data receiving unit receives the data uploaded by each monitoring node and stores it in the database; the signal processing unit preprocesses and extracts features from the sound and flow rate signals; the AI diagnosis model unit runs the trained intelligent diagnosis model to analyze the multi-source features and determine whether there is a blockage and the severity; the early warning release unit generates an early warning notification based on the model output and pushes it to the relevant personnel.

[0031] User terminal: including the management terminal of the monitoring center and the mobile terminal of the management personnel. The management terminal can be a monitoring computer installed in the water conservancy bureau or irrigation management department, which is used to display the monitoring data, diagnosis results and historical trends of each open channel in real time; the mobile terminal can be a mobile phone or tablet, which is used to receive early warning information such as SMS and App push, so that the management personnel can keep abreast of the channel status at any time and anywhere.

[0032] Application example:

[0033] Select a section of an irrigation open channel, deploy the real-time diagnosis system described in this embodiment on the section of the channel. The channel is about 2 kilometers long, 3 meters wide and 1.5 meters deep. A monitoring node is set up every 100 meters along the channel, a total of 20 nodes. Each node includes: a waterproof microphone (frequency response range 20Hz-20kHz, sensitivity -30dB), a water sound sensor installed underwater on the side wall of the channel, used to obtain sound features related to water flow and sediment interaction. An ultrasonic Doppler flow rate sensor (measurement range 0-5m / s, accuracy ±0.01m / s), installed underwater about 0.3 meters from the channel side wall, with a beam pointing to the center of the channel to measure the cross-sectional flow rate distribution; a pressure type water level gauge (range 0-5m, accuracy ±0.1%FS), installed at the bottom of the channel, used to measure the water level. Each node is also equipped with a solar power supply module (including a battery and a solar panel) and a 4G communication module, which uploads the collected data to the cloud server every 5 seconds.

[0034] Example two:

[0035] This embodiment is a real-time diagnosis method for open channel blockage based on the fusion of voiceprint and flow rate using the system described in example one. The steps of the method are as follows, and the flow is shown in Figure 2

[0036] Step 1, data acquisition: the water sound sensor collects underwater sound signals in real time, the Doppler flow rate sensor measures water flow velocity distribution in real time, and the pressure sensor measures water level or water bottom pressure change. The data of each sensor is transmitted to the cloud server in real time through the wireless communication transmission module. Each monitoring node uploads the collected data to the cloud server every 5 seconds. ​

[0037] Step 2, Feature Extraction: The cloud server pre-processes and extracts features from the received sound signals and flow rate signals. For sound signals, Fast Fourier Transform (FFT) or wavelet transform is used to extract spectral features, including sound pressure level distribution in different frequency bands, dominant frequency components, etc. For flow rate signals, the flow rate values at each measurement point and their rate of change over time, as well as the shape features of the flow rate profile (such as maximum flow rate position, flow rate gradient, etc.) are extracted. Pressure sensor data extracts abnormal fluctuation features of water level or pressure.

[0038] The cloud server processes the received sound and flow rate data. For sound signals, the power spectral density is calculated using FFT with Hamming windowing, and the sound pressure level of each 1 / 3 octave frequency band is extracted as a feature. The total sound pressure level and the peak frequency of the dominant frequency of the sound signal are also calculated to represent the intensity and tone changes of the water flow noise. For flow rate signals, the flow rate profile returned by the Doppler sensor is first smoothed, then the flow rate value at each measurement point, as well as the average flow rate, maximum flow rate and its position, etc. are extracted. In addition, the change amount and rate of flow rate between adjacent time periods are calculated to capture the mutation characteristics of flow rate. Pressure level data is used to calculate water level height and water level change rate. All these features are stored in time series and combined into a multi-dimensional feature vector for AI model analysis.

[0039] Step 3, Fusion Diagnosis: The extracted sound spectrum features and flow rate features are input into the trained AI diagnosis model, which evaluates and diagnoses the current siltation and blockage state of the open channel. The AI diagnosis model can be built using machine learning or deep learning algorithms, such as Support Vector Machine (SVM), Random Forest, Neural Network or Convolutional Neural Network (CNN), etc. The model learns the differences between normal and siltation states in historical data through sound and flow rate features, achieving automatic recognition of siltation. Specifically, when siltation occurs near a monitoring node, the water flow is blocked, producing abnormal turbulent noise, the spectral features of which (such as increased high-frequency components, changes in spectral energy distribution, etc.) are captured by the voiceprint recognition module; at the same time, the flow rate distribution at this location also changes (such as a significant increase in local flow rate or the presence of reverse flow), which is detected by the Doppler module. The AI model integrates these two abnormal signals to determine the occurrence of siltation and estimate its severity. If the model determines that there is a siltation risk, it proceeds to the next step of the warning process.

