Error checking method and system for intelligent water meter

Abstract

The invention provides an error checking method and system of an intelligent water meter, and relates to the technical field of error checking of the intelligent water meter, and the method comprises the following steps: collecting multi-modal monitoring data of the intelligent water meter; performing signal preprocessing on the water flow acoustic signal and the mechanical vibration signal to generate an acoustic fusion signal, extracting time-frequency domain features of the acoustic fusion signal related to errors, and constructing an optimal error feature set; performing pattern recognition based on the optimal error feature set and a preset standard verification feature library to generate an error preliminary result; constructing an error prediction and verification model, verifying the error preliminary result in combination with the optimal error feature set and the flow data, and outputting an error verification result; when the error grade reaches a preset error grade threshold value, an early warning mechanism is triggered, and early warning information is pushed, so that the metering performance of the water meter can be evaluated in real time, the error change trend can be dynamically predicted, the timeliness and effectiveness of error verification are effectively guaranteed, and the dynamic error management and control requirements of the intelligent water meter are comprehensively met.

Description

Technical Field

[0001] This invention relates to the field of smart water meter error verification technology, and in particular to a smart water meter error verification method and system. Background Technology

[0002] In the field of modern water resource management and metering, smart water meters are widely used as important metering devices in residential and industrial water metering scenarios.

[0003] However, over long-term use, smart water meters experience increasing metering errors due to factors such as mechanical wear, water quality, and temperature variations, which in turn affects metering accuracy and the fairness of water billing. Traditional smart water meter error verification methods primarily rely on periodic disassembly and laboratory calibration, which is not only labor-intensive and costly but also poses risks of water outages and meter damage. Furthermore, some existing technologies employ fixed-period sampling methods, but these fail to reflect the error variation patterns of smart water meters in real-world usage environments. Meanwhile, other technologies use simple linear models for error estimation, which cannot accurately describe the nonlinear characteristics of smart water meter errors as they change with usage conditions. In addition, existing error verification methods often lack full utilization of historical smart water meter data, preventing the achievement of data-driven intelligent error prediction.

[0004] Therefore, it is necessary to provide an error verification method and system for smart water meters to solve the above-mentioned technical problems. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention provides an error verification method and system for smart water meters, which solves the problems of existing technologies being unable to monitor water meter measurement errors in real time and accurately predict error change trends.

[0006] This invention provides an error verification method for smart water meters, the method comprising: Collect multimodal monitoring data from smart water meters, wherein the multimodal monitoring data includes at least water flow acoustic signals, mechanical vibration signals, and flow reference data; The acoustic signal of the water flow and the mechanical vibration signal are preprocessed to generate an acoustic fusion signal. The time-frequency domain features of the acoustic fusion signal that are highly correlated with the error of the smart water meter are extracted, and the optimal acoustic error feature set is selected and constructed. Based on the optimal acoustic error feature set and the preset standard acoustic test feature library, acoustic pattern recognition is performed to evaluate the operating status of the smart water meter in real time and generate preliminary results of the smart water meter error. A dynamic error prediction and verification model for smart water meters is constructed. Combining the optimal acoustic error feature set and the flow reference data, the preliminary error results of the smart water meters are verified and analyzed, and the smart water meter error verification results, including the smart water meter error level and the smart water meter error prediction trend, are output. When the error level of the smart water meter reaches the preset error level threshold, an early warning mechanism is automatically triggered and an early warning message is pushed.

[0007] Preferably, the step of performing acoustic signal preprocessing on the water flow acoustic signal and the mechanical vibration signal to generate an acoustic fusion signal, extracting time-frequency domain features that are highly correlated with the smart water meter error from the acoustic fusion signal, and screening to construct an optimal acoustic error feature set specifically includes: The acoustic signal preprocessing includes eliminating pipe background noise and electromagnetic interference noise in the water flow acoustic signal and the mechanical vibration signal based on a wavelet threshold denoising algorithm, and then adjusting the denoised water flow acoustic signal and the mechanical vibration signal to the same time dimension through timestamp alignment technology. The water flow acoustic signal and the mechanical vibration signal are fused using an attention-based weighted fusion algorithm. The weights are dynamically assigned according to the contribution of the water flow acoustic signal and the mechanical vibration signal to the error of the smart water meter under different flow levels, and the acoustic fusion signal is generated. A complementary combination of short-time Fourier transform and wavelet packet decomposition is used to extract time-frequency domain features from the acoustic fusion signal. The time-frequency domain features include center frequency, energy entropy, wavelet packet node energy, and short-time energy variance. The correlation coefficient between the time-frequency domain features and the smart water meter error is calculated using the ReliefF algorithm. Time-frequency domain features with correlation coefficients greater than a preset correlation coefficient threshold are retained to generate a preliminary acoustic error feature set. The importance of the time-frequency domain features in the preliminary acoustic error feature set is ranked using the random forest algorithm. Time-frequency domain features in the preliminary acoustic error feature set that meet the preset feature quantity requirements are retained, and finally, the optimal acoustic error feature set is constructed.

[0008] Preferably, the step of performing acoustic pattern recognition based on the optimal acoustic error feature set and the preset standard acoustic verification feature library, and evaluating the operating status of the smart water meter in real time and generating preliminary results of the smart water meter error, specifically includes: The acoustic fingerprint of the smart water meter is extracted based on the optimal acoustic error feature set, and the acoustic fingerprint is normalized. From the preset standard acoustic verification feature library, a standard acoustic fingerprint template that matches the model and service life of the smart water meter is matched. The standard acoustic fingerprint template is associated with the standard error range and normal acoustic feature threshold of the corresponding flow point in the smart water meter. The similarity value between the acoustic fingerprint and the standard acoustic fingerprint template is calculated using a dynamic time warping algorithm. Calculate the Mahalanobis distance between the acoustic fingerprint and the standard acoustic fingerprint template. If the Mahalanobis distance is greater than the normal acoustic feature threshold, then determine that the smart water meter has an abnormal operating state and generate an abnormal state determination result. Based on the mapping relationship between the similarity value and the standard error interval, and combined with the abnormal operating status determination result, the operating status of the smart water meter is evaluated in real time, and a preliminary result of the smart water meter error is generated.

[0009] Preferably, the construction of the smart water meter dynamic error prediction and verification model, combined with the optimal acoustic error feature set and the flow reference data, verifies and analyzes the preliminary results of the smart water meter error, and outputs the smart water meter error verification results including the smart water meter error level and the smart water meter error prediction trend, specifically including: Extract the traffic statistical features from the traffic reference data, and fuse the traffic statistical features with the optimal acoustic error feature set through feature concatenation to generate a multi-dimensional verification input vector. The dynamic error prediction and verification model for the smart water meter is constructed, which includes a basic prediction layer and a verification and correction layer. The basic prediction layer deploys a support vector regression model and a gradient boosting tree model in parallel. The multi-dimensional verification input vector is input into both the support vector regression model and the gradient boosting tree model, and the preliminary error result of the smart water meter is used as the initial supervision signal. The preliminary prediction error values ​​of the support vector regression model and the gradient boosting tree model are output respectively. The verification correction layer introduces an attention mechanism, dynamically allocating the output weights of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistics features to the smart water meter error. Based on the preliminary prediction error value and the corresponding output weights, a fused prediction error value is output. The fusion prediction error value is matched with a preset error level range to determine the error level of the smart water meter; Historical fusion prediction error values ​​are extracted based on an adaptive sliding time window. Combined with these fusion prediction error values, a multi-dimensional time-series trend fitting algorithm is used to generate the smart water meter error prediction trend. The smart water meter error level and the smart water meter error prediction trend are then summarized, and the smart water meter error verification result is output.

