Ultrasonic gas meter mixed signal processing method

Abstract

The invention relates to the field of gas meter signal processing, and discloses an ultrasonic gas meter mixed signal processing method which comprises the following steps: building an experimental platform for simulating the actual operation condition of a gas meter, and collecting ultrasonic original signals under different conditions through the platform to form an original signal data set; preprocessing and expanding the original signal data set, and dividing a training set, a verification set and a test set; training the training set by using a deep learning model to obtain a signal noise reduction model; when the gas meter runs, ultrasonic signals are collected in real time and input into the signal noise reduction model to filter interference signals, and pure signals are output. Through the step-by-step completion of each step, a full-process signal processing method including data acquisition, model training, real-time processing, working condition adaptation and continuous optimization is constructed, and the signal processing capability and metering precision of the ultrasonic gas meter under complex and dynamic working conditions are comprehensively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of gas meter signal processing, and specifically to a method for processing mixed signals of an ultrasonic gas meter. Background Technique

[0002] With the continuous progress of living standards, people are increasingly pursuing environmental protection, and conventional energy sources with serious pollution are gradually abandoned by people. With the construction and popularization of gas transmission pipelines, gas meters have emerged in large numbers. From mechanical diaphragm gas meters to electronic diaphragm gas meters, and from diaphragm gas meters to ultrasonic gas meters, new concepts and new technologies have emerged continuously. In the technical field of gas meter signal processing, accurate measurement of gas flow is crucial for energy management, cost settlement, and safety monitoring. With the development of technology, ultrasonic gas meters have gradually become a research hotspot due to their non-contact measurement, high-precision potential, and low maintenance requirements.

[0003] An ultrasonic gas meter mainly measures the flow rate based on the time difference between the downstream and upstream propagation times of ultrasonic waves in a gas medium. In actual operation, complex working conditions cause a large amount of noise and interference to be mixed in the ultrasonic signal. In the actual application of traditional methods for processing mixed signals of ultrasonic gas meters, although analog filtering technology can initially filter out noise in some frequency bands, it is difficult to accurately separate interference close to the signal frequency band. The component parameters are affected by the environment, resulting in fluctuations in the filtering characteristics and poor stability. Although digital filtering has flexibility and can adjust the filtering parameters through algorithms, sampling quantization introduces errors, and complex algorithms increase the computational burden and processing delay, challenging real-time performance. As the core of flow calculation, the time difference method has difficulty in accurately measuring the time difference of weak and distorted signals. Tiny timing errors cause significant measurement deviations at low flow rates, and a single fixed algorithm cannot adapt to changes in working conditions, resulting in large fluctuations in measurement accuracy and low reliability under different working conditions. Summary of the Invention

[0004] Aiming at the deficiencies of the prior art, the present invention provides a method for processing mixed signals of an ultrasonic gas meter, which solves the problems that the traditional method for processing mixed signals of an ultrasonic gas meter cannot adapt to changes in working conditions, resulting in large fluctuations in measurement accuracy and low reliability under different working conditions.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: A method for processing mixed signals of an ultrasonic gas meter includes the following steps: S1. Build an experimental platform that simulates the actual operating conditions of an analog gas meter, and collect ultrasonic original signals under different conditions through this platform to form an original signal dataset; S2. Preprocess and expand the original signal dataset, and divide it into a training set, a validation set, and a test set; S3. Use a deep learning model to train the training set to obtain a signal denoising model; S4. When the gas meter is running, ultrasonic signals are collected in real time, input into the signal denoising model to filter out interference signals, and pure signals are output; S5. The pure signals are analyzed by the method of wavelet transform and cross-correlation fusion to obtain the signal characteristics under small flow conditions, and the time differences of downstream and upstream propagation are obtained; S6. The working conditions of the gas meter are monitored in real time, and the algorithms in the algorithm library are dynamically switched and fused according to the working condition categories, and the signals are processed to obtain the measurement results; S7. The measurement results and the actual measurement accuracy are transmitted back to the cloud platform, and the algorithm parameters are adjusted based on the big data analysis of the cloud platform for continuous optimization.

