A pressure type liquid level meter remote calibration method and system based on the Internet of Things

Abstract

The application discloses a pressure type liquid level meter remote calibration method and system based on an Internet of Things, and belongs to the field of instrument detection and calibration. The method comprises the following steps: collecting multi-source sensing data containing pressure, medium temperature and ambient atmospheric pressure; performing adaptive median filtering to remove impulse noise and Gaussian weight-based moving average smoothing processing on the pressure data in sequence; based on the preprocessed data, a nonlinear environmental compensation model is constructed by using a recursive least square method with a forgetting factor to dynamically track the nonlinear drift of the sensor; based on the real-time correction coefficient output by the model and the standard calibration curve, the calibration offset is calculated, the calibration instruction is generated, and the parameter update is executed to the terminal; through end-cloud cooperation and multi-source data fusion, the application realizes remote calibration and predictive maintenance with high precision and high safety, and significantly improves the measurement reliability and maintenance efficiency of the liquid level meter under complex working conditions.

Description

Technical Field

[0001] This application belongs to the field of instrumentation testing and calibration, specifically relating to a remote calibration method and system for a pressure level gauge based on the Internet of Things. Background Technology

[0002] With the rapid development of Industrial Internet of Things (IIoT) technology, its application in industrial monitoring and automation control has gradually become an important means to improve production efficiency and management level. Pressure level gauges, as key sensors for measuring liquid height, are widely used in petrochemical, water conservancy, and pharmaceutical industries. In complex industrial environments, the accuracy of level gauges directly affects the safety of the production process and the precision of material measurement, which places higher demands on the stability, real-time performance, and long-term reliability of level monitoring systems.

[0003] Among these technologies, IoT-based remote calibration is a core approach to ensuring the measurement accuracy of pressure level gauges. It enables unified management and parameter correction of level gauges distributed across different geographical locations via network transmission. This technology aims to utilize real-time data collected by sensors to dynamically adjust the zero point, full-scale range, and linearity of the level gauges through a remote control terminal, thereby reducing the cost of manual on-site maintenance and increasing calibration frequency.

[0004] However, traditional calibration methods rely heavily on manual on-site operation, resulting in long maintenance cycles and an inability to cope with sudden measurement deviations. Furthermore, pressure level gauges are susceptible to nonlinear drift due to fluctuations in ambient temperature and changes in medium density during long-term operation. Existing remote calibration solutions lack the ability to deeply perceive multi-source environmental parameters, making it difficult to accurately compensate for measurement errors. In addition, data transmission during calibration is subject to delays and noise interference. Traditional processing logic cannot effectively identify and eliminate abnormal fluctuations in data, leading to insufficiently sensitive feedback adjustment in remote calibration and affecting the measurement accuracy and stability of the level gauge under extreme operating conditions.

[0005] Therefore, a remote calibration method and system for pressure level gauges based on the Internet of Things is desired. Summary of the Invention

[0006] The purpose of this invention is to provide a remote calibration method and system for pressure level gauges based on the Internet of Things, which can effectively solve the problems in the background art.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0008] The first aspect is a remote calibration method for a pressure level gauge based on the Internet of Things disclosed in this application, which includes the following specific steps:

[0009] On the cloud platform side, it receives pressure values, medium temperature values, and ambient atmospheric pressure values ​​with timestamps uploaded by field terminals, forming the original time series;

[0010] The original time series is preprocessed, including: using adaptive median filtering to remove impulse noise, and dynamically expanding the filter window length based on the comparison between the absolute value of the first difference of the pressure signal and the preset fluctuation threshold; then using a moving average algorithm based on Gaussian distribution function to perform secondary smoothing on the signal to suppress high-frequency random noise and reduce phase lag, resulting in the preprocessed data series.

[0011] Based on the preprocessed data sequence, a nonlinear environmental compensation model is constructed using the recursive least squares method with a forgetting factor. The input vector of the nonlinear environmental compensation model includes at least the net pressure value, temperature value and its higher-order terms, so as to dynamically track the nonlinear drift of the sensor caused by environmental changes and long-term operation, and output real-time correction coefficients.

[0012] Based on the real-time correction coefficients, the zero-point offset, full-scale gain error, and linearity deviation are calculated, and a calibration control command containing the zero-point correction value, slope compensation coefficient, and linearized polynomial parameters is generated.

[0013] The calibration control command is sent to the field terminal to implement remote parameter updates.

[0014] Furthermore, adaptive median filtering is used to remove impulse noise, specifically including:

[0015] Real-time calculation of the first-order absolute difference of the pressure signal;

[0016] When the absolute value of the first-order difference exceeds the preset fluctuation threshold, it is determined that the current signal point is in a region of severe fluctuation or is subject to impulse noise interference, and the default filter window length is dynamically expanded from 3 sampling points to 7 sampling points.

[0017] Sort all sampled point values ​​within the expanded window and select the median of the sorted values ​​as the filtered output value at the current time point.

[0018] Furthermore, a moving average algorithm based on Gaussian distribution function weight allocation is used to perform secondary smoothing of the signal, specifically including:

[0019] Define a sliding window of length N, where N is an odd number;

[0020] Based on the time distance between each sampling point in the window and the center point of the window, the corresponding weight is calculated using a Gaussian function, so that the data closer to the center of the window has a greater weight.

[0021] The original pressure values ​​of each sampling point within the window are multiplied by their corresponding weights, summed, and then divided by the sum of all weights to obtain the filtered output value at the current moment.