[0010] Preferably, the multidimensional verification input vector X is input into the support vector regression model and the gradient boosting tree model, and the preliminary error result of the smart water meter is used as the initial supervision signal. Output the preliminary prediction error values ​​of the support vector regression model respectively. The initial prediction error value of the gradient boosting tree model The corresponding calculation formula is as follows: In the formula, Represents the Lagrange multiplier in a support vector regression model; This represents the kernel function in a support vector regression model; represents the i-th multidimensional validation input vector in the training samples of the support vector regression model; b represents the bias term in the support vector regression model; N represents the number of training samples in the support vector regression model; C represents the regularization coefficient in the support vector regression model; This represents the mean vector of the multidimensional validation input vectors in the training samples of the support vector regression model. Denotes the Euclidean norm; This represents the learning rate in a gradient boosting tree model; This represents the predicted output of the t-th decision tree in the gradient boosting tree model for the multidimensional validation input vector X; T represents the total number of decision trees in the gradient boosting tree model. This represents the initial supervision signal in the gradient boosting tree model. The weighting adjustment coefficient; This represents the natural exponential function with base e.

[0011] Preferably, the verification and correction layer introduces an attention mechanism, dynamically allocating the output weights of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistics features to the smart water meter error, and outputting a fused prediction error value based on the preliminary prediction error value and the corresponding output weights, specifically including: The contribution of the optimal acoustic error feature set and the flow statistics feature to the error of the smart water meter is determined by the information gain algorithm. , And satisfy The corresponding calculation formula is as follows: In the formula, Represents the optimal acoustic error feature set Information gain value; Indicates flow statistics characteristics Information gain value; The information entropy represents the error result U of the smart water meter; The error result U of the smart water meter is based on the optimal acoustic error feature set. Conditional information entropy; The error result U of the smart water meter is based on flow statistics characteristics. Conditional information entropy; Based on the contribution , Dynamically allocate the output weights of the support vector regression model , , This represents the feature contribution adjustment coefficient; based on the output weights of the support vector regression model. With the output weights of the gradient boosting tree model , The preliminary prediction error values ​​output by the support vector regression model and the gradient boosting tree model are weighted and fused to output a fused prediction error value R.

[0012] Preferably, the step of extracting historical fusion prediction error values ​​based on an adaptive sliding time window, and combining these fusion prediction error values ​​to generate the smart water meter error prediction trend using a multi-dimensional time-series trend fitting algorithm, specifically includes: An adaptive sliding time window is constructed. The window duration of the adaptive sliding time window is dynamically adjusted according to the fluctuation frequency of the historical fusion prediction error value. After filtering out abnormal error data that exceed the preset deviation range based on the 3σ principle, the continuous historical fusion prediction error values ​​within the window are extracted. Using the multi-dimensional time-series trend fitting algorithm, with the smart water meter error level as the fitting weight, curve fitting is performed on the extracted historical fusion prediction error value and the fusion prediction error value. Simultaneously, extreme outliers in the fitted data are filtered out based on the standard deviation to generate the smart water meter error prediction trend that includes the error change rate and error extreme values.

[0013] An error verification system for smart water meters, the system comprising: The data acquisition module is used to collect multimodal monitoring data from the smart water meter. The multimodal monitoring data includes at least water flow acoustic signals, mechanical vibration signals, and flow reference data. The feature extraction module is used to perform acoustic signal preprocessing on the water flow acoustic signal and the mechanical vibration signal to generate an acoustic fusion signal, extract time-frequency domain features that are highly correlated with the smart water meter error of the acoustic fusion signal, and screen and construct the optimal acoustic error feature set. The error generation module is used to perform acoustic pattern recognition based on the optimal acoustic error feature set and the preset standard acoustic test feature library, to evaluate the operating status of the smart water meter in real time and generate preliminary results of the smart water meter error. The error verification module is used to construct a dynamic error prediction and verification model for smart water meters. Combining the optimal acoustic error feature set and the flow reference data, it verifies and analyzes the preliminary error results of the smart water meters and outputs the smart water meter error verification results, including the smart water meter error level and the smart water meter error prediction trend. The early warning triggering module is used to automatically trigger the early warning mechanism and push early warning information when the error level of the smart water meter reaches a preset error level threshold.

[0014] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the steps of an error verification method for a smart water meter as described in any of the above claims.

[0015] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of an error verification method for a smart water meter as described in any of the above claims.

[0016] Compared with related technologies, the error verification method and system for smart water meters provided by this invention have the following beneficial effects: This invention collects multimodal monitoring data from smart water meters, including at least water flow acoustic signals, mechanical vibration signals, and flow reference data. It preprocesses the water flow acoustic signals and mechanical vibration signals to generate an acoustic fusion signal, extracts time-frequency domain features highly correlated with smart water meter errors from the fusion signal, and constructs an optimal acoustic error feature set. Based on the optimal acoustic error feature set and a preset standard acoustic verification feature library, it performs acoustic pattern recognition to evaluate the smart water meter's operating status in real time and generate preliminary smart water meter error results. It then constructs a dynamic error prediction and verification model for the smart water meter, combines the optimal acoustic error feature set and flow reference data to verify and analyze the preliminary smart water meter error results, and outputs smart water meter error verification results including the smart water meter error level and error prediction trend. When the smart water meter error level reaches a preset error threshold, an early warning mechanism is automatically triggered and an early warning message is pushed. This enables real-time evaluation of the smart water meter's metering performance, dynamic prediction of error change trends, effectively ensuring the timeliness and effectiveness of error verification, and fully meeting the dynamic control requirements of smart water meter errors.