[0006] Preferably, in S1, the experimental platform uses a programmable signal generator to output ultrasonic signals with a frequency range of 50 kHz - 1 MHz and an amplitude accuracy of ±0.01 V, and is equipped with a radio frequency transmitting device that simulates electromagnetic interference of 0 - 200 V / m. The interference frequency is 10 kHz - 1 GHz, and a mechanical vibration table that reproduces pipeline vibration of 5 - 2000 Hz is introduced. Combined with a temperature and humidity control system, the temperature range is controlled at -30°C - 70°C, and the humidity range is 10% - 95%RH. An analog gas pipeline is equipped, and an ultrasonic transducer with a sensitivity of -200 dBre1V / μPa and a data acquisition card with a sampling rate of 5 MHz and a 16-bit accuracy are used to collect the original signals. The single acquisition duration is not less than 30 s. The different conditions include electromagnetic interference intensity, pipeline vibration frequency, flow range, and temperature and humidity conditions.

[0007] Preferably, in S2, the preprocessing and expansion are to calculate the robust mean of the original signals using the robust statistical method and the robust standard deviation , and the signal amplitude is normalized to the range of [-1, 1] according to the formula , where is the value of the th sample point in the original signal sequence, is the value of the th sample point in the signal sequence after standardization processing. Data expansion adopts operations such as random flipping, adding noise, time shifting, and frequency modulation. When adding noise, the dynamic noise standard deviation is set according to the working condition simulation data, and the value range is 0.005 - 0.2. The training set, validation set, and test set are divided in the proportions of 75%, 15%, and 10% in sequence.

[0008] Preferably, in step S3, the training of the training set using the deep learning model is performed by using a deep convolutional neural network. The first layer uses 64 small 3×3 convolutional kernels, paired with the LeakyReLU activation function. Four convolutional modules are stacked in the middle layer, including convolution, instance normalization, LeakyReLU activation, and dilated convolution operations. The dilation rates are set to 1, 2, 3, and 4 in sequence. Dense connections are used between the modules. After adaptive average pooling at the end, a fully connected layer is connected, and the number of neurons is expanded to 256, paired with a Softmax classifier. The focal loss function and the adaptive moment estimation optimizer are selected. The initial learning rate is set to 0.002, and it decays by 0.2 times every 5 epochs. Iterative training is performed until the loss of the validation set does not decrease significantly for 8 consecutive epochs. The early stopping method and the learning rate annealing strategy are enabled, and the optimal model weights are saved.

[0009] Preferably, in step S5, the signal features under small flow conditions are analyzed as follows: Based on wavelet packet transform, the Symlets8 wavelet basis is selected to perform 4-level wavelet packet decomposition on the denoised signal to obtain multi-level detail components and approximation components, and the 3rd and 4th level high-frequency detail components are locked; the high-frequency detail components and the preset standard small flow signal template are subjected to cross-correlation operation, and the cross-correlation result is calculated according to the formula , where is the result of the optimized cross-correlation function, is the high-frequency detail component after wavelet packet transform, is the preset standard small flow signal template, is the index in the signal sequence, is the length of the signal sequence, is the exponential decay term, where is the decay coefficient, is the displacement. The cross-correlation window size is dynamically adjusted to be between 10 and 40 sampling points, and the step size is set to be between 0.5 and 3 sampling points. Combining the simulated annealing algorithm, the value corresponding to the maximum peak is locked to obtain the time differences of forward and reverse propagation .

[0010] Preferably, in step S6, the real-time monitoring of the gas meter working condition is carried out by monitoring the gas meter through a high-precision platinum resistance temperature sensor, a sapphire pressure sensor, and a gas sensor. After the sensor signals are amplified by a low-noise, high-gain conditioning circuit and band-pass filtered, they are connected to a 24-bit high-speed multi-channel A / D conversion chip. The sampling frequency is set to 2 kHz, and the conversion results are read in real time by a microprocessor. A hybrid working condition judgment algorithm based on decision tree and neural network is used to classify the working condition categories.