[0040] Learning Process: The model takes the pre-processed feature vector as input and outputs a judgment of the channel siltation state (such as normal, mild siltation, severe siltation). The following describes two typical machine learning models and their learning processes in detail.

[0041] 1. Support Vector Machine (SVM) Model Learning Process

[0042] Support Vector Machine (SVM) is a powerful binary or multi-class supervised learning model, whose core idea is to find an optimal hyperplane in the feature space to maximize the margin between different classes.

[0043] For the clogging diagnosis problem of the present application, first, the multi-dimensional feature vector of each time point

[0044] is regarded as a point in the feature space, where is the voiceprint feature, is the flow rate feature, is the pressure feature. At the same time, a label

[0045] is assigned to each sample point, for example, -1 represents "normal" and +1 represents "clogging".

[0046] The learning process of SVM is to solve the following optimization problem to find the hyperplane :

[0047]

[0048] The constraint condition is:

[0049] where: is the normal vector of the hyperplane, is the bias term, is the slack variable (allowing some samples to be misclassified), and C is the penalty coefficient, used to balance the maximization of the margin and the minimization of classification errors.

[0050] The learning process is as follows:

[0051] Data preparation: Collect a large amount of historical data with labels (feature vectors under normal and different degrees of clogging state) to form a training set .

[0052] Kernel function selection: Since the relationship between voiceprint and flow rate features and clogging state may be nonlinear, a kernel function (such as radial basis function RBF kernel: is usually selected to map the data to a higher dimensional space, making it linearly separable.

[0053] Model training: Solve the above optimization problem by algorithms such as Sequential Minimal Optimization (SMO) to get the optimal and . This process is essentially to find those "support vectors" that play a decisive role in defining the classification boundary.

[0054] Model application: After training, for a new unlabeled feature vector , the label

[0055] The value is used to determine its category, wherein is the mapping function value. If it is +1, it is diagnosed as "blockage"; if it is -1, it is "normal". For multi-classification problems (such as normal, mild, severe), a "one-to-one" or "one-to-many" strategy can be used to combine multiple SVM classifiers.

[0056] 2. Convolutional Neural Network (CNN) model learning process

[0057] Convolutional Neural Networks are particularly suitable for processing data with spatial or temporal structure. In this invention, the features from the acoustic print (spectrum graph) and flow rate (profile or time series) can be combined into a two-dimensional "feature map" as the input of the CNN.

[0058] The learning process of CNN automatically learns the deep patterns in the data through multiple levels of feature extraction and abstraction. A typical CNN model includes convolutional layers, pooling layers, and fully connected layers.

[0059] Data preparation and input: the acoustic print spectrum features (such as 's matrix, with the number of frequency bands and flow rate features (such as 's matrix, with the number of flow rate measurement points

[0060] Convolutional Layer: This layer uses multiple convolution kernels (filters) to slide on the input feature map and perform convolution operations. Each convolution kernel is responsible for extracting a local feature, such as a specific peak in the spectrum, rapid changes in flow rate, etc. The learning process is through the backpropagation algorithm, constantly adjusting the weights of the convolution kernel, so that it can extract the most useful local patterns to distinguish "blockage" and "normal".

[0061] Pooling Layer: The feature map output by the convolution layer is down-sampled to reduce the data dimension, reduce the amount of calculation, and at the same time preserve the most significant features and give the model certain translation invariance. For example, MaxPooling selects the maximum value in the receptive field as the output.

[0062] Fully Connected Layer: After multiple convolution and pooling layers, the high-level feature maps are flattened into a one-dimensional vector and input into a fully connected layer. The fully connected layer is similar to a traditional neural network, where features are weighted and combined. Finally, a Softmax activation function is applied to output the probability distribution over each congestion level (e.g., normal, mild, severe) .

[0063] The learning process of a CNN is end-to-end: given labeled training data, the model's predicted output is compared to the true labels using a loss function (e.g., cross-entropy loss) and the parameters (weights and biases) are updated using gradient descent and backpropagation algorithms. The learning process mainly includes the following steps:

[0064] 1. Forward Propagation: Input sample X is passed through the network layer by layer, and the output of each layer is calculated until the final predicted probability distribution is obtained .