[0017] This invention acquires multimodal monitoring data to capture acoustic signals of water flow, mechanical vibration signals, and flow reference data during the operation of smart water meters, comprehensively capturing key status information during smart water meter operation and avoiding the limitations of verification under a single data dimension. In the signal processing stage, this invention uses a wavelet threshold denoising algorithm to eliminate pipeline background noise and electromagnetic interference, combined with timestamp alignment technology and an attention mechanism weighted fusion algorithm to generate an acoustic fusion signal, significantly improving signal purity and fusion effectiveness. Then, it extracts time-frequency domain features through complementary extraction of short-time Fourier transform and wavelet packet decomposition, and uses the ReliefF algorithm and random forest algorithm to screen and construct an optimal acoustic error feature set, ensuring high correlation between features and errors. This invention achieves real-time evaluation of the smart water meter's operating status and generates preliminary results of smart water meter errors through acoustic pattern recognition based on the optimal acoustic error feature set and a preset standard acoustic verification feature library. By normalizing acoustic fingerprints, calculating similarity using dynamic time warping algorithms, and determining Mahalanobis distance, it solves the problem of real-time monitoring difficulties in traditional methods. This invention constructs a dynamic error prediction and verification model. The basic prediction layer uses a support vector regression model and a gradient boosting tree model deployed in parallel to achieve initial prediction. The verification and correction layer dynamically allocates weights and outputs a fused prediction error value by introducing an attention mechanism. Combined with an adaptive sliding time window and a multi-dimensional time-series trend fitting algorithm, it generates a prediction trend for smart water meter errors. This solves the problem of inaccurate predictions in traditional linear models, improves prediction accuracy, and precisely determines the error level. Furthermore, this invention employs an automatic early warning mechanism to promptly push warning information when the smart water meter error level reaches a preset error threshold, ensuring metering accuracy and preventing excessive errors from affecting the fairness of water billing. Attached Figure Description

[0018] Figure 1 A flowchart illustrating an error verification method for a smart water meter provided in an embodiment of the present invention; Figure 2 A system block diagram of an error verification system for a smart water meter provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] like Figure 1 The diagram shown is a flowchart of an error verification method for a smart water meter provided in an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S5 are detailed as follows: S1, Collect multimodal monitoring data from the smart water meter, wherein the multimodal monitoring data includes at least water flow acoustic signals, mechanical vibration signals, and flow reference data; Among them, the water flow acoustic signal is the acoustic information generated when water flows through the internal components of the smart water meter during operation. Its frequency and amplitude characteristics are related to the mechanical wear and water flow status of the smart water meter. The mechanical vibration signal is the vibration waveform data generated by the mechanical motion of the internal mechanism and component friction, which directly reflects the operating status of the smart water meter's mechanical structure. The flow reference data is the actual water flow benchmark data in the smart water meter metering scenario, complementing the acoustic and vibration signals.

[0021] In practical applications, multimodal monitoring data acquisition is achieved through adapted sensor components. Acoustic signals of water flow are captured by acoustic sensors deployed near the water flow channel of the smart water meter body. Mechanical vibration signals are acquired by vibration sensors attached to the outer shell of the smart water meter's internal mechanism. Flow reference data is synchronously acquired by a high-precision flow sensor linked to the smart water meter's metering channel, and the acquisition sequence of all three must be aligned in real-time with the actual operating status of the smart water meter. During the acquisition process, the influence of interference factors such as pipeline environmental noise and external vibrations on the data is avoided to ensure that the acquired data accurately reflects the operating characteristics of the smart water meter.

[0022] S2, perform acoustic signal preprocessing on the water flow acoustic signal and the mechanical vibration signal to generate an acoustic fusion signal, extract the time-frequency domain features of the acoustic fusion signal that are highly correlated with the smart water meter error, and screen and construct the optimal acoustic error feature set; The process of preprocessing the acoustic signals of the water flow and the mechanical vibration to generate an acoustic fusion signal, extracting time-frequency domain features highly correlated with the smart water meter error from the acoustic fusion signal, and screening to construct an optimal acoustic error feature set specifically includes: The acoustic signal preprocessing includes eliminating pipe background noise and electromagnetic interference noise in the water flow acoustic signal and the mechanical vibration signal based on a wavelet threshold denoising algorithm, and then adjusting the denoised water flow acoustic signal and the mechanical vibration signal to the same time dimension through timestamp alignment technology. The water flow acoustic signal and the mechanical vibration signal are fused using an attention-based weighted fusion algorithm. The weights are dynamically assigned according to the contribution of the water flow acoustic signal and the mechanical vibration signal to the error of the smart water meter under different flow levels, and the acoustic fusion signal is generated. A complementary combination of short-time Fourier transform and wavelet packet decomposition is used to extract time-frequency domain features from the acoustic fusion signal. The time-frequency domain features include center frequency, energy entropy, wavelet packet node energy, and short-time energy variance. The correlation coefficient between the time-frequency domain features and the smart water meter error is calculated using the ReliefF algorithm. Time-frequency domain features with correlation coefficients greater than a preset correlation coefficient threshold are retained to generate a preliminary acoustic error feature set. The importance of the time-frequency domain features in the preliminary acoustic error feature set is ranked using the random forest algorithm. Time-frequency domain features in the preliminary acoustic error feature set that meet the preset feature quantity requirements are retained, and finally, the optimal acoustic error feature set is constructed.

[0023] Among them, the wavelet threshold denoising algorithm is used to eliminate interference noise in water flow acoustic signals and mechanical vibration signals. Its core is to decompose the original signal into wavelet coefficients of different scales through wavelet transform, set a reasonable threshold for the noisy wavelet coefficients, process them, and then reconstruct the signal through inverse wavelet transform, thereby filtering out pipeline background noise and electromagnetic interference noise and ensuring signal purity.

[0024] Timestamp alignment technology is used to address the timing discrepancy between the acquisition of acoustic signals from water flow and mechanical vibration signals. Since the two types of signals are acquired by different sensors, there may be a time synchronization issue. This technology calibrates the timestamps of the two types of signals after denoising to the same time dimension, avoiding fusion deviations caused by timing misalignment.

[0025] The attention-based weighted fusion algorithm dynamically assigns weights. It analyzes the contribution of acoustic and mechanical vibration signals to the smart water meter's error by considering the operating characteristics of different flow rates. For example, at low flow rates, mechanical vibration signals better reflect errors caused by movement wear, while at high flow rates, acoustic signals better reflect the influence of water flow conditions on the error. Signals with high contribution are assigned higher weights, and signals with low contribution are assigned lower weights. This weighted calculation generates a fused acoustic signal, allowing it to more accurately focus on error-related characteristics.

[0026] The complementary combination of Short-Time Fourier Transform (SFT) and Wavelet Packet Decomposition (WPD) is used to extract the time-frequency domain features of acoustic fusion signals. SFT segments the acoustic fusion signal through a sliding time window, clearly reflecting its frequency distribution across different time windows. WPD performs equal and refined decomposition of the high-frequency and low-frequency bands of the acoustic fusion signal, capturing detailed features across different frequency bands. The combination of these two methods covers the time-frequency information of the acoustic fusion signal from low to high frequencies and from the overall picture to the details, avoiding the limitations of single-method feature extraction.