[0011] Preferably, the dynamic switching and fusion algorithm library in S6 includes Kalman filtering, Hilbert-Huang transform, empirical mode decomposition, and particle filtering. The processing signal obtains the final measurement result by predefining intelligent switching rules according to the fine category of working conditions, and calculates the fusion signal processing result according to the formula where is the dynamic adaptive weighting coefficient, is the final signal after dynamic switching and fusion algorithm processing, is the signal processed by the th algorithm in the algorithm library.

[0012] Preferably, in S7, the measurement result and the actual measurement accuracy are transmitted back to the cloud platform through the built-in 5G wireless module of the gas meter. The processing result, actual measurement accuracy, and working condition data are packaged in JSON format and transmitted back to the cloud platform every 10 minutes. The algorithm parameters are adjusted and continuously optimized based on the big data analysis of the cloud platform. The cloud platform stores the data in the distributed database after receiving and verifying it, analyzes the correlation between errors and working conditions according to the big data analysis framework, adjusts the parameters using genetic and gradient descent methods, initially sends an OTA push update instruction, and upgrades the algorithm during the low-peak gas usage period.

[0013] The present invention provides a method for processing mixed signals of an ultrasonic gas meter. It has the following beneficial effects: 1. Through the gradual completion of each step of the present invention, a full-process signal processing method covering data acquisition, model training, real-time processing, working condition adaptation, and continuous optimization is constructed, comprehensively improving the signal processing ability and measurement accuracy of the ultrasonic gas meter under complex and dynamic working conditions, and solving the problem that the traditional method for processing mixed signals of ultrasonic gas meters cannot adapt to changes in working conditions, with large fluctuations in measurement accuracy and low reliability under different working conditions.

[0014] 2. By using robust statistical methods to calculate the robust mean and robust standard deviation of the original signal and normalizing them, and implementing data augmentation operations, the present invention optimizes from the data source to ensure the efficient and stable operation of the model under complex working conditions. By constructing a customized deep convolutional neural network and supporting optimization strategies, the model is trained efficiently, accurately learns the complex mapping relationship between signals and interferences, constructs a signal denoising model, accurately identifies and efficiently filters out interference signals under various working conditions, significantly improving the purity and stability of ultrasonic signals, and effectively enhancing the operation reliability and measurement accuracy of ultrasonic gas meters.

[0015] 3. The present invention uses wavelet transform and cross-correlation fusion to analyze the characteristics of small flow signals, accurately measures the micro time difference, reduces the measurement error of small flows, extracts the characteristics of ultrasonic signals with high resolution under small flow working conditions, greatly improves the measurement accuracy of the propagation time difference, and thus significantly improves the accuracy and reliability of small flow measurement, expanding the measurement performance boundary of gas meters in low flow scenarios.

[0016] 4. The present invention processes ultrasonic gas meter signals through multi-algorithm collaboration, accurately measures gas flow under complex working conditions, filters out interference with a deep learning model, lays a stable signal foundation for measurement, enables the ultrasonic gas meter to optimize the signal processing strategy in real time under complex and variable working conditions, achieves high-precision measurement under all working conditions, ensures that the measurement accuracy is not affected by fluctuations in flow rate, temperature, pressure, and gas composition, improves operation stability and reliability, thereby expanding the applicable range of the ultrasonic gas meter and meeting the diverse actual working condition requirements and efficient management. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a method flow chart of a method for processing mixed signals of an ultrasonic gas meter proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] Next, the technical solutions of the present invention will be clearly and completely described in conjunction with the drawings of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0019] Embodiment: Please refer to the attached Figure 1 , the embodiment of the present invention provides a method for processing mixed signals of an ultrasonic gas meter, including the following steps: S1. Build an experimental platform that simulates the actual operating conditions of the gas meter, and collect ultrasonic original signals under different conditions through this platform to form an original signal data set; S2. Preprocess and expand the original signal data set, and divide it into a training set, a validation set, and a test set; S3. Use a deep learning model to train the training set to obtain a signal noise reduction model; S4. When the gas meter is operating, collect ultrasonic signals in real time, input them into the signal noise reduction model to filter out interference signals, and output pure signals; S5. Analyze the signal characteristics under small flow conditions for the pure signal by using the method of wavelet transform and cross-correlation fusion to obtain the propagation time differences of downstream and upstream flows; S6. Monitor the working conditions of the gas meter in real time, dynamically switch and fuse the algorithms in the algorithm library according to the working condition categories, and process the signals to obtain the measurement results; S7. Transmit the measurement results and the actual measurement accuracy back to the cloud platform, and adjust the algorithm parameters based on the big data analysis of the cloud platform for continuous optimization.