[0065] 2. Calculate Loss: Use a loss function (e.g., cross-entropy loss) to measure the difference between the predicted value ŷ and the true label y. For a single sample, the cross-entropy loss function is:

[0066]

[0067] where: is the total number of classes, iterates over all classes.

[0068] 3. Backpropagation: Calculate the gradient of the loss function L with respect to all parameters (weights W and biases b) in the network and . According to the chain rule, starting from the output layer.

[0069] 4. Parameter Update: Use an optimization algorithm (e.g., stochastic gradient descent SGD or Adam) to update the network parameters according to the calculated gradients to reduce the loss. The update rule is as follows:

[0070]

[0071]

[0072] where is the learning rate, which controls the step size of each update.

[0073] By iterating the above steps on the entire training set, all parameters of the network (convolution kernel weights, fully connected layer weights, etc.) will be automatically optimized, allowing the model to learn effective features from raw sensor data to distinguish different congestion states, achieving accurate classification of congestion states. ​

[0074] The training of the diagnostic model relies on a high-quality labeled dataset. This dataset needs to cover the data of the target open channel under various typical working conditions, including but not limited to: normal free-flow state, different stages of siltation state, and state before and after dredging, and is associated with manually confirmed labels.

[0075] In feature processing, the application adopts an early fusion strategy of feature layers, that is, multiple source features such as voiceprints and flow rates are spliced into a unified high-dimensional feature vector before being input into the diagnostic model. A late fusion strategy based on a double-flow network can also be adopted, that is, independent feature extraction sub-networks are designed for data of different modalities, and then feature fusion and joint decision are made in a high-level network.

[0076] The AI diagnostic model is specifically constructed as follows:

[0077] A convolutional neural network (CNN) is used to construct the open channel siltation diagnostic model. The input of the model is the fused feature vector, including the sound spectrum features and flow rate features (and pressure features if necessary) of each monitoring node at a certain time. To utilize the time dimension information, the input can also use the feature sequence of the last several time steps. The output of the model is the siltation probability or siltation level (for example, 0 level represents normal, 1 level represents slight siltation, and 2 level represents severe siltation). During the model training phase, historical data of the channel under different dredging states are collected: including sound and flow rate features under the following scenarios: shortly after dredging (normal state), after running for one month (light siltation), after running for three months (heavy siltation), etc., and the corresponding siltation level marked by artificial is used as the supervision signal. Through training of these data, the CNN model learns the differences in feature patterns between normal and siltation states. To prevent overfitting, data augmentation (such as adding random noise to the spectrum features) and cross-validation techniques are used during the training process. The final model achieves a classification accuracy of more than 90% on the test set, and can reliably distinguish between normal and siltation states, and reasonably evaluate the siltation level.

[0078] The AI diagnostic model is obtained by training historical data. The training data includes sound and flow rate feature samples collected under different siltation working conditions, and the corresponding siltation level is labeled. Through supervised learning, the model learns the mapping relationship between sound spectrum and flow rate distribution and siltation state. To improve the generalization ability of the model, ensemble learning or deep learning architecture can be used for joint feature extraction and classification of multi-sensor data.

[0079] The diagnosis can also be combined with pressure sensing information to enhance the reliability of the diagnosis. Specifically, miniature pressure sensors are installed at key positions at the bottom of the channel to monitor the water level or water pressure changes in real time. When a local blockage occurs, the upstream water level may rise and the downstream water level may decrease, forming a local water level difference or pressure difference. These pressure anomaly characteristics can be used as supplementary input to the AI model together with the voiceprint and Doppler information, thereby further improving the accuracy of the blockage detection.