[0027] In the time-frequency domain characteristics, the center frequency refers to the frequency point where the energy of the acoustic fusion signal is concentrated, reflecting the main frequency characteristics of the acoustic fusion signal. For example, when the wear of the smart water meter's movement intensifies, the center frequency of the acoustic fusion signal may shift. Energy entropy describes the degree of disorder in the energy distribution of the acoustic fusion signal. The higher the energy entropy, the more dispersed the signal energy distribution, indirectly reflecting the instability of the smart water meter's operating state. Wavelet packet node energy is the energy value corresponding to each node after wavelet packet decomposition, reflecting the energy changes of signals in different frequency bands. For example, an abnormal increase in energy in a specific frequency band may be related to increased friction in components. Short-time energy variance reflects the amplitude of energy fluctuations in the signal over a short period of time and is directly related to the stability of the smart water meter's operation.

[0028] The ReliefF algorithm is used to filter time-frequency domain features highly correlated with smart water meter errors. It calculates the contribution of each time-frequency domain feature to the error category (correlation coefficient), retains features with correlation coefficients greater than a preset threshold, and removes irrelevant or weakly correlated features, generating a preliminary acoustic error feature set and reducing feature redundancy. The Random Forest algorithm is used to rank the importance of time-frequency domain features in the preliminary acoustic error feature set. By constructing multiple decision trees, it quantifies feature importance based on the influence of time-frequency domain features on the classification results of the decision trees, retaining a preset number of important preceding features, and finally constructing a concise, efficient, and strongly correlated optimal acoustic error feature set.

[0029] Through the above methods, the purification, synergistic fusion, and precise extraction and optimization of time-domain features of water flow acoustic signals and mechanical vibration signals were effectively achieved.

[0030] S3, based on the optimal acoustic error feature set and the preset standard acoustic test feature library, perform acoustic pattern recognition, evaluate the operating status of the smart water meter in real time, and generate preliminary results of smart water meter error. The process of performing acoustic pattern recognition based on the optimal acoustic error feature set and the preset standard acoustic verification feature library, real-time evaluation of the smart water meter's operating status, and generation of preliminary smart water meter error results specifically includes: The acoustic fingerprint of the smart water meter is extracted based on the optimal acoustic error feature set, and the acoustic fingerprint is normalized. From the preset standard acoustic verification feature library, a standard acoustic fingerprint template that matches the model and service life of the smart water meter is matched. The standard acoustic fingerprint template is associated with the standard error range and normal acoustic feature threshold of the corresponding flow point in the smart water meter. The similarity value between the acoustic fingerprint and the standard acoustic fingerprint template is calculated using a dynamic time warping algorithm. Calculate the Mahalanobis distance between the acoustic fingerprint and the standard acoustic fingerprint template. If the Mahalanobis distance is greater than the normal acoustic feature threshold, then determine that the smart water meter has an abnormal operating state and generate an abnormal state determination result. Based on the mapping relationship between the similarity value and the standard error interval, and combined with the abnormal operating status determination result, the operating status of the smart water meter is evaluated in real time, and a preliminary result of the smart water meter error is generated.

[0031] Acoustic fingerprints are feature sets extracted from the optimal acoustic error feature set, characterizing the unique acoustic properties of smart water meters. They contain acoustic information related to errors in the current operating state of the smart water meter, and the acoustic fingerprints of smart water meters differ significantly under different error states. Normalization is used to eliminate the numerical scale bias of acoustic fingerprints caused by different acquisition scenarios. By mapping feature values ​​to a unified dimension range, it ensures that acoustic fingerprints are comparable to standard acoustic fingerprint templates.

[0032] The preset standard acoustic verification feature library is a pre-built reference benchmark library that stores standard acoustic fingerprint templates for smart water meters of different models and service years. It also associates the standard error range of each template's corresponding flow point with the normal acoustic feature threshold. The standard error range represents the reasonable range of error at each flow point during normal operation of the smart water meter. The normal acoustic feature threshold is the acoustic feature boundary value for determining whether the smart water meter is operating normally. The standard acoustic fingerprint templates are standard samples in the preset standard acoustic verification feature library that perfectly match the model and service year of the smart water meter to be verified, and are constructed based on a large amount of normal operation data from similar smart water meters.

[0033] The dynamic time warping algorithm is used to calculate the similarity value between the actual acoustic fingerprint and the standard acoustic fingerprint template, taking into account the possible time axis offset between the two types of signals. For example, the time sequence misalignment of the water flow acoustic signal caused by the instantaneous operating speed fluctuation of the smart water meter can be achieved by dynamically adjusting the time axis to align the signals, thereby accurately calculating the similarity between the two and avoiding misjudgment of similarity caused by time misalignment.

[0034] Mahalanobis distance is used to measure the distance between an actual acoustic fingerprint and a standard acoustic fingerprint template in a multi-feature space. It eliminates interference from correlations between features and more accurately reflects the degree of difference between the two feature sets compared to ordinary distance calculations. Comparing it to a normal acoustic feature threshold, if it exceeds the normal acoustic feature threshold, it indicates that the current acoustic characteristics of the smart water meter deviate from the normal range, and is directly judged as an abnormal operating state. The normal acoustic feature threshold is an upper limit of Mahalanobis distance determined based on statistical analysis of a large amount of normal operating data from similar smart water meters, and is the key boundary distinguishing between normal and abnormal operation of a smart water meter.

[0035] Understandably, a higher similarity value indicates a stronger match between the current acoustic fingerprint and the standard acoustic fingerprint template, corresponding to a lower error range within the standard error interval. If the Mahalanobis distance exceeds the normal acoustic feature threshold, it is considered an abnormal operation. Even if the similarity value maps to the lower error range, the operating status still needs to be assessed as an "abnormal associated error state," and the abnormal attribute should be simultaneously marked in the preliminary results of the smart water meter error to ensure the completeness of the operating status assessment and the objectivity of the preliminary results of the smart water meter error.

[0036] S4. Construct a dynamic error prediction and verification model for smart water meters. Combine the optimal acoustic error feature set with the flow reference data to verify and analyze the preliminary error results of the smart water meters. Output the smart water meter error verification results, including the smart water meter error level and the smart water meter error prediction trend. The construction of the smart water meter dynamic error prediction and verification model, combined with the optimal acoustic error feature set and the flow reference data, verifies and analyzes the preliminary error results of the smart water meter, and outputs the smart water meter error verification results, including the smart water meter error level and the smart water meter error prediction trend, specifically including: Extract the traffic statistical features from the traffic reference data, and fuse the traffic statistical features with the optimal acoustic error feature set through feature concatenation to generate a multi-dimensional verification input vector. The dynamic error prediction and verification model for the smart water meter is constructed, which includes a basic prediction layer and a verification and correction layer. The basic prediction layer deploys a support vector regression model and a gradient boosting tree model in parallel. The multi-dimensional verification input vector is input into both the support vector regression model and the gradient boosting tree model, and the preliminary error result of the smart water meter is used as the initial supervision signal. The preliminary prediction error values ​​of the support vector regression model and the gradient boosting tree model are output respectively. The verification correction layer introduces an attention mechanism, dynamically allocating the output weights of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistics features to the smart water meter error. Based on the preliminary prediction error value and the corresponding output weights, a fused prediction error value is output. The fusion prediction error value is matched with a preset error level range to determine the error level of the smart water meter; Historical fusion prediction error values ​​are extracted based on an adaptive sliding time window. Combined with these fusion prediction error values, a multi-dimensional time-series trend fitting algorithm is used to generate the smart water meter error prediction trend. The smart water meter error level and the smart water meter error prediction trend are then summarized, and the smart water meter error verification result is output.

