Water meter flow real-time correction method and system based on intelligent sensor

Abstract

The present invention relates to the field of intelligent sensor application technology, and discloses a real-time correction method and system for water meter flow based on intelligent sensors. The system includes a flow signal acquisition module, a signal feature decomposition module, a frequency domain feature fusion module, an error compensation decision module, and a correction instruction execution module. The method collects multi-dimensional vibration waveform data of the water meter pipeline through a multi-band sensor, obtains a local frequency domain feature tensor set through time-frequency domain decomposition and energy spectrum feature extraction, and then obtains a local optimized feature tensor set through dynamic noise suppression and frequency band correlation optimization. Global feature reconstruction is performed based on frequency domain energy distribution to generate an error compensation decision feature map, and finally a real-time correction control signal is generated. The present invention improves the accuracy and dynamic adaptability of flow measurement in complex pipe network environments through multi-dimensional signal acquisition and intelligent feature processing. It is suitable for high-precision flow correction of intelligent water meters and has the advantages of strong anti-interference ability and accurate compensation.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent sensor application, and in particular to a method and system for real-time correction of water meter flow based on intelligent sensors. Background Art

[0002] With the advancement of smart city construction, smart water meters, as core equipment for water resource measurement, have a direct impact on the level of refined water resource management due to their measurement accuracy and stability. Traditional water meter flow measurement mainly relies on mechanical structures or single sensors to collect signals. In complex pipe network environments (such as water flow pulsation, pipe vibration, fluid viscosity changes, etc.), they are easily affected by noise interference and multi-factor coupling, resulting in large measurement errors. Specifically, the existing technology has the following prominent problems:

[0003] Traditional signal acquisition methods have significant limitations. Single-frequency band or single-type sensors struggle to fully capture the complex fluid dynamics within pipelines. For example, the vibrations generated by water flowing through valves and elbows contain multi-band energy distribution. A single sensor can only capture signal data in a localized dimension and cannot reflect the complete spectral characteristics of the flow state, resulting in a lack of comprehensive information support for subsequent error compensation.

[0004] Signal processing methods lack robustness. Traditional time-frequency domain analysis methods (such as Fourier transform and wavelet transform) struggle to dynamically suppress interference from ambient noise and inherent pipeline vibration when processing non-stationary signals. For example, when a brief surge of flow occurs in a pipeline network, the noise energy can mask the frequency domain characteristics of the actual flow signal, leading to feature extraction errors and, in turn, affecting the accuracy of error compensation.

[0005] Frequency-domain feature fusion mechanisms suffer from significant drawbacks. Existing technologies typically employ fixed-weight feature fusion strategies, which are unable to adapt to changes in pipeline network operating conditions. For example, under different flow conditions (e.g., low-flow trickle flow versus high-flow turbulence), the energy correlations between frequency bands vary significantly. Fixed weights cannot dynamically optimize the coupling relationships between frequency bands, resulting in the fused features being unable to accurately represent actual flow conditions.

[0006] Error compensation strategies lack intelligence. Traditional methods often rely on empirical formulas or simple linear regression models for error correction, lacking in-depth understanding of frequency-domain energy distribution patterns. For example, when scaling on the inner wall of a pipeline causes changes in the fluid boundary layer, the frequency-domain energy distribution of the flow signal undergoes nonlinear changes. Traditional models struggle to capture these complex changes, leading to delayed or inaccurate compensation decisions.

[0007] Finally, real-time performance and dynamic adaptability are insufficient. Existing systems are unable to quickly complete the entire process of signal acquisition, feature analysis, and compensation command generation when faced with sudden flow fluctuations or changes in pipe network topology. For example, during the flow recovery phase following emergency repairs of a water supply pipe burst, traditional systems can take a long time to adjust to accurate metering, impacting real-time monitoring and scheduling of water resources. Summary of the Invention

[0008] The purpose of the present invention is to provide a real-time correction method and system for water meter flow based on an intelligent sensor to solve the problems raised in the above background technology.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a real-time water meter flow correction system based on an intelligent sensor, the system comprising:

[0010] Flow signal acquisition module, used to obtain multi-dimensional vibration waveform data of water meter pipelines through multi-band sensors;

[0011] a signal feature decomposition module, configured to perform time-frequency domain decomposition and energy spectrum feature extraction on the multi-dimensional vibration waveform data to obtain a set of local frequency domain feature tensors of the flow state;

[0012] a frequency domain feature fusion module, configured to perform dynamic noise suppression and frequency band correlation optimization on the set of local frequency domain feature tensors of the flow state to obtain a set of local optimized feature tensors of the flow state;

[0013] An error compensation decision module is used to perform global feature reconstruction based on frequency domain energy distribution on the set of local optimized feature tensors of the flow state to generate a flow error compensation decision feature map;

[0014] The correction instruction execution module is used to generate a real-time flow correction control signal according to the flow error compensation decision characteristic diagram.

[0015] Preferably, the signal feature decomposition module includes:

[0016] a waveform data segmentation unit, configured to dynamically divide the multi-dimensional vibration waveform data according to a preset time window to obtain a set of waveform data segments;

[0017] The feature extraction unit is used to input each segment of the waveform data segment set into a frequency domain encoder based on a multi-scale decomposition network to obtain a set of local frequency domain feature tensors of the flow state.

[0018] Preferably, the frequency domain feature fusion module includes:

[0019] A frequency domain tensor expansion unit, configured to perform matrix expansion on the flow state local frequency domain feature tensor along a frequency band dimension to obtain a set of flow state local frequency domain feature vectors;

[0020] a frequency band energy correlation calculation unit, configured to calculate a frequency domain correlation coefficient between any two eigenvectors in the set of local frequency domain eigenvectors of the flow state to generate a frequency band energy correlation matrix;

[0021] a dynamic noise suppression unit, configured to perform weight correction on the frequency band energy correlation matrix according to the frequency domain differences of adjacent time windows in the set of local frequency domain feature vectors of the flow state to obtain a frequency band energy constraint correlation matrix;

[0022] The feature optimization unit is used to perform cross-band convolution fusion on the frequency band energy constraint association matrix and the set of local frequency domain feature vectors of the flow state to obtain the local optimized feature tensor of the flow state.

[0023] Preferably, the frequency band energy correlation calculation unit includes:

[0024] A frequency domain projection subunit, configured to map each feature vector in the set of local frequency domain feature vectors of the flow state to an orthogonal frequency domain space to obtain a set of projected frequency domain feature vectors;

[0025] The frequency domain correlation analysis subunit is used to calculate the spectrum energy similarity between any two projected feature vectors in the set of projected frequency domain feature vectors to generate the frequency band energy correlation matrix composed of multiple frequency band energy correlation values.