[0022] Furthermore, a nonlinear environmental compensation model is constructed using the recursive least squares method with a forgetting factor, specifically including:

[0023] Construct an input vector, which includes a bias term, net pressure value, temperature value, a term with the square of the net pressure value, a term with the square of the temperature value, and a term with the product of the net pressure value and the temperature value.

[0024] By introducing a forgetting factor to assign decreasing weights to historical data, and updating the model parameter vector through recursive calculation, the model output can be made to approximate the true liquid level value.

[0025] The net pressure value is obtained by subtracting the real-time ambient atmospheric pressure from the pressure value.

[0026] Furthermore, a calibration control command is generated, including zero-point correction values, slope compensation coefficients, and linearized polynomial parameters, specifically including:

[0027] Set the zero-point correction value Z to the opposite of the zero-point offset;

[0028] The slope compensation coefficient K is calculated based on the full-scale gain error. The calculation formula is K=1 / (1+ΔG), where ΔG is the full-scale gain error.

[0029] By performing polynomial fitting on the deviation data at each feature point, the linearized polynomial parameters are obtained. The deviation data is the difference between the ideal output value and the actual output value corresponding to the standard liquid level.

[0030] Furthermore, after the calibration control command is sent to the field terminal, a closed-loop verification step performed by the field terminal is also included:

[0031] The field terminal receives and decrypts the calibration control command, and writes the new calibration parameters into the non-volatile memory;

[0032] Enter self-test mode, collect the current pressure signal, medium temperature and ambient atmospheric pressure, and apply new calibration parameters to correct the uncalibrated liquid level value, obtain the calibrated measured liquid level value and upload it to the cloud platform;

[0033] The cloud platform receives multiple calibrated measured liquid level values ​​within a preset time period, compares them with the corresponding reference liquid level values, and calculates the sum of squared residuals.

[0034] The residual sum of squares is compared with a preset residual threshold. If the residual sum of squares is less than the threshold, the calibration is considered successful; if it is greater than or equal to the threshold, the calibration is considered unsuccessful, and a fault alarm is triggered when the number of consecutive failures reaches a preset value.

[0035] Furthermore, the method also includes predictive maintenance steps:

[0036] Extract historical calibration records of the level gauge terminal from the cloud database. Each record includes at least the calibration timestamp, zero offset, full-scale gain error, and number of ambient temperature cycles since the last calibration.

[0037] By using multiple consecutive calibration records as input features, training samples are constructed and fed into a long short-term memory neural network for training to obtain a predictive maintenance model.

[0038] Using a trained predictive maintenance model, based on the most recent calibration records of the current terminal, the time interval for the next recommended calibration is predicted, and an early warning message is generated when the remaining time is less than a preset lead time.

[0039] Furthermore, the number of ambient temperature cycles is defined as the number of times the ambient temperature crosses a preset threshold range within a statistical period.

[0040] Furthermore, before receiving the timestamped pressure value, medium temperature value, and ambient atmospheric pressure value uploaded by the field terminal, the process also includes:

[0041] The liquid level gauge terminal is equipped with a high-precision pressure sensing unit, a platinum resistance temperature sensor, and a MEMS absolute pressure sensor to simultaneously collect pressure, medium temperature, and ambient atmospheric pressure.

[0042] Each frame of data collected is timestamped using a GPS or BeiDou timing module.

[0043] The timestamped data is encapsulated and a checksum is generated using the CRC-16 cyclic redundancy check algorithm. The data is then transmitted to the cloud platform via a narrowband IoT protocol.

[0044] Secondly, the IoT-based remote calibration system for pressure level gauges disclosed in this application includes:

[0045] This includes on-site terminals and cloud platforms;

[0046] The field terminal includes:

[0047] A high-precision pressure sensing unit is used to collect pressure values;

[0048] Platinum resistance temperature sensor, used to acquire medium temperature values;

[0049] MEMS absolute pressure sensor, used to collect ambient atmospheric pressure values;

[0050] The time synchronization module is used to add a high-precision timestamp to each frame of data collected;

[0051] The first processing unit is used to encapsulate the collected raw data and upload it to the cloud platform through the narrowband IoT communication module; and to receive and execute the calibration control instructions issued by the cloud platform and write the new calibration parameters into the non-volatile memory.

[0052] The cloud platform includes:

[0053] The data receiving module is used to receive pressure, temperature and atmospheric pressure data with timestamps uploaded by the field terminal;

[0054] The data preprocessing module is used to perform adaptive median filtering and Gaussian weighted moving average filtering on the original time series.

[0055] The model building module is used to construct a nonlinear environmental compensation model based on preprocessed data using the recursive least squares method with a forgetting factor, and output real-time correction coefficients.

[0056] The calibration decision module is used to calculate the zero-point offset, full-scale gain error and linearity deviation based on real-time correction coefficients, and generate calibration control instructions that include zero-point correction values, slope compensation coefficients and linearized polynomial parameters.

[0057] The command issuance module is used to encrypt calibration control commands and then issue them to the corresponding field terminals.

[0058] The closed-loop verification module is used to receive the calibrated measurement values ​​fed back by the field terminal, calculate the sum of squares of the residuals between the measured values ​​and the reference liquid level values, and determine whether the calibration is successful based on the comparison results with the preset threshold.

[0059] The predictive maintenance module is used to predict the next recommended calibration interval and generate early warning information based on the zero offset, full-scale gain error and ambient temperature cycle number in the historical calibration records, using a long short-term memory neural network.

[0060] In summary, this application includes at least one of the following beneficial technical effects:

[0061] 1. This invention achieves dynamic compensation for temperature drift and pressure fluctuations in pressure level gauges under complex industrial environments by constructing a multi-source sensing data acquisition sequence and a nonlinear environmental compensation model. The overall measurement error of the calibrated system is significantly reduced, and the zero-point drift rate during long-term operation is greatly decreased, significantly enhancing the reliability of level monitoring.