[0037] Among them, flow statistics features are characteristics extracted from flow reference data that characterize the flow operation pattern, such as average flow, instantaneous peak flow, and time-period fluctuation range, reflecting the impact of different flow conditions on smart water meter errors. The feature splicing method integrates the flow statistics features and the optimal acoustic error feature set sequentially according to feature dimensions, so that the two types of error-related features form a unified input structure, avoiding information loss caused by feature fragmentation, and thus generating a multi-dimensional verification input vector.

[0038] The basic prediction layer employs a parallel deployment of Support Vector Regression (SVR) and Gradient Boosting Tree (GPRS) models, fully leveraging the advantages of both. The SVR model uses a kernel function to achieve a non-linear mapping of multi-dimensional verification input vectors, adapting to the non-linear characteristics of smart water meter errors varying with operating conditions. The GPRS model effectively captures complex interactions between features by iteratively constructing weak learners and weighted fusion. The parallel computation of both models covers different error impact scenarios, reducing the prediction bias of a single model. The initial supervision signal, i.e., the preliminary result of the smart water meter error, provides a prediction benchmark for the basic prediction layer, improving the reasonableness of the initial prediction error value.

[0039] The verification and correction layer introduces an attention mechanism, which dynamically adjusts the output weights of the two basic models based on the contribution of the optimal acoustic error feature set and the flow statistics features to the error. The fusion prediction error value is obtained through weighted calculation, and this value is closer to the actual error situation than the result of a single model.

[0040] The preset error level range is a pre-defined error range based on the metering accuracy standards of smart water meters. By matching the fused prediction error value with the preset error level range, the error level of the smart water meter is determined. An adaptive sliding time window is used to dynamically extract historical fused prediction error values. The window duration is adjusted according to the error fluctuation frequency to ensure data timeliness. A multi-dimensional time-series trend fitting algorithm fits curves to historical and fused prediction error values, removes outliers, captures the error change pattern, and generates a smart water meter error prediction trend that includes the error change rate and future extreme values.

[0041] In practical applications, extracting flow statistics features needs to be adapted to the actual water usage conditions of smart water meters to ensure the capture of flow patterns under different time periods and water usage intensities. When concatenating features, the dimensional specifications of the flow statistics features and the optimal acoustic error feature set are unified to avoid format differences that could cause multi-dimensional verification input vectors to fail. The support vector regression model and gradient boosting tree model of the basic prediction layer are deployed in parallel, and the computational logic is optimized based on the hardware performance of the smart water meter's edge computing module to prevent excessive resource consumption from affecting real-time verification. The contribution calculation of the attention mechanism in the verification correction layer adjusts parameters according to the flow level characteristics of different water usage scenarios, such as residential and industrial use. The preset error level range is set with reference to industry water meter metering accuracy standards to ensure compliance in error level determination. The adaptive sliding time window duration is dynamically adapted based on the fluctuation frequency of historical error data, combined with a multi-dimensional time-series trend fitting algorithm to ensure the timeliness of error prediction trends, ultimately outputting smart water meter error verification results that meet actual verification requirements.

[0042] The multidimensional verification input vector X is input into the support vector regression model and the gradient boosting tree model, and the preliminary error result of the smart water meter is used as the initial supervision signal. Output the preliminary prediction error values ​​of the support vector regression model respectively. The initial prediction error value of the gradient boosting tree model The corresponding calculation formula is as follows: In the formula, Represents the Lagrange multiplier in a support vector regression model; This represents the kernel function in a support vector regression model; represents the i-th multidimensional validation input vector in the training samples of the support vector regression model; b represents the bias term in the support vector regression model; N represents the number of training samples in the support vector regression model; C represents the regularization coefficient in the support vector regression model; This represents the mean vector of the multidimensional validation input vectors in the training samples of the support vector regression model. Denotes the Euclidean norm; This represents the learning rate in a gradient boosting tree model; This represents the predicted output of the t-th decision tree in the gradient boosting tree model for the multidimensional validation input vector X; T represents the total number of decision trees in the gradient boosting tree model. This represents the initial supervision signal in the gradient boosting tree model. The weighting adjustment coefficient; This represents the natural exponential function with base e.

[0043] Among them, the Lagrange multiplier is used to quantify the influence weight of the multidimensional verification input vector in the training samples on the prediction results of the support vector regression model, thereby improving the model's ability to fit the nonlinear correlation of smart water meter errors. The kernel function is used to realize the nonlinear mapping of the multidimensional verification input vector, mapping the error correlation features that are difficult to distinguish linearly in the low-dimensional feature space to the high-dimensional space to construct a linear decision boundary, adapting to the nonlinear characteristics of smart water meter errors as they change with operating conditions.

[0044] The bias term adjusts the prediction baseline of the support vector regression model, ensuring that the prediction results more closely match the actual error range of smart water meters. The regularization coefficient controls the complexity of the support vector regression model, balancing feature fitting accuracy with the risk of overfitting, preventing the model from over-relying on noise information in the training samples, and ensuring adaptability to different smart water meter operating scenarios. The mean vector of the multidimensional validation input vector is used to calculate the deviation between the multidimensional input vector to be validated and the training sample vectors, providing a standardized distance metric for the kernel function and optimizing the nonlinear mapping effect.

[0045] The learning rate is used to adjust the contribution strength of each decision tree in the gradient boosting tree model. By limiting the weight proportion of a single tree, it prevents the gradient boosting tree model from rapidly converging to a local optimum, thus improving the stability of error prediction. The decision tree prediction output is the error prediction result of a single decision tree on a multi-dimensional verification input vector. The iterative fusion of multiple trees gradually reduces prediction bias and accurately captures the complex interaction relationships of error-related features. The weight adjustment coefficient is used to dynamically adapt the influence of the initial supervision signal, so that the prediction of the gradient boosting tree model both references historical verification experience and conforms to real-time operating conditions.

[0046] By relying on the Lagrange multipliers and kernel function of the support vector regression model to adapt to the nonlinear correlation of smart water meter errors, and combining the learning rate and number of decision trees of the gradient boosting tree model to ensure prediction stability, the preliminary prediction error values ​​of the two models are output using the preliminary results of smart water meter errors as the initial supervision signal.