[0026] Preferably, the feature optimization unit is specifically implemented as follows:

[0027] Performing a dilated convolution process on the frequency band energy constraint association matrix to obtain a frequency band energy constraint feature matrix;

[0028] Inputting the set of local frequency domain feature vectors of the traffic state and the frequency band energy constraint feature matrix into a bidirectional recursive coding network to obtain a set of local context feature vectors of the traffic state;

[0029] Matrix reconstruction is performed on the set of the traffic state local context feature vectors to obtain the traffic state local optimized feature tensor.

[0030] Preferably, the error compensation decision module includes:

[0031] A local feature compression unit, configured to perform mean pooling processing in the frequency domain on each tensor in the set of the flow state local optimization feature tensors to obtain a set of flow state local optimization feature vectors;

[0032] A frequency domain energy entropy calculation unit, configured to calculate the energy entropy value of each eigenvector in the set of local optimized eigenvectors of the flow state to generate a flow dynamic energy entropy set;

[0033] A reference frequency band determination unit, configured to select a flow state local optimization feature vector corresponding to a minimum energy entropy value in the flow dynamic energy entropy set as an initial compensation reference vector;

[0034] a dynamic compensation weight calculation unit, configured to calculate the dynamic compensation weight of each eigenvector in the set of the flow state local optimization eigenvectors and the initial compensation reference vector according to the frequency domain distance between each eigenvector and the initial compensation reference vector and the energy entropy value of each eigenvector to generate a dynamic compensation weight set;

[0035] A global reconstruction unit is used to perform weighted superposition on the set of local optimization feature vectors of the flow state using the dynamic compensation weight set to generate the flow error compensation decision feature map.

[0036] Preferably, the frequency domain energy entropy calculation unit is specifically implemented as follows:

[0037] Calculating the median vector and the range vector of the local optimized characteristic vector of the flow state;

[0038] Performing element-by-element difference calculation on the local optimized feature vector of the flow state and the median vector, and performing a cubic power operation on the difference result to obtain a flow feature energy difference vector;

[0039] Calculate the overall mean of the flow characteristic energy difference vector; perform a ratio operation on the mean and the cube value of the extreme difference vector, and input the result into a normalization function to obtain the energy entropy value.

[0040] Preferably, the dynamic compensation weight calculation unit is specifically implemented as follows:

[0041] Multiplying the energy entropy value of the local optimization characteristic vector of the flow state and the energy entropy value of the initial compensation reference vector by a first compensation coefficient to obtain a first dynamic compensation factor;

[0042] Multiplying the absolute value of the Manhattan distance between the local optimized characteristic vector of the flow state and the initial compensation reference vector by a second compensation coefficient to obtain a second dynamic compensation factor;

[0043] The first dynamic compensation factor is multiplied by the second dynamic compensation factor to obtain the dynamic compensation weight.

[0044] Preferably, the correction instruction execution module is specifically implemented as follows:

[0045] The flow error compensation decision feature map is input into a correction signal generator based on random forest to obtain the real-time flow correction control signal, which is used to indicate the calibration strategy of the flow metering device.

[0046] Preferably, the present invention also includes a real-time correction method for water meter flow based on an intelligent sensor, the method comprising:

[0047] Collect multi-dimensional vibration waveform data of water meter pipes through multi-band sensors;

[0048] Performing time-frequency domain decomposition and energy spectrum feature extraction on the multi-dimensional vibration waveform data to obtain a set of flow state local frequency domain feature tensors;

[0049] Performing dynamic noise suppression and frequency band correlation optimization on the set of local frequency domain feature tensors of the flow state to obtain a set of local optimized feature tensors of the flow state;

[0050] Performing global feature reconstruction based on frequency domain energy distribution on the set of local optimized feature tensors of the flow state to generate a flow error compensation decision feature map;

[0051] A real-time flow correction control signal is generated according to the flow error compensation decision characteristic diagram.

[0052] Compared with the prior art, the present invention has the following beneficial effects:

[0053] In terms of signal acquisition and feature processing, the application of multi-band sensors breaks through the limitations of traditional single sensors and can fully capture the multi-dimensional vibration signals generated by fluid movement in the pipeline, covering the energy distribution characteristics of different frequency bands. Through time-frequency domain decomposition and energy spectrum feature extraction, the system can convert raw vibration data into a collection of local frequency domain feature tensors of the flow state, achieving a refined characterization of the flow signal and providing rich feature information for subsequent error compensation. Compared with traditional single-dimensional signal processing methods, this method can more accurately reveal the inherent correlation between flow changes and vibration characteristics, effectively improving the comprehensiveness and accuracy of feature extraction.

[0054] In the frequency domain feature fusion link, dynamic noise suppression and frequency band correlation optimization technology achieves adaptive suppression of noise components and dynamic optimization of the coupling relationship between frequency bands by constructing a frequency band energy correlation matrix and performing weight correction. Specifically, by calculating the frequency domain correlation coefficient between feature vectors and adjusting the weights based on the frequency domain differences of adjacent time windows, the system can filter the interference of environmental noise and inherent vibration of the pipeline in real time, while strengthening the energy correlation between effective frequency bands. This data-driven feature fusion mechanism enables the fused local optimized feature tensor of the flow state to more realistically reflect the actual flow state, significantly improving the robustness and anti-interference ability of the features.

[0055] In the error compensation decision-making process, the global feature reconstruction technology based on frequency domain energy distribution realizes intelligent analysis and precise compensation of flow errors by introducing key parameters such as energy entropy value and dynamic compensation weight. By calculating the energy entropy value of each eigenvector to determine the initial compensation reference vector, and dynamically adjusting the compensation weight according to the frequency domain distance and energy entropy value, the system can adapt to the changes in frequency domain energy distribution under different flow conditions, avoiding the limitations of traditional fixed weight compensation strategies. This compensation decision-making mechanism based on global feature reconstruction can more accurately capture the nonlinear change pattern of flow signals, thereby generating a flow error compensation decision feature map that better meets actual needs, significantly improving the accuracy and dynamic adaptability of error compensation.