[0062] 2. This invention's IoT-based remote calibration mechanism completely revolutionizes the traditional manual on-site maintenance model. Maintenance personnel no longer need to carry specialized calibration equipment to geographically remote monitoring points; they can complete parameter calibration across the entire measurement range via a cloud platform. The calibration operation time for a single terminal is significantly reduced compared to the traditional model, greatly improving maintenance efficiency while eliminating subjective errors caused by manual operation.

[0063] 3. The predictive maintenance module of this invention, integrating a long short-term memory neural network, can anticipate potential faults and provide calibration suggestions based on the historical performance degradation trends of equipment. This shift from reactive maintenance to proactive prevention significantly improves equipment uptime. Simultaneously, the closed-loop verification mechanism ensures the validity of each parameter update, providing high-quality, highly reliable data support for industrial automation control.

[0064] 4. The system of this invention adopts a distributed architecture and narrowband IoT communication technology, supporting smooth access for a massive number of terminals and meeting the monitoring needs of large-scale industrial sites. The introduction of multiple encryption algorithms and secure handshake protocols ensures the security of the remote calibration link, guaranteeing that industrial production data is not leaked or illegally tampered with in complex network environments, and has extremely high engineering application value. Attached Figure Description

[0065] Figure 1 This is an overall schematic diagram of a remote calibration method for pressure level gauges based on the Internet of Things.

[0066] Figure 2 This is a schematic diagram of the data flow in the remote calibration method for level gauges;

[0067] Figure 3 This is a flowchart of constructing a mathematical model for compensating for temperature drift and pressure fluctuation of a pressure level gauge using the recursive least squares method to calculate the real-time correction coefficient;

[0068] Figure 4 It is a flowchart that calculates the calibration offset based on the real-time correction coefficient and issues control commands to implement remote parameter updates and closed-loop verification. Detailed Implementation

[0069] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute any limitation on the scope of protection of this invention.

[0070] The core of the IoT-based remote calibration method for pressure level gauges disclosed in the application lies in constructing an edge-cloud collaborative intelligent calibration system to achieve accurate and efficient remote parameter correction for widely distributed level gauges. The following will combine... Figures 1 to 4 The flowchart shown illustrates the specific implementation steps of this method in detail.

[0071] Firstly, regarding step S1, constructing a multi-source sensing data acquisition sequence is the foundation of the entire remote calibration method. Its core lies in deploying level gauge terminals with multi-parameter sensing capabilities to simultaneously collect pressure, medium temperature, ambient atmospheric pressure, and precise time information, and transmitting this raw data to the cloud platform via a reliable IoT link. Step S1 aims to provide high-quality, highly synchronized data raw materials for subsequent data preprocessing, environmental compensation, and remote calibration. The implementation process of this step will be described in detail below with reference to specific embodiments.

[0072] Step S101: Configure a high-precision pressure sensing unit. Deploy a pressure-type level gauge terminal at the monitoring site. Its core pressure sensing component uses a diffused silicon pressure-sensitive chip, which is encapsulated in a probe cavity with an IP68 protection rating. The front end contacts the measured medium through a stainless steel diaphragm to sense the static pressure generated by the medium.

[0073] Depending on the specific application scenario, such as monitoring water levels in reservoirs or channels in water conservancy projects, the sensor's measurement range is adapted to 0 to 50 meters of water column. To ensure the reliability of the basic measurement data, the overall accuracy level of the sensitive element is set to 0.1. During data acquisition, the constant current source drive circuit inside the terminal provides a stable excitation current to the Wheatstone bridge inside the pressure sensor, generating a millivolt-level differential voltage signal proportional to the applied pressure.

[0074] The signal is first conditioned and amplified by a high-input-impedance, low-drift instrumentation amplifier, and then digitized by a 24-bit resolution analog-to-digital converter. The data sampling frequency is fixed at 100 Hz to capture minute transient pressure changes caused by liquid surface fluctuations.

[0075] Step S102: Simultaneously acquire medium temperature and ambient atmospheric pressure. To achieve accurate compensation for the pressure measurement, the medium temperature and ambient atmospheric pressure must be acquired simultaneously with the pressure measurement. Inside the probe, close to the diffused silicon pressure chip, a high-precision platinum resistance temperature sensor, such as PT100 or PT1000, is integrated for real-time monitoring of the temperature of the measured medium.

[0076] Meanwhile, inside the electrical housing of the level gauge terminal, a MEMS absolute pressure sensor is deployed at a location far from heat sources such as the main control chip and communication module. This sensor is specifically designed to collect real-time atmospheric pressure data above the liquid surface, which will serve as the benchmark for deducting the influence of atmospheric pressure in subsequent compensation calculations.

[0077] Step S103: Assign a precise time stamp to each frame of measurement data. The level gauge terminal integrates a timing module that supports GPS or BeiDou satellite navigation systems. This module provides a high-precision time reference for the local system clock after locking onto the satellite signal.

[0078] After each set of data on pressure, medium temperature, and atmospheric pressure is collected, the system clock immediately adds a timestamp to it. The synchronization accuracy of this timestamp is better than 10 milliseconds, which provides an accurate time basis for subsequent analysis of the time series characteristics of the data and the fusion of data from multiple terminals.

[0079] The steps S101 to S103 above together form the foundation for underlying data acquisition, enabling simultaneous perception of four dimensions: pressure, temperature, air pressure, and time, providing rich and accurate raw data for subsequent remote calibration.