[0047] The verification and correction layer introduces an attention mechanism, dynamically allocating the output weights of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistics features to the smart water meter error, and outputting a fused prediction error value based on the preliminary prediction error value and the corresponding output weights, specifically including: The contribution of the optimal acoustic error feature set and the flow statistics feature to the error of the smart water meter is determined by the information gain algorithm. , And satisfy The corresponding calculation formula is as follows: In the formula, Represents the optimal acoustic error feature set Information gain value; Indicates flow statistics characteristics Information gain value; The information entropy represents the error result U of the smart water meter; The error result U of the smart water meter is based on the optimal acoustic error feature set. Conditional information entropy; The error result U of the smart water meter is based on flow statistics characteristics. Conditional information entropy; Based on the contribution , Dynamically allocate the output weights of the support vector regression model , , This represents the feature contribution adjustment coefficient; based on the output weights of the support vector regression model. With the output weights of the gradient boosting tree model , The preliminary prediction error values ​​output by the support vector regression model and the gradient boosting tree model are weighted and fused to output a fused prediction error value R.

[0048] The information gain algorithm calculates the degree of uncertainty reduction in the smart water meter error result before and after the introduction of the optimal acoustic error feature set and flow statistics features, i.e., the information gain value, which objectively reflects the feature's ability to represent the error. The information gain value of the optimal acoustic error feature set directly reflects the strength of the acoustic feature's influence on the error, while the information gain value of the flow statistics features reflects the degree of error correlation of the flow features. Together, they determine the importance ratio of the two types of features in error prediction.

[0049] Information entropy is an indicator that measures the uncertainty of error results from smart water meters. The more dispersed the error results are and the higher the uncertainty, the larger the information entropy value. Conditional information entropy is the residual uncertainty of the error results given an optimal set of acoustic error features or flow statistics.

[0050] The feature contribution adjustment coefficient is used to fine-tune the impact of feature contribution on the output weights of the support vector regression model and the gradient boosting tree model. It is dynamically adjusted according to the actual difference in the role of the optimal acoustic error feature set and the flow statistics feature in the actual application scenario of smart water meters, so as to avoid the weight deviation caused by a single fixed allocation logic.

[0051] By employing the above method, the contribution of the optimal acoustic error feature set and flow statistics features to the smart water meter error is quantified using the information gain algorithm, ensuring the objectivity of feature importance assessment. Furthermore, based on the contribution of the two types of features, the output weights of the support vector regression model are dynamically adapted using a feature contribution adjustment coefficient, avoiding excessive dominance of a single feature or model in the prediction results. Finally, the preliminary prediction error values ​​of the two models are weighted and fused. The resulting fused prediction error value fully integrates the advantages of both models and dual-dimensional feature information, better reflecting the actual error state of the smart water meter.

[0052] The step of extracting historical fusion prediction error values ​​based on an adaptive sliding time window, and combining these fusion prediction error values ​​with a multi-dimensional time-series trend fitting algorithm to generate the smart water meter error prediction trend, specifically includes: An adaptive sliding time window is constructed. The window duration of the adaptive sliding time window is dynamically adjusted according to the fluctuation frequency of the historical fusion prediction error value. After filtering out abnormal error data that exceed the preset deviation range based on the 3σ principle, the continuous historical fusion prediction error values ​​within the window are extracted. Using the multi-dimensional time-series trend fitting algorithm, with the smart water meter error level as the fitting weight, curve fitting is performed on the extracted historical fusion prediction error value and the fusion prediction error value. Simultaneously, extreme outliers in the fitted data are filtered out based on the standard deviation to generate the smart water meter error prediction trend that includes the error change rate and error extreme values.

[0053] The 3σ principle, based on the normal distribution of data, classifies values ​​exceeding the sum of the historical fusion prediction error mean and three times the standard deviation, or falling below the difference between the mean and three times the standard deviation, as anomalous error data. The multi-dimensional time-series trend fitting algorithm considers both the time dimension and the error correlation feature dimension of the error data during the fitting process. It captures the changing patterns of smart water meter error levels over time through curve fitting, using the smart water meter error level as the fitting weight. Higher smart water meter error levels result in a larger weight for the corresponding error data in the fitting, ensuring that the fitted trend primarily reflects the changing patterns of high-risk error states and improving the trend's adaptability to key error scenarios.

[0054] Extreme outliers are individual data points that deviate from the overall error trend after curve fitting. They may be caused by interference factors not fully filtered out during the fitting process. These points are eliminated by screening using the standard deviation to prevent distortion of the fitted curve and ensure the accuracy and reliability of the generated smart water meter error prediction trend. The error change rate reflects the magnitude of error increase or decrease per unit time, used to determine how fast the error develops. Error extremes are the maximum or minimum values ​​that the error may reach in future periods.

[0055] In practical applications, taking the error control of smart water meters in residential communities as an example, during the morning peak water usage period, frequent water flow switching leads to increased fluctuations in historical fusion prediction error values. The adaptive sliding time window automatically shortens its duration, while the 3σ principle filters out abnormal error data caused by sudden pipeline vibrations, ensuring that the extracted historical fusion prediction error values ​​closely match the real-time error status. During the fitting phase, if the smart water meter error level is close to the preset error threshold for a certain period, the error data for that period is given increased weight in the multi-dimensional time series trend fitting. After removing extreme outliers by combining the standard deviation, the predicted trend of smart water meter errors for a future period is accurately generated.

[0056] S5. When the error level of the smart water meter reaches the preset error level threshold, the early warning mechanism is automatically triggered and an early warning message is pushed.

[0057] In practical applications, when the error level of a smart water meter reaches a preset error level threshold, the system first automatically captures key information such as the smart water meter's unique serial number, installation location, current error value, and preset error level threshold, integrating it into structured early warning content. Subsequently, the warning information is pushed through multiple channels, such as to maintenance personnel responsible for the area, clearly informing them of the smart water meter information requiring priority verification and the error exceeding the standard. Simultaneously, the warning information also synchronizes historical error data fragments from associated smart water meters, providing maintenance personnel with a reference for preliminary judgment of the error cause, ensuring that management and maintenance personnel can respond quickly and preventing the continued expansion of errors from affecting metering fairness.

[0058] like Figure 2 The diagram shown is a system block diagram of an error verification system for a smart water meter according to an embodiment of the present invention. The system includes: The data acquisition module is used to collect multimodal monitoring data from the smart water meter. The multimodal monitoring data includes at least water flow acoustic signals, mechanical vibration signals, and flow reference data. The feature extraction module is used to perform acoustic signal preprocessing on the water flow acoustic signal and the mechanical vibration signal to generate an acoustic fusion signal, extract time-frequency domain features that are highly correlated with the smart water meter error of the acoustic fusion signal, and screen and construct the optimal acoustic error feature set. The error generation module is used to perform acoustic pattern recognition based on the optimal acoustic error feature set and the preset standard acoustic test feature library, to evaluate the operating status of the smart water meter in real time and generate preliminary results of the smart water meter error. The error verification module is used to construct a dynamic error prediction and verification model for smart water meters. Combining the optimal acoustic error feature set and the flow reference data, it verifies and analyzes the preliminary error results of the smart water meters and outputs the smart water meter error verification results, including the smart water meter error level and the smart water meter error prediction trend. The early warning triggering module is used to automatically trigger the early warning mechanism and push early warning information when the error level of the smart water meter reaches a preset error level threshold.