[0056] In terms of system real-time performance and engineering application, the correction instruction execution module generates real-time flow correction control signals by introducing a random forest algorithm, achieving a rapid response from feature analysis to compensation instruction generation. This algorithm possesses efficient nonlinear modeling capabilities and parallel computing characteristics, enabling it to complete feature map processing and control signal generation in a short period of time, meeting the strict real-time requirements of smart water meters. Furthermore, the entire system utilizes a modular design, allowing each functional module (such as signal acquisition, feature decomposition, feature fusion, error compensation, and instruction execution) to be independently optimized and upgraded, facilitating engineering deployment and maintenance. The system is highly versatile and scalable, making it suitable for different types of water supply networks and flow metering scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 This is a working principle diagram of the water meter flow real-time correction system based on intelligent sensors according to the present invention;

[0058] Figure 2 This is the working principle diagram of the frequency domain feature fusion module;

[0059] Figure 3 This is the working principle diagram of the error compensation decision module;

[0060] Figure 4 This is the working principle diagram of the frequency domain energy entropy calculation unit. DETAILED DESCRIPTION

[0061] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0062] See also Figure 1-Figure 4The present invention relates to a real-time correction system for water meter flow based on an intelligent sensor. The system includes the following functional modules that cooperate with each other to achieve real-time correction of water meter flow. The specific implementation steps are as follows:

[0063] The flow signal acquisition module uses a multi-band sensor deployed on the water meter pipeline to continuously acquire multi-dimensional vibration waveform data caused by the water flow within the pipeline. This sensor can collect vibration signals across different frequency ranges, such as simultaneously collecting vibration signals in the low, medium, and high frequency bands to comprehensively reflect the dynamic characteristics of the water flow in the pipeline. This multi-dimensional vibration waveform data includes, but is not limited to, radial, axial, and circumferential vibration signals of the pipeline. The spatial arrangement of multiple sensors enables simultaneous acquisition of vibrations in different directions.

[0064] Signal feature decomposition module: performs time-frequency domain decomposition and energy spectrum feature extraction on the acquired multi-dimensional vibration waveform data. Specifically, the continuous vibration waveform data is first divided into multiple waveform data segments according to a preset time window. The length of each time window can be set according to the water flow characteristics and subsequent processing requirements, for example, it can be set to a value between 0.1 seconds and 1 second. Then, each waveform data segment is input into a frequency domain encoder based on a multi-scale decomposition network. The encoder performs time-frequency domain decomposition of the waveform data segment at different scales through multi-layer convolution and pooling operations, extracts the energy spectrum features of each frequency band, and finally obtains a set of local frequency domain feature tensors of the flow state. Each feature tensor contains the energy distribution information of the corresponding time window and frequency band.

[0065] Frequency Domain Feature Fusion Module: This module dynamically suppresses noise and optimizes frequency band correlations for a collection of local frequency domain feature tensors of the traffic state. By analyzing the energy correlations and time series differences between different frequency bands, it suppresses noise interference and enhances the correlations between effective frequency bands, resulting in a collection of optimized local feature tensors of the traffic state. This optimized feature tensor more accurately reflects the actual traffic state, reducing the impact of noise and redundant information.

[0066] The Error Compensation Decision Module reconstructs global features based on the frequency domain energy distribution of a collection of locally optimized flow state feature tensors. By integrating the frequency domain energy information from each local feature, a flow error compensation decision feature map is generated. This feature map intuitively displays the error distribution and compensation direction in the current flow measurement, providing a basis for subsequent correction decisions.

[0067] The correction instruction execution module generates a real-time flow correction control signal based on the generated flow error compensation decision characteristic map. This signal is transmitted to the flow metering device, instructing it to execute the corresponding calibration strategy to achieve real-time correction of the water meter flow measurement error.

[0068] The present invention will be further described below in conjunction with Examples 1 to 5:

[0069] Example 1:

[0070] This embodiment describes in detail the specific structure and workflow of the signal feature decomposition module. The signal feature decomposition module includes a waveform data segmentation unit and a feature extraction unit, which work together to achieve time-frequency domain decomposition and energy spectrum feature extraction of multi-dimensional vibration waveform data.

[0071] The core function of the waveform data segmentation unit is to dynamically segment multidimensional vibration waveform data into preset time windows. The setting of the preset time window requires a comprehensive consideration of water flow characteristics and the real-time requirements of subsequent processing. For example, in industrial water use scenarios where the flow velocity in the pipeline is high and the state changes rapidly, the time window can be set to 0.2 seconds to ensure that the segmented waveform data segments can capture transient changes in water flow vibrations. In conventional residential water use scenarios, where the flow state is relatively stable, the time window can be appropriately increased to 0.8 seconds to reduce the number of data segments and the subsequent computational complexity. During the dynamic segmentation process, the waveform data segmentation unit performs a sliding intercept of the original vibration signal using continuous time windows. The waveform data segments within each time window contain complete multidimensional vibration information (e.g., radial, axial, and circumferential vibration signals), forming an ordered set of waveform data segments. It is important to note that adjacent time windows can overlap as needed (e.g., 10% of the window length) to prevent loss of boundary information due to signal truncation.

[0072] The feature extraction unit's primary function is to input each segment of a collection of waveform data into a frequency-domain encoder based on a multi-scale decomposition network to obtain a set of local frequency-domain feature tensors of the flow state. The multi-scale decomposition network employs a hierarchical convolutional architecture, with its core consisting of multiple convolutional layers of varying scales, activation layers, and pooling layers. Specifically, the input waveform data first enters the first convolutional layer, which uses a large convolution kernel (e.g., 5×5) to perform preliminary feature extraction on the input data, focusing on capturing global vibration characteristics in the low-frequency range, such as the overall low-frequency vibration trend of the pipeline. After the convolution operation, the output feature map is passed to an activation layer (e.g., a ReLU activation function) to introduce nonlinearity and enhance the model's ability to represent complex signals. Subsequently, the feature map is downsampled by a pooling layer (e.g., max pooling) to reduce the feature dimensionality while retaining key features.

[0073] The second convolutional layer uses medium-sized convolution kernels (e.g., 3×3) to extract local features in the mid-frequency band, such as the mid-frequency vibration components generated by water impacting pipe joints. After the same convolution, activation, and pooling operations, the output feature map is further passed to subsequent layers. Subsequent convolutional layers can use smaller convolution kernels (e.g., 1×1) as needed to capture detailed features in the high-frequency band, such as high-frequency vibration noise caused by turbulent water flow. The stride parameters of the convolution kernels in each layer can be adjusted according to feature extraction requirements. For example, setting a stride of 2 in the first layer to quickly reduce the dimension and a stride of 1 in subsequent layers to retain more details.

[0074] Through this multi-scale layered processing, the frequency domain encoder can perform multi-resolution time-frequency domain decomposition on the input waveform data segment. The feature maps output by each convolutional layer correspond to energy spectrum features at different frequency scales. For example, the first layer corresponds to low-frequency features, the second layer corresponds to medium-frequency features, and subsequent layers correspond to high-frequency features. These feature maps are stacked in the depth dimension to form a multi-dimensional feature tensor. For each waveform data segment, the frequency domain encoder ultimately outputs a local frequency domain feature tensor of the flow state containing the time dimension, frequency dimension, and energy dimension. This tensor records the energy distribution of different frequency bands within the time window. After all waveform data segments are processed by the frequency domain encoder, a set of local frequency domain feature tensors of the flow state is formed.