[0020] Specifically, by preprocessing and augmenting the original signal dataset, the training set, validation set, and test set are divided to optimize the data distribution, enhance data diversity and representativeness, improve the quality of the training data for the deep learning model, lay a foundation for the subsequent model to accurately learn signal features and interference patterns, avoid overfitting, and improve the generalization ability of the model.

[0021] By using the deep learning model to train the training set, a signal noise reduction model is obtained, thereby deeply mining the internal features and mapping relationships between signals and interference, constructing an efficient interference suppression model, accurately identifying and strongly filtering out multi-source interference signals under complex working conditions, and significantly improving the purity and stability of ultrasonic signals.

[0022] When the gas meter is running, ultrasonic signals are collected in real time, input into the signal noise reduction model to filter out interference signals, and pure signals are output, thereby purifying the signals in real time during the actual operation of the gas meter, ensuring the reliability of the input signals in the metering link, providing a stable and interference-free signal source for high-precision flow calculation, and improving metering accuracy and reliability.

[0023] By using the method of wavelet transform and cross-correlation fusion for the pure signal to analyze the signal characteristics under small flow conditions, the time differences of downstream and upstream propagation are obtained, thereby focusing on the weak characteristics of small flow signals, combining multi-scale decomposition with precise correlation analysis, accurately extracting the propagation time differences, significantly improving the metering resolution and accuracy under small flow conditions, and expanding the low-flow metering performance of the gas meter.

[0024] By real-time monitoring the working conditions of the gas meter, dynamically switching and fusing the algorithms in the algorithm library according to the working condition categories, and processing the signals to obtain the metering results, thereby dynamically and intelligently adapting the best algorithm combination according to the working conditions, ensuring accurate and efficient signal processing under all working conditions, and stably guaranteeing metering accuracy and reliability and enhancing applicability in wide-range flow, variable temperature, pressure, and composition scenarios.

[0025] By transmitting the metering results and the actual metering accuracy back to the cloud platform, adjusting the algorithm parameters based on big data analysis on the cloud platform for continuous optimization, thereby constructing a closed-loop optimization system for gas meter signal processing, deeply mining the relationship between working conditions and metering, and adaptively evolving the algorithm performance to maintain a high-precision and stable metering level of the gas meter in the long term.

[0026] By gradually completing S1 to S7, a full-process signal processing method covering data acquisition, model training, real-time processing, working condition adaptation, and continuous optimization is constructed, comprehensively improving the signal processing ability and metering accuracy of the ultrasonic gas meter under complex and dynamic working conditions, thus solving the problems that the traditional mixed signal processing method of ultrasonic gas meters cannot adapt to changes in working conditions, and the metering accuracy fluctuates greatly and the reliability is low under different working conditions.

[0027] In S1, the experimental platform uses a programmable signal generator to output ultrasonic signals with a frequency range of 50 kHz - 1 MHz and an amplitude accuracy of ±0.01 V, and is equipped with a radio frequency transmitting device that simulates electromagnetic interference of 0 - 200 V / m. The interference frequency is 10 kHz - 1 GHz. A mechanical vibration table that reproduces pipeline vibrations of 5 - 2000 Hz is introduced, combined with a temperature and humidity control system that regulates the temperature range from -30°C to 70°C and the humidity range from 10% to 95% RH. An analog gas pipeline is equipped. An ultrasonic transducer with a sensitivity of -200 dBre1V / μPa and a data acquisition card with a sampling rate of 5 MHz and 16-bit precision are used to collect the original signals. The single acquisition duration is not less than 30 s. Different conditions include electromagnetic interference intensity, pipeline vibration frequency, flow range, and temperature and humidity conditions.