[0080] Step S104: Encapsulate the collected data and reliably transmit it via the Internet of Things. Packet the aforementioned timestamped pressure, temperature, and atmospheric pressure data according to a predefined message format.

[0081] At the end of the message, the system uses the CRC-16 cyclic redundancy check algorithm to generate a 16-bit checksum for the receiver to verify the integrity of the data. The encapsulated data packet is then processed by the low-power wide-area network (LPWAN) communication module, which supports the NB-IoT protocol. Its uplink transmit power is configurable to 23dBm, and its receive sensitivity reaches -129dBm, ensuring reliable data transmission even in weak signal environments such as complex industrial sites or underground facilities.

[0082] After the data packet is sent to the cloud IoT platform via the wireless network, the platform first checks the integrity of the data packet based on the check bit. If the check fails, a retransmission request is sent to the terminal, thereby ensuring the quality of the data entering the analysis stage from the source.

[0083] In summary, through step S1 above, the system successfully constructed a multi-source sensing data acquisition sequence, laying a solid data foundation for subsequent data preprocessing, environmental compensation, and remote calibration. These steps are closely linked and together constitute the first stage of the entire remote calibration method.

[0084] For step S2, data preprocessing and noise suppression are crucial steps to ensure the accuracy of subsequent environmental compensation and calibration. The raw data sequences received by the cloud platform often contain various types of noise caused by electromagnetic interference, bubble impaction, or electronic thermal noise; therefore, data cleaning and filtering are essential. The implementation process of step S2 will be described in detail below with reference to specific embodiments.

[0085] Step S201: Data Unpacking and Sequence Extraction. After receiving the raw data packets from the field terminals, the cloud-based IoT platform first unpacks the data packets according to a pre-agreed communication protocol, reconstructing pressure values, medium temperature values, and ambient atmospheric pressure values ​​with precise timestamps, forming three independent raw time series. These sequences will serve as input for all subsequent signal processing and analysis.

[0086] Step S202: An adaptive median filtering algorithm is used to remove impulse noise. To address isolated impulse noise that may exist in the pressure signal due to brief electromagnetic interference or bubble impacts on the sensor diaphragm, this embodiment employs an improved adaptive median filtering algorithm.

[0087] In practice, the system calculates the first-order difference of the pressure signal, i.e., the difference between adjacent sampling points, in real time, and compares the absolute value of this difference with a preset fluctuation threshold. This preset fluctuation threshold is calibrated based on the base noise level of the sensor in a static state, and can be set to, for example, the rate of change corresponding to 0.5% of the full-scale output value.

[0088] When the absolute value of the first-order difference exceeds the threshold, the system determines that the current signal point is in a region of violent fluctuation or is affected by impulse noise. At this time, the length of the filtering window is dynamically expanded from the default 3 sampling points to 7 sampling points to enhance the coverage of abnormal impulses.

[0089] After determining the window, the values ​​of all sampling points within the window are sorted, and the median of the sorted values ​​is selected as the filtered output value at the current time point, thereby effectively eliminating isolated pulse spikes.

[0090] Step S203 involves smoothing high-frequency random noise using a Gaussian weighted moving average algorithm. After median filtering, the signal may still contain high-frequency random fluctuations caused by electronic thermal noise or slight liquid surface oscillations. To further suppress this type of noise while preserving the true dynamic characteristics of liquid level changes to the greatest extent possible, this embodiment employs a moving average filtering method based on Gaussian distribution function weight allocation for secondary smoothing.

[0091] Specifically, a sliding window of length N is defined, and the weight corresponding to each sampling point within the window is... The weight is calculated using a Gaussian function based on its time distance from the window center point. The weight calculation formula is:

[0092]

[0093] Where i is the index of the sampling point within the window, from 1 to N; μ is the index of the center position of the window, and when N is an odd number, ... σ is the preset standard deviation of the Gaussian kernel width, used to control the decay rate of the weights.

[0094] In this embodiment, the window length N can be set according to the sampling frequency and the frequency characteristics of liquid level changes. For example, at a sampling rate of 100 Hz, N=11 is taken, which corresponds to a time window of about 0.11 seconds, which can effectively smooth noise without excessive lag; σ is set to 2.0 based on experience to ensure that the sampling points near the center point receive higher weights.

[0095] Based on the above weights, the filtered output value y(n) at the current time n is calculated by the weighted average of the sampling points within the window:

[0096]

[0097] In the formula, This represents the original pressure value of the i-th sampling point from the oldest to the newest within the window. The denominator is the sum of all weights, used for normalization.

[0098] This weighting method allows data closer to the center of the window to contribute more to the output, thus effectively suppressing high-frequency noise while significantly reducing the signal phase lag caused by the traditional moving average method.

[0099] Step S204: Store the preprocessed data. The pressure, temperature, and atmospheric pressure time series data, after the above two filtering steps, are considered to have largely eliminated noise interference and can accurately reflect the physical processes at the site. This data is then stored in a temporary buffer in the cloud database, awaiting the next stage of nonlinear environmental compensation calculations.

[0100] In summary, through step S2, the system completes the cleaning and noise reduction of the raw sensing data, providing high-quality data input for the subsequent establishment of an accurate environmental compensation model. These preprocessing operations are prerequisites for ensuring the reliability of remote calibration and can effectively avoid the adverse effects of noise on the calculation of calibration parameters.

[0101] For step S3, establishing a nonlinear environmental compensation model is the core algorithmic step of the entire remote calibration method. Its purpose is to eliminate the interference of environmental temperature fluctuations and atmospheric pressure changes on pressure measurement, and to dynamically track the performance degradation of the sensor due to long-term operation. The implementation process of this step will be described in detail below with reference to specific embodiments.