[0075] During implementation, the frequency domain encoder's network parameters are optimized using training data. During training, historically collected vibration waveform data and its corresponding true flow rate errors serve as input and labels. A backpropagation algorithm is used to adjust convolution kernel weights, biases, and other parameters, enabling the network to accurately extract frequency domain features related to flow rate errors from the vibration signal. For example, for waveform data segments containing flow rate measurement errors, the network must learn to enhance characteristic frequency bands associated with the error (e.g., abnormal vibration energy at specific frequencies) while suppressing irrelevant noise bands.

[0076] Furthermore, the feature extraction unit can configure multiple parallel frequency-domain encoder branches based on actual needs, each corresponding to a different multi-scale decomposition network structure, to achieve independent feature extraction for multi-dimensional vibration data (such as vibration signals in different directions). For example, separate encoder branches are designed for radial and axial vibration signals, each outputting a corresponding local frequency-domain feature tensor. Ultimately, the tensors from each branch are concatenated across dimensions to form a comprehensive tensor set containing multi-directional features. This parallel processing approach fully utilizes the spatial information of multi-dimensional data, improving the comprehensiveness of feature extraction.

[0077] The signal feature decomposition module transforms the raw vibration signal into a structured frequency-domain feature tensor through dynamic time windowing in the waveform data segmentation unit and multi-scale frequency-domain encoding in the feature extraction unit. This process not only discretizes the continuous time signal into processable segments but also, through hierarchical feature extraction using a multi-layer convolutional network, enables refined analysis of the different frequency components in the vibration signal, providing critical foundational data for subsequent frequency-domain feature fusion and error compensation decisions.

[0078] Example 2:

[0079] This example details the composition and working process of the frequency domain feature fusion module. The module includes a frequency domain tensor expansion unit, a frequency band energy correlation calculation unit, a dynamic noise suppression unit, and a feature optimization unit. Each unit implements dynamic noise suppression and frequency band correlation optimization for frequency domain features through a series of operations.

[0080] The function of the frequency domain tensor expansion unit is to perform matrix expansion on the local frequency domain feature tensor of the flow state along the frequency band dimension. The local frequency domain feature tensor of the flow state is a multi-dimensional data structure output by the feature extraction unit, which usually contains multiple dimensions such as time, frequency, and energy. When expanding along the frequency band dimension, the unit arranges the eigenvalues ​​of each frequency band of each tensor in the frequency band dimension in order and converts them into a one-dimensional feature vector. For example, if the frequency band dimension of a tensor contains 100 frequency bands, each tensor is expanded into a feature vector of length 100, and each element in the vector corresponds to the eigenvalue of a frequency band. Through this operation, the multi-dimensional feature tensor is converted into a two-dimensional matrix form, forming a set of local frequency domain feature vectors of the flow state, where each row of the matrix represents the eigenvector corresponding to a time window, and each column represents the eigenvalue distribution of a frequency band in different time windows.

[0081] The frequency band energy correlation calculation unit is used to calculate the frequency domain correlation coefficient between any two eigenvectors in the eigenvector set, generating a frequency band energy correlation matrix. This unit first maps each eigenvector into an orthogonal frequency domain space using the frequency domain projection subunit. Orthogonal frequency domain mapping can be achieved through orthogonal transformations, such as using a Fourier transform to convert eigenvectors from the time domain to the frequency domain, ensuring that different frequency bands are mathematically orthogonal, thereby eliminating redundant correlations between frequency bands. In the projected frequency domain eigenvector set after mapping, the dimensions (frequency bands) of each vector are independent of each other, facilitating subsequent correlation analysis.

[0082] The frequency domain correlation analysis subunit calculates the spectral energy similarity between any two projected eigenvectors based on the projected eigenvectors. Spectral energy similarity can be measured using methods such as vector inner product or cosine similarity. Essentially, it assesses the degree of similarity in the energy distributions of two frequency bands. For example, for two projected eigenvectors A and B, their cosine similarity is calculated. If the similarity is close to 1, the energy distributions of the two frequency bands are highly similar, potentially corresponding to the same physical phenomenon (such as vibration caused by water impacting the same component). If the similarity is close to 0, the energy distributions differ significantly, potentially representing different signal components (such as a valid signal and noise). The similarity values ​​of all pairwise eigenvectors are arranged in order to form an N×N frequency band energy correlation matrix (N is the number of frequency bands). The element in the i-th row and j-th column of this matrix represents the strength of the energy correlation between the i-th and j-th frequency bands.

[0083] The dynamic noise suppression unit modifies the weights of the frequency band energy correlation matrix based on the frequency domain differences between adjacent time windows in the eigenvector set. Frequency domain differences between adjacent time windows reflect the stability of the signal in the time series. If the eigenvalues ​​of a frequency band in adjacent windows fluctuate dramatically (e.g., a sudden increase or decrease in energy), the band may be subject to noise interference (e.g., transient pipeline vibration or external impact), and its correlation weight should be suppressed. Conversely, if the eigenvalues ​​change gradually, the band is considered to be a stable and valid signal, and its weight can be maintained or enhanced. The specific modification method is to calculate the absolute value of the eigenvalue difference between adjacent time windows for each frequency band and set a threshold (e.g., twice the average difference determined based on historical data). For frequency bands whose differences exceed the threshold, their weights in the correlation matrix are reduced proportionally to the difference (e.g., the larger the difference, the larger the weight reduction coefficient). This method generates a frequency band energy constrained correlation matrix that suppresses the correlation strength of noise bands while preserving the correlation of valid bands.

[0084] The feature optimization unit performs cross-band convolution on the frequency band energy constraint correlation matrix and the feature vector set. First, the frequency band energy constraint correlation matrix is ​​subjected to dilated convolution. Dilated convolution expands the receptive field of the convolution kernel by inserting zero values ​​between the weights of the standard convolution kernel, thereby capturing a wider range of frequency band correlation information without increasing the amount of computation. For example, using a 3×3 convolution kernel with a dilation ratio of 2, its actual receptive field is equivalent to that of a 5×5 convolution kernel, but the number of parameters is still 3×3. After dilated convolution, the correlation matrix is ​​converted into a frequency band energy constraint feature matrix, which contains long-range correlation information between different frequency bands.