[0080] Under normal operating conditions, the sound spectrum and flow rate distribution of each monitoring node are relatively stable, and the AI model output is "normal" or level 0. Once a blockage occurs near a node, for example, due to an increase in upstream sediment forming a sand dam downstream of the node, causing the local water section to narrow, the system will react as follows: First, the water sound sensor at the node detects that the water flow noise has significantly increased, with an increase in high-frequency components in the frequency spectrum (this is due to the generation of turbulent flow and cavitation noise by the water flow through the narrow section); at the same time, the Doppler flow rate sensor measures that the average flow rate and maximum flow rate of the section have significantly increased, exceeding the historical same period level; the pressure water level gauge shows that the water level on the upstream side of the node has risen slightly. These abnormal characteristics are transmitted to the cloud in real time and input to the AI model. After analysis by the model, it is determined that the probability of a blockage occurring in the region of the node is high, and the blockage level is output as level 2 (severe blockage). The system immediately sends an early warning to the mobile phone of the channel manager through a short message, and the manager receives the warning and immediately goes to the scene to check that a large amount of sediment has indeed accumulated, causing the water flow to be obstructed. Subsequently, the management department organizes a dredging operation to remove the blockage, and the channel resumes normal water flow.

[0081] Step 4, warning and positioning: once the AI diagnosis model determines that there is a blockage in the region of a monitoring node, the system immediately generates a warning message. The warning message includes the time of the occurrence of the blockage, the estimated location (determined by the location of the monitoring node), and the severity level. The warning message is sent to the relevant management personnel through a short message, a mobile application, or a monitoring center interface, reminding them to take timely dredging measures. The system can also store historical monitoring data and diagnosis results in a database for subsequent analysis and model optimization.

[0082] Finally, it should be noted that the above is only used to illustrate the technical solutions of the present application and is not limiting. Although the present application has been described in detail with reference to the preferred arrangement, those skilled in the art should understand that the technical solutions of the present application (such as the form of the channel being detected, the use of various models and formulas, the order of the steps, etc.) can be modified or replaced equivalently without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. A real-time diagnosis system for open channel siltation based on the fusion of voiceprint and flow velocity, comprising: A section of several kilometers of irrigation canal channel, a monitoring node is set every 100 meters along the channel, at least 20 monitoring nodes are arranged, each monitoring node at least includes: a waterproof microphone for collecting water flow noise, the waterproof microphone is installed about 0.5 meters above the water surface of the channel side wall; an ultrasonic Doppler flow velocity sensor, the ultrasonic Doppler flow velocity sensor is installed about 0.3 meters underwater on the channel side wall, the beam is directed to the center of the channel to measure the cross-sectional flow velocity distribution; a pressure type water level gauge for measuring water level, the pressure type water level gauge is installed at the bottom of the channel, each node is also provided with a solar power supply module and a communication module, each monitoring node is connected with a cloud server by wire or wireless, the cloud server is connected with at least one client, the cloud server is provided with a data receiving and storage unit, a signal processing unit, an AI diagnosis model unit and an early warning release unit.

2. A real-time diagnosis method for open channel siltation based on the fusion of voiceprints and flow velocity using the system of claim 1, characterized in that, The method comprises the following steps: Step 1, data acquisition: the underwater sound signal is collected by the underwater sound sensor in real time, the flow velocity distribution is measured by the Doppler flow velocity sensor in real time, the water level or the change of the water bottom pressure is measured by the pressure sensor, and the data of each sensor is sent to the cloud server in real time through the wireless communication transmission module, and the data collected by each monitoring node is uploaded to the cloud server every 5 seconds; Step 2, feature extraction: the cloud server respectively pre-processes and extracts features from the received sound signal and flow velocity signal; for the sound signal, the frequency spectrum features are extracted by using fast Fourier transform or wavelet transform, including the sound pressure level distribution of different frequency bands and the main frequency component; for the flow velocity signal, the flow velocity value and its change rate with time of each measuring point, and the shape feature of the flow velocity profile are extracted, and the abnormal fluctuation features of the water level or the pressure are extracted from the pressure sensor data; Step 3, fusion diagnosis: the extracted sound frequency spectrum features and flow velocity features are input into the trained AI diagnosis model, and the current siltation and blockage state of the open channel is evaluated and diagnosed by the model; the AI diagnosis model is constructed by using machine learning or deep learning algorithm, and support vector machine, random forest, neural network or convolutional neural network; the model realizes automatic identification of siltation by learning the differences between sound and flow velocity features in normal state and siltation state in historical data; Step 4, early warning and positioning: once the AI diagnosis model determines that there is siltation in the area of a monitoring node, the diagnosis system immediately generates early warning information; the early warning information includes the time of siltation occurrence, the estimated position and the severity level, the early warning information is notified to the relevant management personnel through short message, mobile application or monitoring center interface, reminding to take dredging measures in time, and the monitoring data and diagnosis results are stored in the database as historical data for subsequent analysis and model optimization and training.

Images