[0102] Step S301: Extract preprocessed environmental characteristic parameters. The system reads the current medium temperature and ambient atmospheric pressure values ​​in real time from the data sequence processed in step S2. These two parameters will serve as the main input variables for subsequent compensation calculations.

[0103] Step S302: Preliminary physical compensation is performed based on the principle of hydrostatic pressure. The measurement of the pressure level gauge is based on the basic formula of statics. The measured absolute pressure is equal to the atmospheric pressure plus the product of the medium density, gravitational acceleration, and liquid level height.

[0104] Therefore, the uncompensated liquid level can be obtained by subtracting atmospheric pressure from absolute pressure and then dividing by the product of density and gravitational acceleration. However, since the density of the medium changes with temperature, and the pressure sensor itself has temperature drift and nonlinear characteristics, the initially calculated liquid level needs to be further corrected.

[0105] Step S303: Perform temperature compensation to correct the effects of changes in medium density. The system pre-builds and stores a fluid physical property database, which covers the density data of commonly used media in industrial settings, such as pure water, crude oil, and various acid, alkali, and salt solutions, in the range of -20 degrees Celsius to 80 degrees Celsius, with a temperature step size set to 0.1 degrees Celsius.

[0106] For each medium, the database stores a temperature-density lookup table. When performing temperature compensation, the system first selects the corresponding density table based on the type of medium being measured by the current level gauge, and then uses the current temperature as an index to look up the density value at that temperature.

[0107] If the current temperature is exactly equal to a certain temperature node in the table, the corresponding density is read directly; if it is between two temperature nodes, the accurate density is calculated using a linear interpolation method.

[0108] Step S304: Perform atmospheric pressure compensation. The system directly utilizes the ambient atmospheric pressure collected in real time by the absolute pressure sensor integrated inside the terminal. During each liquid level calculation, the atmospheric pressure is subtracted from the pressure sensor measurement value to obtain the net pressure value, thereby eliminating the influence of atmospheric pressure fluctuations on liquid level measurement.

[0109] Step S305: An adaptive compensation model is constructed using the recursive least squares method to track the nonlinear drift and aging trend of the sensor itself. This model takes the pre-compensated net pressure value, medium temperature, and their higher-order and cross-terms as inputs, and outputs the final corrected liquid level height.

[0110] Specifically, the model input vector includes a bias term, net pressure value, temperature value, a term consisting of the square of the net pressure value, the square of the temperature value, and a product term of the net pressure value and temperature. These terms are used to fit the nonlinear characteristics of the sensor and measurement system. The model output is assumed to be a linear combination of the input vector and the model parameter vector.

[0111] The model updates the parameter vector in real time using recursive least squares. A forgetting factor is introduced during the update process, with a value set between 0.95 and 0.98. This forgetting factor is used to assign decreasing weights to historical data, making the algorithm pay more attention to recent observations to track the time-varying characteristics of the system.

[0112] The recursive calculation process includes calculating the gain vector based on the current input vector and the error covariance matrix of the previous time step, calculating the prior error based on the error between the current reference liquid level value and the model output value, updating the parameter vector using the gain vector and the prior error, and finally updating the error covariance matrix for use in the next time step.

[0113] This recursive process model continuously absorbs new observation data and adjusts parameters to make the output approximate the true liquid level, thereby effectively compensating for zero-point drift, sensitivity changes, and nonlinear distortions that occur during long-term sensor operation. The reference liquid level value can be obtained from a high-precision reference device or a known liquid level calibration for supervised learning. When no real-time reference value is available, the system can use historical calibration data or samples obtained through periodic manual comparison as a reference.

[0114] Step S306: The final corrected liquid level height is used as the accurate measurement value at the current moment for subsequent calibration decisions. At the same time, the model parameters at the current moment are stored in the cloud as the basis for equipment health status assessment.

[0115] In summary, through step S3, the system constructs an environmental compensation model that integrates physical mechanisms and adaptive learning. This model can dynamically eliminate measurement errors caused by temperature, air pressure, and sensor aging, providing accurate and reliable liquid level data for remote calibration. These compensation operations are the core steps in ensuring the effectiveness of calibration and directly determine the measurement accuracy of the entire system.

[0116] Step S4, calculating the calibration offset and issuing control commands, is a crucial step in translating the environmental compensation results into specific calibration actions. Based on the real-time correction coefficients output in step S3, the cloud platform compares them with the inherent standard characteristics of the equipment to generate calibration parameters for correcting sensor measurement errors, and then sends the commands to the field terminal via a secure link. The implementation process of this step is described in detail below with reference to a specific embodiment.

[0117] Step S401: Obtain the real-time correction coefficients and standard calibration curve. The cloud platform reads the real-time correction coefficients of the current level gauge terminal from the adaptive compensation model in step S3. These coefficients are represented in a set of numerical values, specifically including the model parameter vector Θ updated by the recursive least squares method, which contains weights corresponding to the input features.

[0118] At the same time, the preset standard calibration curve calibrated at the factory is retrieved from the device file of the terminal. This curve is the ideal response characteristic obtained through multi-point testing under standard laboratory environmental conditions. It is stored in the cloud database in the form of a series of input and output data tables of characteristic points. Each characteristic point records the standard liquid level value and its corresponding ideal output value.

[0119] Step S402: Select comparison feature points and obtain ideal output values. Extract multiple key feature points from the standard calibration curve, specifically including the zero point (0% range), 25% range, 50% range, 75% range, and full scale (100% range). Record the ideal output value corresponding to each feature point. ,in These represent the feature points.