[0085] The set of feature vectors and the frequency band energy constraint feature matrix are input into a bidirectional recursive encoding network. The bidirectional recursive encoding network consists of a forward recursive layer and a backward recursive layer: the forward recursive layer processes the feature vector sequence from front to back, capturing the impact of past time windows on the current window; the backward recursive layer processes from back to front, capturing the dependency of future time windows on the current window. Through this bidirectional processing, the network can learn the contextual dependencies between feature vectors, such as whether the feature changes in a certain frequency band are correlated with the feature changes in other frequency bands in the previous and next time windows. The output of the bidirectional recursive layer is a set of local context feature vectors of the traffic state, each of which integrates the frequency band correlation information of the current time window and the previous and next windows.

[0086] Finally, the feature optimization unit performs matrix reconstruction on the context feature vector set. Matrix reconstruction converts the two-dimensional feature vector set (time window × frequency band dimension) back into a multidimensional tensor, restoring information about time, frequency band, and other dimensions to form a locally optimized feature tensor for the traffic state. This tensor not only preserves the time-frequency domain information of the original features but also enhances the reliability and semantic information of the features through frequency band correlation optimization and noise suppression, providing higher-quality input for subsequent error compensation decisions.

[0087] In specific implementations, the matrix expansion method of the frequency-domain tensor expansion unit must strictly match the output dimensions of the feature extraction unit to ensure accurate correspondence between the eigenvalues ​​of each frequency band. The orthogonal projection and similarity calculations of the frequency-band energy correlation calculation unit can be implemented using existing mathematical libraries (such as Python's NumPy library), avoiding complex formula derivations. The bidirectional recurrent encoding network can adopt structures such as long short-term memory (LSTM) networks or gated recurrent units (GRU). By setting the appropriate hidden layer dimensions and number of iterations, computational efficiency and feature fusion effectiveness can be balanced.

[0088] Example 3:

[0089] This embodiment focuses on the specific implementation of the error compensation decision module. The module includes a local feature compression unit, a frequency domain energy entropy calculation unit, a reference frequency band determination unit, a dynamic compensation weight calculation unit, and a global reconstruction unit. Each unit generates a flow error compensation decision feature map by compressing, analyzing, and reconstructing frequency domain features.

[0090] The function of the local feature compression unit is to perform mean pooling processing on each tensor in the set of local optimized feature tensors of the flow state in the frequency domain dimension. The local optimized feature tensor of the flow state is a multi-dimensional data structure, which contains dimensions such as time, frequency band, and energy. The mean pooling processing is aimed at the frequency domain dimension, and the average value of all frequency band feature values ​​of each tensor in the frequency band dimension is calculated to compress the multi-dimensional tensor into a one-dimensional feature vector. For example, if the frequency band dimension of a tensor has 200 frequency bands, then after mean pooling, a mean value vector of length 200 is generated, which retains the overall energy distribution information of the tensor in the frequency domain dimension and eliminates the detailed fluctuations of the local frequency band. Through this operation, all tensors are converted into a set of local optimized feature vectors of the flow state, and the feature dimension is reduced from multi-dimensional to one-dimensional, which simplifies the complexity of subsequent calculations while retaining the main frequency domain energy characteristics.

[0091] The frequency domain energy entropy calculation unit is used to calculate the energy entropy value of each eigenvector in the eigenvector set and generate a flow dynamic energy entropy set. The specific implementation process is as follows: First, the median vector and range vector are calculated for each eigenvector. The median vector is a vector formed by sorting the elements in the eigenvector by numerical size and taking the middle value, reflecting the central tendency of the eigenvalue; the range vector is the difference between the maximum and minimum values ​​in the eigenvector, reflecting the fluctuation range of the eigenvalue. Next, the eigenvector and the median vector are differenced element by element to obtain the degree of deviation of each element from the median. The difference result is then cubed to amplify the larger deviation value, highlight the unevenness of the energy distribution, and form a flow characteristic energy difference vector. For example, if the difference between an element and the median is 2, it becomes 8 after the cubed power operation, and the difference is -1, it becomes -1. This nonlinear transformation enhances the impact of outliers.

[0092] Subsequently, the overall mean of the flow characteristic energy difference vector is calculated. This mean reflects the average degree of deviation of the elements in the eigenvector from the median. The mean is then compared to the cube of the range vector to obtain a dimensionless value that combines the degree of deviation and the fluctuation range of the eigenvalue. Finally, this value is input into a standardization function (such as a normalization function) to map it to the interval [0,1] to obtain the energy entropy value. A larger energy entropy value indicates a more uneven energy distribution of the eigenvector, which may contain more noise or abnormal features. A smaller energy entropy value indicates a more concentrated and stable energy distribution, which is more likely to correspond to the true flow state characteristics.

[0093] The reference frequency band determination unit selects the eigenvector corresponding to the minimum energy entropy value from the flow dynamic energy entropy set as the initial compensation reference vector. The minimum energy entropy value indicates that the eigenvector has the most uniform energy distribution and the highest stability, and therefore serves as the reference for error compensation. For example, if there are 100 eigenvectors in the energy entropy set, and the 35th vector has the smallest entropy value, this vector is selected as the initial compensation reference vector, representing the most reliable frequency domain feature under the current flow state.

[0094] The dynamic compensation weight calculation unit calculates dynamic compensation weights based on the frequency domain distance and energy entropy between each eigenvector and the initial compensation reference vector, generating a dynamic compensation weight set. The specific steps are as follows: First, the energy entropy value of the eigenvector is calculated and multiplied by the energy entropy value of the initial compensation reference vector. This is then multiplied by a first compensation coefficient to obtain a first dynamic compensation factor. The first compensation coefficient is an adjustable parameter (e.g., set to 0.5) that controls the influence of the energy entropy value on the compensation weight. If the energy entropy value of a particular eigenvector is less than that of the reference vector, indicating a more stable energy distribution, the first dynamic compensation factor will increase; otherwise, it will decrease.

[0095] Next, the absolute value of the Manhattan distance between the eigenvector and the initial compensated reference vector is calculated. This distance measures the absolute difference between the two vectors in the frequency domain. The Manhattan distance is calculated as the sum of the absolute values ​​of the differences between the corresponding elements. For example, the Manhattan distance between vectors A = [a1, a2, ..., an] and vector B = [b1, b2, ..., bn] is |a1-b1| + |a2-b2| + ... + |an-bn|. The absolute value of this distance is multiplied by the second compensation coefficient (for example, set to 0.1) to obtain the second dynamic compensation factor, which reflects the degree of similarity between the eigenvector and the reference vector. Smaller distances correspond to smaller factors, indicating greater similarity.