[0028] Specifically, by building this experimental platform that highly simulates actual working conditions, complex environmental factors such as electromagnetic interference, pipeline vibration, temperature and humidity changes, and different flow conditions faced during the operation of gas meters can be accurately reproduced. It can comprehensively cover the ultrasonic signal characteristics and interference patterns under various working conditions, providing a large amount of, diverse, and practical original data samples for the subsequent research and development of signal processing technologies, laying a solid data foundation for the innovation of gas meter mixed signal processing technologies, enabling the developed algorithms and models to deeply adapt to the complexity and diversity of real working conditions, ensuring seamless connection from laboratory research to actual application, avoiding performance degradation of technologies in actual scenarios due to data deviation, and the technical solutions trained with the data collected by this platform can stably and accurately process signals in complex and changing real working conditions, effectively improving the measurement reliability and universality of gas meters.

[0029] In S2, preprocessing and augmentation involve using robust statistical methods to calculate the robust mean of the original signal and the robust standard deviation , and normalizing the signal amplitude to the range of [-1, 1] according to the formula , where is the value of the th sample point in the original signal sequence, and is the value of the th sample point in the signal sequence after standardization processing. Data augmentation uses operations such as random flipping, adding noise, time shifting, and frequency modulation. When adding noise, the dynamic noise standard deviation is set according to the working condition simulation data, and the value range is 0.005 - 0.2. The training set, validation set, and test set are divided in the proportions of 75%, 15%, and 10% in sequence.

[0030] Specifically, by using robust statistical methods to calculate the robust mean and robust standard deviation of the original signal and normalize them, and implementing data augmentation operations, the signal amplitude range is accurately calibrated, the interference of outliers is effectively suppressed, and at the same time, the data diversity and complexity are enriched. Robust statistics ensures the reliability of data preprocessing, and multi-method augmentation enables the data to cover more potential working condition changes, comprehensively improving the data quality and representativeness. A high-quality and strongly adaptable training dataset is created for the deep learning model, helping it accurately capture the core features and interference patterns of the signal, avoiding the risk of overfitting, strengthening the generalization performance of the model, ensuring stable and accurate signal processing under actual complex working conditions, optimizing from the data source, ensuring the efficient and stable operation of the model under complex working conditions, and steadily improving the accuracy and reliability benchmark of gas meter signal processing.

[0031] In S3, the use of a deep learning model to train the training set is to adopt a deep convolutional neural network. The first layer uses 3×3 small-size convolutional kernels, with a quantity of 64, paired with the LeakyReLU activation function. The middle layer stacks 4 convolutional modules, including convolution, instance normalization, LeakyReLU activation, and dilated convolution operations. The dilation rates are set to 1, 2, 3, and 4 in sequence. Dense connections are used between the modules. After adaptive average pooling at the end, it is connected to a fully connected layer, and the number of neurons is expanded to 256, paired with a Softmax classifier. The focal loss function and the adaptive moment estimation optimizer are selected, the initial learning rate is set to 0.002, and it decays by 0.2 times every 5 epochs. Iterative training is performed until the loss of the validation set does not decrease significantly for 8 consecutive epochs. The early stopping method and the learning rate annealing strategy are enabled, and the optimal model weights are saved.