[0120] Step S403: Calculate the actual output value of each feature point under the current operating condition. For each feature point, the system calculates the actual output value corresponding to the same standard liquid level under the current operating condition based on the real-time correction coefficient, i.e., the model parameter vector Θ. .

[0121] The specific calculation method is as follows: For a given standard liquid level value First, the corresponding net pressure value is deduced by using the relationship between the net pressure value and the liquid level obtained in step S304. That is, according to the static pressure formula ,in The density of the medium can be obtained by querying the density value at the current temperature from the temperature compensation database in step S303. The local gravitational acceleration is a known constant.

[0122] Then, based on this net pressure value and the current medium temperature... Constructing input vectors The input vector contains bias term 1 and net pressure value. ,temperature The square of the net pressure value The square of temperature and the product of net pressure and temperature. .

[0123] Finally, the formula is output through the model. The current actual output value is calculated.

[0124] Step S404: Calculate the three types of core biases. Based on the ideal output value at each feature point. Compared with the actual output value The system calculates the zero-point offset, full-scale gain error, and linearity deviation, respectively. Zero-point offset Take the zero point The actual output value at the corresponding feature point minus the ideal output value, i.e. .

[0125] Full-scale gain error By comparing the full-scale point, i.e. The difference between the ratio of the actual output change to the ideal output change at the corresponding feature point and 1 is obtained, and the calculation formula is as follows: This value reflects the overall change in sensor sensitivity.

[0126] linearity deviation Quantification is achieved by calculating the percentage of the maximum deviation of the actual output value from the ideal straight line at each feature point relative to the full-scale output range. First, an ideal straight line is constructed, typically using the endpoint connection method to obtain the straight line equation. ,in This is the full-scale liquid level value.

[0127] Then calculate the absolute value of the deviation at each point. Take the maximum value and divide by the full-scale output range. Multiply by 100% to get the linearity deviation percentage.

[0128] Step S405: Generate calibration control command parameters. Based on the three types of deviations calculated in step S404, the system derives specific calibration parameters used to correct the sensor's measurement characteristics. Zero-point correction value. Take the opposite of the zero offset directly, that is This is used to subtract the fixed offset during subsequent measurements.

[0129] Slope compensation coefficient Calculated based on full-scale gain error This coefficient is used to adjust the overall gain of the measurement results. The linearized polynomial parameter is obtained by adjusting the deviation data at each feature point. Polynomial fitting was performed, and a least squares method was used to fit a value related to the liquid level. polynomial The polynomial order is set according to actual needs, for example, third order. The goal of fitting is to make the polynomial output as close as possible to the deviation values ​​at each point. The coefficients obtained from the fitting are... These are the parameters of the linearized polynomial.

[0130] Step S406: Encapsulate, encrypt, and issue calibration commands. The zero-point correction value generated in step S405... Slope compensation coefficient and linearized polynomial parameters Packets are assembled according to the agreed message format, and the target terminal's unique identifier and current timestamp are added to the instruction packet for the purpose of preventing replay attacks and instruction tracing.

[0131] To ensure the confidentiality and integrity of the instructions during transmission, the AES-256 symmetric encryption algorithm is used to encrypt the instruction packet. The key width is 256 bits, the encryption mode is CBC cipher block chaining mode, and the initialization vector IV is randomly generated and appended to the ciphertext.

[0132] The encryption key is pre-stored in a secure storage area on the cloud platform and the field terminal, and key negotiation and distribution are completed during device registration. The encrypted command packet is sent to the corresponding IoT gateway through the downlink channel of the IoT cloud platform. After receiving the downlink data, the gateway parses the terminal physical address in the command packet and, based on the locally maintained terminal network topology information, accurately forwards the command to the target level gauge terminal through the narrowband IoT backhaul link.

[0133] If the terminal is in sleep mode, the gateway can temporarily store the instruction and wait for the terminal to wake up before sending it.

[0134] Through step S4 above, the system completes the entire process from deviation calculation to the safe issuance of calibration commands, ensuring the accurate generation and reliable transmission of calibration parameters, and laying a solid foundation for subsequent parameter updates and closed-loop verification on the terminal side. These operations constitute the decision-making and command issuance links of the remote calibration method, directly determining the accuracy and safety of the calibration actions.

[0135] Step S5, which involves remote parameter updates and closed-loop verification, is the final execution stage of the entire calibration process. Its purpose is to reliably write the calibration instructions generated in the cloud into the field terminal and to quantitatively evaluate the calibration effect through real-time data feedback, ensuring the effectiveness of the calibration operation. The implementation process of step S5 will be described in detail below with reference to a specific embodiment.

[0136] Step S501: The terminal receives and decrypts the calibration command. After receiving the encrypted calibration command packet from the cloud platform via the narrowband IoT backhaul link, the on-site level gauge terminal first extracts the Initialization Vector (IV) from the command packet header and then decrypts the encrypted portion of the command packet using a 256-bit AES key pre-stored in the terminal's secure storage area. Decryption is performed in CBC mode to restore the zero-point correction value. Slope compensation coefficient Linearized polynomial parameters And plaintext instructions for timestamps and terminal identifiers.

[0137] The terminal then performs a CRC-16 cyclic redundancy check on the plaintext command again, comparing the calculated 16-bit checksum with the checksum appended to the end of the command packet to verify whether any data errors or loss occurred during transmission. If the check fails, the terminal discards the command and sends a retransmission request to the cloud.