[0096] Finally, the first dynamic compensation factor is multiplied by the second dynamic compensation factor to obtain the dynamic compensation weight. For example, if the first dynamic compensation factor of a eigenvector is 0.8 and the second dynamic compensation factor is 0.3, the dynamic compensation weight is 0.24. The dynamic compensation weight combines the energy entropy value and the frequency domain distance: eigenvectors with lower energy entropy values ​​and smaller frequency domain distances have greater weights and a higher proportion in error compensation. Conversely, eigenvectors with lower weights have lower reliability or greater deviation from the benchmark, and thus contribute less to compensation.

[0097] The global reconstruction unit uses the dynamic compensation weight set to perform weighted superposition on the feature vector set to generate a flow error compensation decision feature map. The specific operation is: each element of each feature vector is multiplied by the corresponding dynamic compensation weight to obtain a weighted feature vector, and then all weighted feature vectors are added element by element to form the final feature map. For example, if there are M feature vectors, each of length N, then an N-dimensional feature map is generated after weighted superposition, in which the value of each element is the cumulative sum of the elements at the corresponding position of all feature vectors multiplied by the weight. Through weight distribution, this feature map highlights the contribution of feature vectors with high reliability and similarity to the benchmark, suppresses the influence of noise and abnormal features, and thus intuitively reflects the distribution and compensation direction of errors in the current flow measurement.

[0098] In specific implementations, the mean pooling of the local feature compression unit can be efficiently implemented through matrix operations, such as using the mean function of the tensor library to calculate along a specified dimension. The median vector, range vector, and normalization function of the frequency domain energy entropy calculation unit can be implemented using existing data processing tools, avoiding the derivation of complex formulas. The Manhattan distance and coefficient product in the dynamic compensation weight calculation can be completed using the vector operation library to ensure computational efficiency. The weighted superposition operation of the global reconstruction unit requires attention to vector dimension alignment to ensure that the elements of each feature vector correctly correspond to the weights.

[0099] Example 4:

[0100] This embodiment details the operating principle and implementation of the correction instruction execution module. The core function of the correction instruction execution module is to convert the flow error compensation decision characteristic map into a real-time flow correction control signal, which instructs the flow metering device to execute the calibration strategy. Its specific implementation relies on a correction signal generator based on random forests.

[0101] The flow error compensation decision feature map is a multidimensional data structure output by the error compensation decision module, usually represented in matrix or vector form, where each element corresponds to error compensation information in the frequency domain or time dimension. Because the random forest algorithm requires a one-dimensional feature vector as input, the feature map needs to be preprocessed. The preprocessing process includes dimensionality compression and normalization: Dimensionality compression converts the multidimensional feature map into a one-dimensional vector through a flattening operation, for example, expanding an M×N matrix into an M×N vector by row or column; normalization maps the element values ​​in the vector to a specific range (such as [0,1]) to prevent the impact of eigenvalue scale differences on model training and prediction results.

[0102] The random forest-based correction signal generator consists of multiple decision trees, which improve the model's generalization and robustness through ensemble learning. During the training phase, a historical training dataset is prepared, where each piece of data contains two parts: a preprocessed feature vector (corresponding to a historical sample of the flow error compensation decision feature map) and a corresponding label (i.e., the actual correction control signal to be generated). The form of the label is determined by the calibration strategy of the flow metering equipment. For example, for electromagnetic water meters, the label can be defined as the discrete value of the sensitivity adjustment coefficient (such as -3, -2, -1, 0, +1, +2, +3, corresponding to different degrees of gain or attenuation, respectively); for mechanical water meters, the label can represent the adjustment gear of the gear ratio.

[0103] During training, each decision tree is constructed using bootstrap sampling: a subset of samples (typically 60%-80% of the original dataset) and features (typically the square root of the total number of features) are randomly sampled from the training data for training, reducing the risk of overfitting and increasing model diversity. Each node in the decision tree selects the optimal splitting feature and split point using metrics such as Gini Impurity or Information Gain, partitioning the sample into different sub-nodes until a preset maximum depth or minimum number of samples is reached. For example, a node selects a threshold T for feature X as the splitting condition. Samples with feature values ​​less than T are assigned to the left subtree, while samples with feature values ​​greater than or equal to T are assigned to the right subtree. This recursive splitting process builds the complete tree structure.

[0104] After all decision trees have been trained, the correction signal generator has predictive capabilities. For the input real-time flow error compensation decision feature map, the same preprocessing operations as the training data are first performed to obtain a one-dimensional normalized feature vector. This vector is then input into the random forest model, and each decision tree makes an independent prediction, outputting a discrete correction signal value. The final real-time flow correction control signal is determined by voting: for classification tasks, the category predicted by the majority of decision trees is selected as the result; for regression tasks, the average of the predicted values ​​of all decision trees is taken. For example, if 4 out of 7 decision trees predict a correction signal of +2, 2 predict +1, and 1 predicts +3, the final signal is +2.

[0105] After the real-time flow correction control signal is generated, it needs to be transmitted to the flow metering device to execute the calibration strategy. Different types of flow metering devices correspond to different calibration mechanisms:

[0106] Electromagnetic water meters: Their operating principle is based on Faraday's law of electromagnetic induction. Calibration strategies can be implemented by adjusting the sensor's excitation current or the gain of the signal amplification circuit. For example, a calibration signal of +1 indicates a 10% increase in the excitation current to improve the sensor's sensitivity to low flow rates; a signal of -2 indicates a 20% decrease in the signal amplification gain to suppress signal saturation at high flow rates.

[0107] Mechanical water meters primarily measure flow through a mechanical transmission mechanism. Calibration strategies can be implemented by adjusting the gear ratio between the impeller shaft and the counter. For example, a calibration signal of +3 indicates switching to a gear set with a larger transmission ratio (e.g., from 1:100 to 1:103), allowing the counter to record more pulses at the same flow rate, thereby compensating for negative errors. A signal of -1 indicates a reduction in the transmission ratio (e.g., from 1:100 to 1:97) to compensate for positive errors.

[0108] Ultrasonic water meters measure flow rate based on the propagation time difference of ultrasonic waves in the fluid. Calibration strategies can be implemented by adjusting signal filtering parameters or the sound velocity compensation factor. For example, if the calibration signal indicates a positive error, increasing the sound velocity compensation factor will reduce the calculated flow rate, thereby reducing the metering result.

[0109] In practice, the parameters of the random forest model need to be adjusted based on actual needs, such as the number of decision trees (typically 50-200), maximum depth (typically 5-20 layers), and minimum number of sample splits (typically 2-10). Parameter adjustment can be accomplished through cross-validation, which involves partitioning the training data into validation sets to test the model's prediction accuracy under different parameter combinations and select the optimal parameter combination. Furthermore, the correction signal generator can be deployed on edge computing devices or cloud servers, communicating with the front-end feature generation module and back-end flow metering equipment via a real-time data interface to ensure the real-time and reliability of correction instructions.