[0059] Figure 2 The apparatus of the illustrated embodiment can be used to perform corresponding actions. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0060] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor performs the steps of an error verification method for a smart water meter as described in any of the above claims.

[0061] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein... The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0062] Processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0063] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0064] When the memory 32 is a device independent of the processor 31, the device may further include: Bus 33 is used to connect the memory 32 and the processor 31.

[0065] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of an error verification method for a smart water meter as described in any of the above claims.

[0066] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0067] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0068] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0069] Through the above embodiments, this invention collects multimodal monitoring data from smart water meters, including at least water flow acoustic signals, mechanical vibration signals, and flow reference data. It preprocesses the water flow acoustic signals and mechanical vibration signals to generate an acoustic fusion signal, extracts time-frequency domain features highly correlated with smart water meter errors from the acoustic fusion signal, and constructs an optimal acoustic error feature set. Based on the optimal acoustic error feature set and a preset standard acoustic verification feature library, it performs acoustic pattern recognition to evaluate the smart water meter's operating status in real time and generate preliminary smart water meter error results. It constructs a dynamic error prediction and verification model for smart water meters, combines the optimal acoustic error feature set and flow reference data to verify and analyze the preliminary smart water meter error results, and outputs smart water meter error verification results including the smart water meter error level and the smart water meter error prediction trend. When the smart water meter error level reaches a preset error threshold, it automatically triggers an early warning mechanism and pushes early warning information. This enables real-time evaluation of the smart water meter's metering performance, dynamic prediction of error change trends, effectively ensuring the timeliness and effectiveness of error verification, and fully meeting the dynamic control requirements of smart water meter errors.

[0070] This invention acquires multimodal monitoring data to capture acoustic signals of water flow, mechanical vibration signals, and flow reference data during the operation of smart water meters, comprehensively capturing key status information during smart water meter operation and avoiding the limitations of verification under a single data dimension. In the signal processing stage, this invention uses a wavelet threshold denoising algorithm to eliminate pipeline background noise and electromagnetic interference, combined with timestamp alignment technology and an attention mechanism weighted fusion algorithm to generate an acoustic fusion signal, significantly improving signal purity and fusion effectiveness. Then, it extracts time-frequency domain features through complementary extraction of short-time Fourier transform and wavelet packet decomposition, and uses the ReliefF algorithm and random forest algorithm to screen and construct an optimal acoustic error feature set, ensuring high correlation between features and errors. This invention achieves real-time evaluation of the smart water meter's operating status and generates preliminary results of smart water meter errors through acoustic pattern recognition based on the optimal acoustic error feature set and a preset standard acoustic verification feature library. By normalizing acoustic fingerprints, calculating similarity using dynamic time warping algorithms, and determining Mahalanobis distance, it solves the problem of real-time monitoring difficulties in traditional methods. This invention constructs a dynamic error prediction and verification model. The basic prediction layer uses a support vector regression model and a gradient boosting tree model deployed in parallel to achieve initial prediction. The verification and correction layer dynamically allocates weights and outputs a fused prediction error value by introducing an attention mechanism. Combined with an adaptive sliding time window and a multi-dimensional time-series trend fitting algorithm, it generates a prediction trend for smart water meter errors. This solves the problem of inaccurate predictions in traditional linear models, improves prediction accuracy, and precisely determines the error level. Furthermore, this invention employs an automatic early warning mechanism to promptly push warning information when the smart water meter error level reaches a preset error threshold, ensuring metering accuracy and preventing excessive errors from affecting the fairness of water billing.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for error checking of a smart water meter, characterized by, The method comprises: Collecting multi-modal monitoring data of the intelligent water meter, the multi-modal monitoring data at least comprising water flow acoustic signals, mechanical vibration signals, and flow reference data; Performing acoustic signal preprocessing on the water flow acoustic signals and the mechanical vibration signals to generate acoustic fusion signals, extracting time-frequency domain features of the acoustic fusion signals highly correlated with errors of the intelligent water meter, and screening and constructing an optimal acoustic error feature set; Performing acoustic pattern recognition based on the optimal acoustic error feature set and a preset standard acoustic verification feature library, real-time evaluating an operating state of the intelligent water meter, and generating an initial result of errors of the intelligent water meter; Constructing a dynamic error prediction and verification model of the intelligent water meter, combining the optimal acoustic error feature set and the flow reference data, verifying and analyzing the initial result of errors of the intelligent water meter, and outputting an error verification result of the intelligent water meter including an error level of the intelligent water meter and a prediction trend of errors of the intelligent water meter; When the error level of the intelligent water meter reaches a preset error level threshold, triggering an early warning mechanism automatically and pushing early warning information.

2. The error checking method of a smart water meter according to claim 1, wherein, The acoustic signal preprocessing comprises eliminating pipeline background noise and electromagnetic interference noise in the water flow acoustic signals and the mechanical vibration signals based on a wavelet threshold denoising algorithm, and adjusting the denoised water flow acoustic signals and the mechanical vibration signals to the same time dimension through a time stamp alignment technology. The water flow acoustic signals and the mechanical vibration signals are fused and processed based on an attention mechanism weighted fusion algorithm, weights are dynamically allocated according to contributions of the water flow acoustic signals and the mechanical vibration signals to errors of the intelligent water meter under different flow ranges of the intelligent water meter, and the acoustic fusion signals are generated. Time-frequency domain features are extracted from the acoustic fusion signals by using a method of complementary combination of short-time Fourier transform and wavelet packet decomposition, the time-frequency domain features comprising a center frequency, an energy entropy, a wavelet packet node energy, and a short-time energy variance. Correlation coefficients of the time-frequency domain features and the errors of the intelligent water meter are calculated through a ReliefF algorithm, the time-frequency domain features with correlation coefficients greater than a preset correlation coefficient threshold are retained to generate a preliminary screening acoustic error feature set, the time-frequency domain features in the preliminary screening acoustic error feature set are sorted in importance by using a random forest algorithm, the time-frequency domain features in the preliminary screening acoustic error feature set meeting a preset feature quantity requirement are retained, and finally the optimal acoustic error feature set is constructed. The acoustic pattern recognition based on the optimal acoustic error feature set and the preset standard acoustic verification feature library, and the real-time evaluation of the operating state of the intelligent water meter and the generation of the initial result of errors of the intelligent water meter, specifically comprise:

3. The error checking method of a smart water meter according to claim 1, wherein, Extracting acoustic fingerprints of the intelligent water meter based on the optimal acoustic error feature set, and performing normalization processing on the acoustic fingerprints; ​ From the preset standard sound school test feature library, the standard acoustic fingerprint template consistent with the model and service life of the intelligent water meter is matched, and the standard acoustic fingerprint template is associated with the standard error interval and the normal acoustic feature threshold of the corresponding flow point in the intelligent water meter; The similarity value of the acoustic fingerprint and the standard acoustic fingerprint template is calculated by using a dynamic time warping algorithm; The Mahalanobis distance of the acoustic fingerprint and the standard acoustic fingerprint template is calculated, and if the Mahalanobis distance is greater than the normal acoustic feature threshold, it is determined that the intelligent water meter has an abnormal running state and an abnormal state judgment result is generated; According to the mapping relationship between the similarity value and the standard error interval, the running state of the intelligent water meter is evaluated in real time in combination with the abnormal running state judgment result, and an initial error result of the intelligent water meter is generated.