[0138] In step S502, the new calibration parameters are written to the non-volatile memory. After successful verification, the microcontroller inside the terminal initiates the EEPROM erase / write process. This EEPROM uses a high-reliability electrically erasable programmable read-only memory, designed with an erase / write life of no less than 1 million cycles to meet the needs of long-term, frequent calibration.

[0139] The microcontroller accesses the EEPROM via an I²C or SPI interface, locates the specific address space storing the current calibration parameters, first performs an erase operation to clear the data in that area, and then performs the new zero-point correction value. Slope compensation coefficient The linearized polynomial parameters are written sequentially to the corresponding addresses, and a readback check is performed to ensure that the data is written correctly.

[0140] After the write operation is complete, the system updates the active parameter field in memory, making the new parameters take effect immediately.

[0141] Step S503: Enter self-test mode and provide real-time measurement data. After the parameters are updated, the terminal automatically enters self-test mode. In this state, the microcontroller acquires the pressure signal at the current moment according to the normal measurement procedure. medium temperature and ambient atmospheric pressure .

[0142] First, the pressure signal is sequentially subjected to the adaptive median filtering described in step S202 and the weighted moving average filtering described in step S203 to obtain a smoothed pressure value. .

[0143] Then, temperature compensation is performed using the method in step S303: based on the current temperature. Query the fluid physical properties database to obtain the medium density If the temperature is between nodes, linear interpolation is used.

[0144] Then, atmospheric pressure compensation is performed according to step S304 to obtain the net pressure value. Calculate the uncalibrated liquid level value based on the hydrostatic formula. ,in The local gravitational acceleration is a known constant.

[0145] Finally, apply the newly written calibration parameters to... The correction is made, and the correction formula is:

[0146]

[0147] in The polynomial value calculated based on the parameters of the linearized polynomial, i.e.:

[0148]

[0149] This is the calibrated measured liquid level value. This measured value, along with the corresponding acquisition timestamp, is packaged together according to the message format defined in step S104 and uploaded to the cloud platform via the IoT link as preliminary feedback on the calibration effect.

[0150] Step S504: The cloud platform initiates the closed-loop verification process. After receiving the first frame of measurement data from the terminal, the cloud platform starts a closed-loop verification timer for a preset time period. This time period can be set according to the on-site working conditions, for example, 10 minutes.

[0151] During this period, the platform continuously receives real-time measurement data uploaded by the terminal and collects no less than 50 valid sample points. The threshold for the number of sample points can be adjusted according to the statistical significance requirements.

[0152] Step S505: Calculate the sum of squared residuals to quantify calibration accuracy. After the verification period ends, the cloud platform performs a calibration quality assessment on all collected sample points.

[0153] For each sample point Extract its measured liquid level value and compared with the reference liquid level value obtained by other means at the same time. Compare them.

[0154] The reference liquid level value can be obtained synchronously by a high-precision reference liquid level gauge temporarily set up on site, or by manual measurement or a known liquid level gauge. The platform calculates the residual sum of squares (RSS) using the following formula:

[0155]

[0156] in, The total number of sample points participating in the evaluation, and ; For the first The measured liquid level values ​​at each sample point after calibration are in meters. For the first The reference liquid level value corresponding to each sample point is in the same unit as the measured value; the unit of RSS is square meters, which characterizes the overall deviation between the calibrated measured value and the true value.

[0157] Step S506: Perform subsequent processing based on the evaluation results. The platform compares the calculated RSS value with a preset residual threshold, which is determined according to the measurement accuracy requirements, for example, set to 0.001 square meters. If the RSS is less than the threshold, the remote calibration is considered successful. The system automatically updates the terminal's device operation file, recording the timestamp of this calibration, the old parameters before calibration, the new parameters written after calibration, and the verified RSS value, and marks the device status as normal.

[0158] If the RSS is greater than or equal to the threshold, the calibration effect is deemed unsatisfactory, and the platform automatically records a failed attempt. When the same terminal fails to achieve the required RSS after three consecutive calibration attempts, the system determines that the terminal may have physical hardware damage, severe contamination of the sensor diaphragm, or strong interference in the environment that cannot be compensated for by the algorithm. In this case, a fault alarm message is automatically generated and pushed to the mobile terminal of the maintenance personnel responsible for the area via an instant messaging protocol, prompting them to arrange on-site inspection and maintenance as soon as possible.

[0159] Through step S5 above, the system completes a full closed loop from command reception and parameter update to effect verification, ensuring the controllability and effectiveness of each remote calibration operation and providing a reliable data foundation for subsequent predictive maintenance.

[0160] To further enhance the system's intelligence, this embodiment also integrates a predictive maintenance module based on a long short-term memory neural network. This module, through in-depth analysis of the equipment's historical calibration records, can predict the performance degradation trend of the level gauge in advance and provide a suggested calibration time, thereby shifting from reactive maintenance to proactive prevention. The following detailed description, using a specific embodiment, elaborates on the construction, training, and application process of this predictive maintenance module.

[0161] First, constructing the training dataset is fundamental to neural network applications. The system extracts historical calibration records from the full lifecycle operational archives of each level gauge terminal stored in the cloud database. Each record contains relevant data for a calibration event, specifically including the timestamp of this calibration and the zero-point offset at the time of calibration. Full-scale gain error Number of ambient temperature cycles since the last calibration and the time interval between this calibration and the last calibration. This refers to the actual calibration cycle. The number of ambient temperature cycles is defined as the number of times the ambient temperature crosses a preset threshold range within a statistical period. For example, the process of the temperature rising from below 15 degrees Celsius to above 25 degrees Celsius and then falling back is counted as one cycle.