[0110] Example 5:

[0111] This example focuses on the adaptive calibration mechanism of a smart water meter system. This mechanism achieves intelligent adaptation to different water flow environments by dynamically adjusting parameter weights and calibration cycles. The adaptive calibration mechanism includes an environmental feature extraction unit, a parameter weight adjustment unit, a calibration cycle decision unit, and a feedback verification unit. These units work together to optimize the system's real-time calibration performance.

[0112] The environmental feature extraction unit continuously collects pipeline environmental parameters, including temperature, pressure, water hardness, and pipeline material information. These parameters are obtained through sensors distributed at different locations in the pipeline and converted into digital signals for transmission to the system. The temperature sensor uses a platinum resistance thermometer, whose resistance value changes linearly with temperature. By measuring the resistance value and converting it into a digital signal, accurate water temperature data can be obtained. The pressure sensor is based on the principle of piezoresistive effect, converting pressure changes into electrical signal changes, and obtaining the pressure value after amplification and digital processing. The water hardness sensor is based on the principle of ion-selective electrode, and determines the water hardness level by detecting the concentration of calcium and magnesium ions in the water. Pipeline material information is obtained through pre-configuration or self-learning. For example, during the system initialization phase, the user enters the pipeline material type, or the system automatically identifies the pipeline material by analyzing the characteristics of the water flow signal.

[0113] The parameter weight adjustment unit dynamically adjusts the weights of each calibration parameter based on environmental characteristics. This unit first preprocesses the environmental characteristics, including data cleaning, normalization, and feature selection. Data cleaning removes outliers and noise from sensor data, for example, by smoothing temperature and pressure data using a sliding window filter algorithm. Normalization maps feature values ​​of varying ranges to a uniform interval, such as [0, 1], to eliminate the impact of feature scale differences on weight adjustment. Feature selection selects key features that significantly influence calibration parameters from the original environmental characteristics. For example, in some scenarios, temperature and pressure are the primary factors affecting flow measurement, while water hardness has a smaller impact, so the water hardness feature can be ignored.

[0114] Subsequently, the parameter weight adjustment unit inputs the preprocessed environmental feature vector into the pre-trained weight mapping model. The model can be expressed as:

[0115]

[0116] in, is the calibration parameter weight vector, is the preprocessed environmental feature vector, Represents the weight mapping function. This weight mapping function is trained using historical data. Its specific form can be selected based on the actual situation, such as a multilayer perceptron or a decision tree. For example, when environmental characteristics indicate low water temperature, the model increases the weight of temperature-related calibration parameters to compensate for the impact of low temperatures on flow measurement. In practice, the weight mapping model uses incremental learning, continuously updating model parameters based on newly collected data to adapt to environmental changes.

[0117] The calibration cycle decision unit dynamically adjusts the calibration cycle based on flow stability. Flow stability is assessed by analyzing the amplitude and frequency of the flow signal's fluctuations. The specific steps are as follows: First, a sliding time window is set in the calibration cycle decision unit to capture a segment of the flow signal. Then, the standard deviation of the flow signal within this time period is calculated as an indicator of the fluctuation amplitude. A larger standard deviation indicates more severe flow signal fluctuations and poorer flow stability.

[0118] To analyze the frequency components of the flow signal, the calibration cycle decision unit performs a Fourier transform on the intercepted flow signal. The Fourier transform converts the time-domain signal into a frequency-domain signal, thereby determining the strength of the different frequency components within the signal. By analyzing the frequency-domain signal, the dominant fluctuation frequency of the flow signal is determined. A high dominant fluctuation frequency indicates rapid flow fluctuations and unstable flow conditions.

[0119] The calibration cycle decision unit pre-sets fluctuation amplitude thresholds and primary fluctuation frequency thresholds. When the calculated fluctuation amplitude exceeds the amplitude threshold, or the primary fluctuation frequency exceeds the primary fluctuation frequency threshold, it indicates unstable flow conditions. In this case, the calibration cycle decision unit shortens the calibration cycle and increases the calibration frequency to track changes in flow conditions and ensure accurate flow measurement.

[0120] Conversely, when both the fluctuation amplitude and the primary fluctuation frequency are below their respective thresholds, the flow state is relatively stable. The calibration cycle decision unit sets a longer calibration cycle for stable flow conditions. This is because when the flow is stable, flow measurement errors change more slowly, making frequent calibration unnecessary. Extending the calibration cycle reduces system energy consumption, minimizes the impact of calibration operations on normal flow measurement, and extends the lifespan of system components.

[0121] The sliding time window length of the calibration cycle decision unit can be adjusted according to the actual application scenario. For example, in residential water use scenarios, water flow fluctuations are relatively small and change slowly. The sliding time window length can be set to 10 minutes to more comprehensively capture the changing trends of water flow. In industrial water use scenarios, water flow fluctuations may be large and change rapidly. In order to promptly detect changes in water flow status, the sliding time window length can be set to 5 minutes. In this way, the calibration cycle decision unit can flexibly adjust the calibration cycle according to different application scenarios and water flow characteristics, optimizing the system's operating efficiency while ensuring measurement accuracy.

[0122] The feedback verification unit verifies calibration results in real time to ensure calibration validity. This unit compares calibrated flow data with a reference standard and calculates the deviation. If the deviation exceeds the acceptable range, a recalibration process is triggered, and calibration parameters are corrected. The reference standard can be obtained from a high-precision flow meter or through statistical analysis of historical data. For example, in industrial water use scenarios, an electromagnetic flowmeter can be used as a reference standard, and calibration results can be regularly compared to ensure long-term system stability.

[0123] In specific implementations, the sensors of the environmental feature extraction unit need to be calibrated regularly to ensure data accuracy. The calibration process usually uses standard substances or known parameters for comparison, and adjusts the output value of the sensor to make it consistent with the standard value. The weight mapping model of the parameter weight adjustment unit can adopt an incremental learning method to continuously update the model parameters based on the newly collected data. The sliding time window length of the calibration cycle decision unit can be adjusted according to the actual application scenario, for example, it can be set to 10 minutes in the residential water use scenario and 5 minutes in the industrial water use scenario. The deviation threshold of the feedback verification unit can be set according to the measurement accuracy requirements, for example, it can be set to ±2%.

[0124] The adaptive calibration mechanism dynamically adjusts calibration parameters and cycles through real-time perception and analysis of environmental characteristics, enabling the system to adapt to varying flow environments and operating conditions. This intelligent adaptability improves the pertinence and effectiveness of calibration, ensuring accurate and reliable flow measurement while reducing system energy consumption and maintenance costs. In practical applications, this mechanism can significantly improve the performance of smart water meters in complex environments, providing strong support for the precise measurement and management of water resources.