[0032] Specifically, by constructing a customized deep convolutional neural network and supporting optimization strategies, the multi-layer architecture and unique convolutional kernel settings of the deep convolutional neural network can automatically extract hierarchical features of the ultrasonic signal; instance normalization and activation functions improve the model convergence speed and expressiveness; dilated convolution captures multi-scale features, and dense connections promote feature reuse; the focal loss function focuses on difficult-to-classify samples, and the adaptive moment estimation optimizer collaborates with the dynamic learning rate adjustment mechanism to efficiently train the model, thereby accurately learning the complex mapping relationship between the signal and interference, and constructing a powerful signal denoising model. This model can accurately identify and efficiently filter out interference signals under various working conditions, significantly improving the purity and stability of the ultrasonic signal, laying a foundation for subsequent accurate measurement, thus solving the problem that traditional filtering methods are difficult to handle interference in complex working conditions, effectively making up for the limitations of general signal processing methods in processing the mixed signals of ultrasonic gas meters, providing key technical support for high-precision measurement of gas meters in a strong interference environment, and strongly improving the operation reliability and measurement accuracy of gas meters.

[0033] The signal characteristics analyzed in S5 under low flow conditions are to perform 4-level wavelet packet decomposition on the denoised signal using the Symlets8 wavelet basis based on wavelet packet transform, obtain multi-level detail components and approximate components, and lock the third and fourth level high-frequency detail components; compare the high-frequency detail components with the preset standard small flow signal template Perform cross-correlation operation according to the formula Calculate the cross-correlation result, where is the optimized cross-correlation function result, is the high-frequency detail component after wavelet packet transform, It is a preset standard small flow signal template. is the index in the signal sequence, is the length of the signal sequence, is an exponential decay term, where is the attenuation coefficient, The cross-correlation window size is dynamically adjusted to 10-40 sampling points, and the step size is set to 0.5-3 sampling points. The simulated annealing algorithm is combined to lock the maximum peak value corresponding to Value, get the downstream and upstream propagation time difference .

[0034] Specifically, by selecting the Symlets8 wavelet basis based on wavelet packet transform to perform 4-level decomposition and optimize the cross-correlation operation on the denoised signal, the wavelet packet transform multi-scale decomposition can deeply analyze the signal frequency domain structure, explore the weak characteristics of small flow signals, and accurately lock the key high-frequency detail components; the attenuation coefficient and intelligent window adjustment strategy are introduced to optimize the cross-correlation operation, and the simulated annealing algorithm is used to overcome noise interference, and the signal characteristic peaks corresponding to the downstream and upstream propagation time differences are accurately locked. Therefore, under small flow conditions, ultrasonic signal characteristics can be extracted with high resolution, and the measurement accuracy of the propagation time difference can be greatly improved, thereby significantly improving the accuracy and reliability of small flow measurement, and expanding the metering performance boundary of gas meters in low flow scenarios, thereby solving the core problem of traditional algorithms in small flow conditions, that is, weak signals, noise influence, fuzzy feature extraction, large time difference measurement errors, and serious lack of metering accuracy. This provides a key technical breakthrough for gas meters to accurately measure small flow gas, ensuring that users have fair cost measurement and accurate energy management in low consumption scenarios.

[0035] The real-time monitoring of the gas meter working condition in S6 is carried out through high-precision platinum resistance temperature sensors, sapphire pressure sensors, and gas sensors. The sensor signal is amplified and band-pass filtered by a low-noise, high-gain conditioning circuit, and then connected to a 24-bit high-speed multi-channel A / D conversion chip. The sampling frequency is set to 2kHz, and the conversion result is read in real time by a microprocessor. The hybrid working condition judgment algorithm based on decision tree and neural network is used to divide the working condition category.

[0036] The dynamic switching and fusion algorithm library in S6 includes Kalman filtering, Hilbert-Huang transform, empirical mode decomposition, and particle filtering. The signal is processed to obtain the final measurement result. The intelligent switching rules are predefined according to the fine classification of the working condition. According to the formula Calculate the fusion signal processing result, where is the dynamic adaptive weighting coefficient, The final signal after dynamic switching and fusion algorithm processing, Algorithm library The signal processed by the algorithm.