4. The error checking method of a smart water meter according to claim 1, wherein, The intelligent water meter dynamic error prediction and verification model is constructed, the initial error result of the intelligent water meter is verified and analyzed in combination with the optimal acoustic error feature set and the flow reference data, and an intelligent water meter error verification result including an intelligent water meter error level and an intelligent water meter error prediction trend is output, specifically including: The flow statistical features of the flow reference data are extracted, the flow statistical features and the optimal acoustic error feature set are fused through feature splicing, and a multi-dimensional verification input vector is generated; The intelligent water meter dynamic error prediction and verification model is constructed, and the intelligent water meter dynamic error prediction and verification model includes a basic prediction layer and a verification correction layer; Wherein, the basic prediction layer is arranged in parallel with a support vector regression model and a gradient boosting tree model, the multi-dimensional verification input vector is input into the support vector regression model and the gradient boosting tree model, the initial supervision signal is taken as the initial error value of the support vector regression model and the gradient boosting tree model, and the initial prediction error value of the support vector regression model and the gradient boosting tree model is output respectively; the verification correction layer introduces an attention mechanism, dynamically allocates the output weight of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistical features to the intelligent water meter error, and outputs a fusion prediction error value based on the initial prediction error value and the corresponding output weight; The fusion prediction error value is matched with a preset error level interval to determine the intelligent water meter error level; Based on the adaptive sliding time window, the historical fusion prediction error value is extracted, the fusion prediction error value is combined, and the intelligent water meter error prediction trend is generated through a multi-dimensional time series trend fitting algorithm, the intelligent water meter error level and the intelligent water meter error prediction trend are summarized, and the intelligent water meter error verification result is output.

5. The error checking method of a smart water meter according to claim 4, wherein, inputting the multi-dimensional check input vector X into the support vector regression model and the gradient boosting tree model, and taking the smart water meter error preliminary result as an initial supervision signal , respectively outputting the preliminary prediction error value of the support vector regression model and the preliminary prediction error value of the gradient boosting tree model , and the corresponding calculation formulas are as follows: wherein, denotes the Lagrange multiplier in the support vector regression model; denotes the kernel function in the support vector regression model; denotes the i-th multi-dimensional check input vector in the training sample of the support vector regression model; b denotes the bias term in the support vector regression model; N denotes the number of training samples of the support vector regression model; C denotes the regularization coefficient in the support vector regression model; denotes the mean vector of the multi-dimensional check input vectors in the training sample of the support vector regression model; denotes the Euclidean norm; denotes the learning rate in the gradient boosting tree model; denotes the predicted output of the t-th decision tree in the gradient boosting tree model for the multi-dimensional check input vector X; T denotes the total number of decision trees in the gradient boosting tree model; denotes the weight adjustment coefficient of the initial supervisory signal in the gradient boosting tree model; denotes the natural exponential function with base e.

6. The error checking method of a smart water meter according to claim 4, wherein, The verification correction layer introduces an attention mechanism, dynamically allocates the output weight of the support vector regression model and the gradient boosting tree model based on the contribution of the optimal acoustic error feature set and the flow statistical features to the intelligent water meter error, and outputs a fusion prediction error value based on the initial prediction error value and the corresponding output weight, specifically including: Contribution degrees of the optimal acoustic error feature set and the flow statistical feature to the error of the smart water meter are determined by an information gain algorithm , , and satisfy The corresponding calculation formula is as follows: wherein, represents the information gain value of the optimal acoustic error feature set ; represents the information gain value of the flow statistics feature ; represents the information entropy of the smart water meter error result U; represents the conditional information entropy of the smart water meter error result U based on the optimal acoustic error feature set ; represents the conditional information entropy of the smart water meter error result U based on the flow statistics feature ; based on the contribution degree , dynamically allocating output weights of the support vector regression model , , representing a feature contribution degree adjustment coefficient; according to the output weights of the support vector regression model and the output weights of the gradient boosting tree model , , the preliminary prediction error values respectively output by the support vector regression model and the gradient boosting tree model are weighted and fused to output a fusion prediction error value R.

7. The error checking method of a smart water meter according to claim 4, wherein, The adaptive sliding time window-based historical fusion prediction error value is extracted, the fusion prediction error value is combined, and the intelligent water meter error prediction trend is generated through a multi-dimensional time series trend fitting algorithm. Specifically, the adaptive sliding time window is constructed, the window length of the adaptive sliding time window is dynamically adjusted according to the fluctuation frequency of the historical fusion prediction error value, after abnormal error data exceeding a preset deviation range is filtered based on the 3σ principle, the continuous historical fusion prediction error values in the window are extracted; The multi-dimensional time series trend fitting algorithm is used, the intelligent water meter error level is used as a fitting weight, the extracted historical fusion prediction error values and the fusion prediction error values are curve fitted, extreme outliers in the fitting data are simultaneously screened and removed based on a standard deviation, and the intelligent water meter error prediction trend including an error change rate and an error extreme value is generated. The system comprises:

8. An error checking system of a smart water meter, applied to the error checking method of a smart water meter according to any one of claims 1-7, characterized in that, A data acquisition module is configured to acquire multi-modal monitoring data of an intelligent water meter, wherein the multi-modal monitoring data at least includes water flow acoustic signals, mechanical vibration signals, and flow reference data. A feature extraction module is configured to perform acoustic signal preprocessing on the water flow acoustic signals and the mechanical vibration signals to generate acoustic fusion signals, extract time-frequency domain features of the acoustic fusion signals that are highly correlated with intelligent water meter errors, and screen and construct an optimal acoustic error feature set. An error generation module is configured to perform acoustic pattern recognition based on the optimal acoustic error feature set and a preset standard acoustic verification feature library, to evaluate the running state of the intelligent water meter in real time and generate an intelligent water meter error preliminary result. An error verification module is configured to construct an intelligent water meter dynamic error prediction and verification model, to perform verification analysis on the intelligent water meter error preliminary result in combination with the optimal acoustic error feature set and the flow reference data, and to output an intelligent water meter error verification result including an intelligent water meter error level and an intelligent water meter error prediction trend. A warning triggering module is configured to automatically trigger a warning mechanism and push a warning information when the intelligent water meter error level reaches a preset error level threshold. When the processor runs the computer program stored in the memory, the processor performs the steps of the error verification method of the intelligent water meter according to any one of claims 1-7. 9.An electronic device comprising a memory and a processor, the memory having stored therein a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the error verification method of the intelligent water meter according to any one of claims 1-7.

10. A readable storage medium, in which a computer program is stored, characterized in that, ​

Images