[0125] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0126] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A real-time correction system for water meter flow based on intelligent sensors, characterized in that: include: Flow signal acquisition module, used to obtain multi-dimensional vibration waveform data of water meter pipelines through multi-band sensors; a signal feature decomposition module, configured to perform time-frequency domain decomposition and energy spectrum feature extraction on the multi-dimensional vibration waveform data to obtain a set of local frequency domain feature tensors of the flow state; a frequency domain feature fusion module, configured to perform dynamic noise suppression and frequency band correlation optimization on the set of local frequency domain feature tensors of the flow state to obtain a set of local optimized feature tensors of the flow state; An error compensation decision module is used to perform global feature reconstruction based on frequency domain energy distribution on the set of local optimized feature tensors of the flow state to generate a flow error compensation decision feature map; a correction instruction execution module, configured to generate a real-time flow correction control signal according to the flow error compensation decision characteristic diagram; The frequency domain feature fusion module includes: A frequency domain tensor expansion unit, configured to perform matrix expansion on the flow state local frequency domain feature tensor along a frequency band dimension to obtain a set of flow state local frequency domain feature vectors; a frequency band energy correlation calculation unit, configured to calculate a frequency domain correlation coefficient between any two eigenvectors in the set of local frequency domain eigenvectors of the flow state to generate a frequency band energy correlation matrix; a dynamic noise suppression unit, configured to perform weight correction on the frequency band energy correlation matrix according to the frequency domain differences of adjacent time windows in the set of local frequency domain feature vectors of the flow state to obtain a frequency band energy constraint correlation matrix; a feature optimization unit, configured to perform cross-band convolution fusion on the frequency band energy constraint association matrix and the set of local frequency domain feature vectors of the flow state to obtain a local optimized feature tensor of the flow state; The error compensation decision module includes: A local feature compression unit, configured to perform mean pooling processing in the frequency domain on each tensor in the set of the flow state local optimization feature tensors to obtain a set of flow state local optimization feature vectors; A frequency domain energy entropy calculation unit, configured to calculate the energy entropy value of each eigenvector in the set of local optimized eigenvectors of the flow state to generate a flow dynamic energy entropy set; A reference frequency band determination unit, configured to select a flow state local optimization feature vector corresponding to a minimum energy entropy value in the flow dynamic energy entropy set as an initial compensation reference vector; a dynamic compensation weight calculation unit, configured to calculate the dynamic compensation weight of each eigenvector in the set of the flow state local optimization eigenvectors and the initial compensation reference vector according to the frequency domain distance between each eigenvector and the initial compensation reference vector and the energy entropy value of each eigenvector to generate a dynamic compensation weight set; A global reconstruction unit, configured to perform weighted superposition on the set of local optimized feature vectors of the flow state using the dynamic compensation weight set to generate the flow error compensation decision feature map; The correction instruction execution module is specifically implemented as follows: The flow error compensation decision feature map is input into a correction signal generator based on random forest to obtain the real-time flow correction control signal, which is used to indicate the calibration strategy of the flow metering device.

2. The water meter flow real-time correction system based on intelligent sensor according to claim 1 is characterized in that: The signal feature decomposition module includes: a waveform data segmentation unit, configured to dynamically divide the multi-dimensional vibration waveform data according to a preset time window to obtain a set of waveform data segments; The feature extraction unit is used to input each segment of the waveform data segment set into a frequency domain encoder based on a multi-scale decomposition network to obtain a set of local frequency domain feature tensors of the flow state.

3. The water meter flow real-time correction system based on intelligent sensor according to claim 1 is characterized in that: The frequency band energy correlation calculation unit includes: A frequency domain projection subunit, configured to map each feature vector in the set of local frequency domain feature vectors of the flow state to an orthogonal frequency domain space to obtain a set of projected frequency domain feature vectors; The frequency domain correlation analysis subunit is used to calculate the spectrum energy similarity between any two projected feature vectors in the set of projected frequency domain feature vectors to generate the frequency band energy correlation matrix composed of multiple frequency band energy correlation values.

4. The water meter flow real-time correction system based on intelligent sensor according to claim 3 is characterized in that: The feature optimization unit is specifically implemented as follows: Performing a dilated convolution process on the frequency band energy constraint association matrix to obtain a frequency band energy constraint feature matrix; Inputting the set of local frequency domain feature vectors of the traffic state and the frequency band energy constraint feature matrix into a bidirectional recursive coding network to obtain a set of local context feature vectors of the traffic state; Matrix reconstruction is performed on the set of the traffic state local context feature vectors to obtain the traffic state local optimized feature tensor.

5. The water meter flow real-time correction system based on intelligent sensor according to claim 1 is characterized in that: The frequency domain energy entropy calculation unit is specifically implemented as follows: Calculating the median vector and the range vector of the local optimized characteristic vector of the flow state; Performing element-by-element difference calculation on the local optimized feature vector of the flow state and the median vector, and performing a cubic power operation on the difference result to obtain a flow feature energy difference vector; Calculating the overall mean of the flow characteristic energy difference vector; A ratio operation is performed on the mean value and the cube value of the extreme difference vector, and the result is input into a normalization function to obtain the energy entropy value.

6. The water meter flow real-time correction system based on intelligent sensor according to claim 5 is characterized in that: The dynamic compensation weight calculation unit is specifically implemented as follows: Multiplying the energy entropy value of the local optimization characteristic vector of the flow state and the energy entropy value of the initial compensation reference vector by a first compensation coefficient to obtain a first dynamic compensation factor; Multiplying the absolute value of the Manhattan distance between the local optimized characteristic vector of the flow state and the initial compensation reference vector by a second compensation coefficient to obtain a second dynamic compensation factor; The first dynamic compensation factor is multiplied by the second dynamic compensation factor to obtain the dynamic compensation weight.

7. The correction method of the water meter flow real-time correction system based on the intelligent sensor according to any one of claims 1 to 6, characterized in that: include: Collect multi-dimensional vibration waveform data of water meter pipes through multi-band sensors; Performing time-frequency domain decomposition and energy spectrum feature extraction on the multi-dimensional vibration waveform data to obtain a set of flow state local frequency domain feature tensors; Performing dynamic noise suppression and frequency band correlation optimization on the set of local frequency domain feature tensors of the flow state to obtain a set of local optimized feature tensors of the flow state; Performing global feature reconstruction based on frequency domain energy distribution on the set of local optimized feature tensors of the flow state to generate a flow error compensation decision feature map; A real-time flow correction control signal is generated according to the flow error compensation decision characteristic diagram.

Images