[0037] Specifically, multiple types of high-precision sensors are used to accurately capture operating condition information such as temperature, pressure, and gas composition, and after circuit optimization processing and high-speed A / D conversion, reliable real-time data is provided to the microprocessor; the mixed operating condition judgment algorithm deeply mines the operating condition characteristics and accurately defines the operating condition category; the weights and combinations of various algorithms such as Kalman filtering are intelligently adjusted according to the operating conditions, and the signals are adaptively processed, so that the ultrasonic gas meter can optimize the signal processing strategy in real time under complex and changeable operating conditions, achieve high-precision measurement under all operating conditions, ensure that the measurement accuracy is not affected by fluctuations in flow, temperature, pressure, and gas composition, and improve operational stability and reliability, thereby expanding the scope of application of ultrasonic gas meters and meeting the diverse actual operating conditions and efficient management needs.

[0038] In S7, the metering results and actual metering accuracy are transmitted back to the cloud platform through the built-in 5G wireless module of the gas meter. The processing results, actual metering accuracy and operating condition data are packaged in JSON format and transmitted back to the cloud platform every 10 minutes. The algorithm parameters are adjusted based on the big data analysis of the cloud platform for continuous optimization. After the cloud platform receives and verifies them, they are stored in the distributed database. The relationship between errors and operating conditions is analyzed according to the big data analysis framework, and the parameters are adjusted using genetic and gradient descent methods. The update instructions are initially pushed through OTA to upgrade the algorithm during low-peak gas consumption periods.

[0039] Specifically, by utilizing the high-speed and stable communication capabilities of 5G, key data of gas meters are periodically collected and efficiently transmitted back to the cloud platform. After receiving the data, the cloud platform strictly verifies and stores it, and uses big data analysis to explore the deep correlation between metering errors and operating conditions. Genetic and gradient descent algorithms are used to accurately optimize algorithm parameters, and then upgrade instructions are safely pushed through OTA to update the algorithm seamlessly during low-peak gas consumption periods, thereby building a closed-loop continuous optimization system for gas meter signal processing technology, dynamically improving algorithm performance based on actual operation feedback, ensuring that gas meters maintain high-precision metering levels in the long term, and adapting to dynamic changes in the gas supply environment and user gas usage patterns, thereby improving user satisfaction and energy management efficiency, and promoting the intelligent and adaptive development of gas metering technology.

[0040] Although embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for processing mixed signals of an ultrasonic gas meter, characterized in that: The following steps are involved: S1. Build an experimental platform to simulate the actual operating conditions of the gas meter, and use the platform to collect ultrasonic raw signals under different conditions to form a raw signal data set; S2, preprocessing and expanding the original signal data set to divide it into a training set, a validation set and a test set; S3, using the deep learning model to train the training set to obtain a signal noise reduction model; S4, when the gas meter is running, the ultrasonic signal is collected in real time, input into the signal noise reduction model to filter out the interference signal, and output a pure signal; S5. The pure signal is analyzed by wavelet transform and cross-correlation fusion in the small flow condition to obtain the propagation time difference between the upstream and downstream. S6, real-time monitoring of the working condition of the gas meter, dynamic switching according to the working condition category, integrating the algorithms in the algorithm library, processing the signal to obtain the metering result; S7. Send the measurement results and actual measurement accuracy back to the cloud platform, and adjust the algorithm parameters based on the cloud platform big data analysis for continuous optimization.

2. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: The experimental platform in S1 uses a programmable signal generator to output an ultrasonic signal with a frequency range of 50kHz-1MHz and an amplitude accuracy of ±0.01V, and is equipped with a radio frequency transmitter that simulates 0-200V / m electromagnetic interference, with an interference frequency of 10kHz-1GHz. A mechanical vibration table that reproduces 5-2000Hz pipeline vibration is introduced, and a temperature and humidity control system is combined to adjust the temperature range to -30℃-70℃ and the humidity range to 10%-95%RH. It is also equipped with a simulated gas pipeline, and an ultrasonic transducer with a sensitivity of -200dBre1V / μPa and a data acquisition card with a sampling rate of 5MHz and 16-bit accuracy are used to collect the original signal. The single collection time is not less than 30s. The different conditions include electromagnetic interference intensity, pipeline vibration frequency, flow range, and temperature and humidity conditions.

3. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: The preprocessing and expansion in S2 is to use a robust statistical method to calculate the robust mean of the original signal Robust standard deviation , according to the formula Normalize the signal amplitude to the range [-1,1], where is the first The value of the sample points, is the first The data is expanded by random flipping, noise addition, time shift, and frequency modulation. When adding noise, the dynamic noise standard deviation is set according to the working condition simulation data, with a value range of 0.005-0.

2. The training set, validation set, and test set are divided into 75%, 15%, and 10% respectively.

4. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: In the S3, the deep learning model used to train the training set is a deep convolutional neural network. The first layer uses 3×3 small-size convolution kernels, with 64 kernels, and a LeakyReLU activation function. The middle layer stacks 4 convolution modules, including convolution, instance normalization, LeakyReLU activation, and hole convolution operations. The hole rates are set to 1, 2, 3, and 4, respectively. The modules are densely connected, and the fully connected layer is connected after adaptive average pooling at the end. The number of neurons is expanded to 256, and a Softmax classifier is used. The focal loss function and the adaptive moment estimation optimizer are selected. The initial learning rate is set to 0.002, and it is decayed by 0.2 times every 5 epochs. The iterative training is carried out until the validation set loss has no obvious decrease for 8 consecutive epochs. The early stopping method and the learning rate annealing strategy are enabled to save the optimal model weights.

5. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: The signal characteristics analyzed in S5 under the condition of small flow rate are to perform 4-level wavelet packet decomposition on the denoised signal using the Symlets8 wavelet basis based on wavelet packet transform, obtain multi-level detail components and approximate components, and lock the 3rd and 4th level high-frequency detail components; compare the high-frequency detail components with the preset standard small flow signal template Perform cross-correlation operation according to the formula Calculate the cross-correlation result, where is the optimized cross-correlation function result, is the high-frequency detail component after wavelet packet transform, It is a preset standard small flow signal template. is the index in the signal sequence, is the length of the signal sequence, is an exponential decay term, where is the attenuation coefficient, The cross-correlation window size is dynamically adjusted to 10-40 sampling points, and the step size is set to 0.5-3 sampling points. The simulated annealing algorithm is combined to lock the maximum peak value corresponding to Value, get the downstream and upstream propagation time difference .

6. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: The real-time monitoring of the working condition of the gas meter in S6 is to monitor the gas meter through a high-precision platinum resistance temperature sensor, a sapphire pressure sensor, and a gas sensor. The sensor signal is amplified and band-pass filtered by a low-noise, high-gain conditioning circuit, and then connected to a 24-bit high-speed multi-channel A / D conversion chip. The sampling frequency is set to 2kHz, and the conversion result is read in real time by a microprocessor. The hybrid working condition judgment algorithm based on a decision tree and a neural network is used to divide the working condition category.

7. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: The dynamic switching and fusion algorithm library in S6 includes Kalman filtering, Hilbert-Huang transform, empirical mode decomposition, and particle filtering. The processing signal to obtain the final metering result is to predefine intelligent switching rules according to the fine classification of working conditions, according to the formula Calculate the fusion signal processing result, where is the dynamic adaptive weighting coefficient, The final signal after dynamic switching and fusion algorithm processing, Algorithm library The signal processed by the algorithm.

8. The method for processing mixed signals of an ultrasonic gas meter according to claim 1, characterized in that: In the S7, the metering results and actual metering accuracy are transmitted back to the cloud platform through the built-in 5G wireless module of the gas meter. The processing results, actual metering accuracy and operating condition data are packaged in JSON format and transmitted back to the cloud platform every 10 minutes. The cloud platform big data analysis is based on which the algorithm parameters are adjusted for continuous optimization. The cloud platform receives and verifies the parameters and stores them in the distributed database. The error is analyzed in accordance with the big data analysis framework. The genetic and gradient descent methods are used to adjust the parameters. The update instructions are initially pushed via OTA to upgrade the algorithm during the off-peak gas consumption period.

Images