[0162] Arrange the above historical records in chronological order to form a multidimensional time series. For each time window, select continuous... The calibration record is used as the input feature, and the window length is... This can be set based on experience, for example, taking 10 calibration records. The feature vector of each calibration record contains the zero-point offset. Gain error Temperature cycle number Three values.

[0163] Therefore, the input to each training sample is a shape of... A two-dimensional matrix representing the past The device status change trajectory for each calibration. The sample label is the actual time interval between the next calibration that immediately follows this window. The output is a scalar representing the number of days remaining until the next calibration. A large number of training samples can be generated by sliding a window across the complete time series.

[0164] Next, we design the LSTM network structure. The input layer receiver shape is as follows: The time series data is then processed. One or more LSTM hidden layers are then connected to capture long-term dependencies and non-linear trends in the time series. In this embodiment, two LSTM layers are stacked. The first LSTM layer contains 128 memory units and returns the output sequence at each time step; the second LSTM layer contains 64 memory units and returns only the output of the last time step. A fully connected layer with 32 neurons is then connected after the LSTM layers, using the ReLU linear rectified activation function.

[0165] Finally, the output layer contains one neuron with no activation function; it directly outputs the predicted next calibration time interval. The unit is days. To avoid overfitting, Dropout layers can be added between LSTM layers and after fully connected layers, with a dropout rate set to 0.2.

[0166] The loss function is designed as the mean squared error (MSE) between the predicted and actual values, and its calculation formula is as follows:

[0167]

[0168] in The total number of training samples, For the first The prediction calibration interval for each sample. This corresponds to the actual calibration interval. The goal of training is to minimize this loss function by optimizing the network weights, that is, to make the predicted next calibration time as close as possible to the actual calibration time. The optimizer used is the Adam adaptive moment estimator, with an initial learning rate set to 0.001, a training epoch of 100, and an early stopping method is used to terminate training when the validation set loss no longer decreases for 10 consecutive epochs to prevent overfitting.

[0169] During training, the dataset is divided into training, validation, and test sets in a 7:2:1 ratio, which are used for model learning, hyperparameter tuning, and final performance evaluation, respectively.

[0170] After training, the trained LSTM model is deployed to a cloud platform. For each operational level gauge terminal, the system periodically extracts its most recent... The calibration records are used to construct the input feature matrix, which is then input into the model for forward computation to obtain the predicted next calibration interval. .

[0171] The predicted value is compared with the time difference between the current time and the last calibration. If the remaining time is less than a preset lead time, such as 15 days, the system automatically generates an early warning message to remind maintenance personnel to arrange the calibration plan in advance. The prediction results are also stored in the device file for subsequent model iteration and optimization.

[0172] In this way, the predictive maintenance module can provide personalized calibration suggestions based on the historical performance degradation trend of the equipment, thereby proactively intervening before measurement accuracy declines significantly, and significantly improving equipment uptime and maintenance efficiency.

[0173] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0174] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A remote calibration method for a pressure level gauge based on the Internet of Things, characterized in that, Includes the following steps: On the cloud platform side, it receives pressure values, medium temperature values, and ambient atmospheric pressure values ​​with timestamps uploaded by field terminals, forming the original time series; The original time series is preprocessed, including: using adaptive median filtering to remove impulse noise, and dynamically expanding the filter window length based on the comparison between the absolute value of the first difference of the pressure signal and the preset fluctuation threshold; then using a moving average algorithm based on Gaussian distribution function to perform secondary smoothing on the signal to suppress high-frequency random noise and reduce phase lag, resulting in the preprocessed data series. Based on the preprocessed data sequence, a nonlinear environmental compensation model is constructed using the recursive least squares method with a forgetting factor. The input vector of the nonlinear environmental compensation model includes at least the net pressure value, temperature value and its higher-order terms, so as to dynamically track the nonlinear drift of the sensor caused by environmental changes and long-term operation, and output real-time correction coefficients. Based on the real-time correction coefficients, the zero-point offset, full-scale gain error, and linearity deviation are calculated, and a calibration control command containing the zero-point correction value, slope compensation coefficient, and linearized polynomial parameters is generated. The calibration control command is sent to the field terminal to implement remote parameter updates; Specifically, the adaptive median filtering for eliminating impulse noise includes: calculating the absolute value of the first-order difference of the pressure signal in real time; when the absolute value of the first-order difference exceeds a preset fluctuation threshold, dynamically expanding the default filter window length from 3 sampling points to 7 sampling points; sorting all the sampling point values ​​within the expanded window, and selecting the median of the sorted values ​​as the filter output value at the current time point. The specific steps for constructing a nonlinear environmental compensation model using the recursive least squares method with a forgetting factor include: constructing an input vector, which includes a bias term, net pressure value, temperature value, a squared term of net pressure value, a squared term of temperature, and a product term of net pressure value and temperature; introducing a forgetting factor to assign decreasing weights to historical data, and updating the model parameter vector through recursive calculation to make the model output approximate the true liquid level value; wherein, the net pressure value is obtained by subtracting the real-time collected ambient atmospheric pressure from the pressure value; The specific steps of generating the calibration control command, which includes the zero-point correction value, slope compensation coefficient, and linearized polynomial parameters, include: [The text abruptly ends here, likely due to an incomplete sentence or missing information.] Set it as the opposite of the zero-point offset; calculate the slope compensation coefficient based on the full-scale gain error. The calculation formula is: ,in The full-scale gain error is calculated by performing polynomial fitting on the deviation data at multiple preset standard liquid level points within the liquid level measurement range to obtain linearized polynomial parameters. The deviation data is the difference between the ideal output value and the actual output value corresponding to the standard liquid level.

2. The remote calibration method for a pressure level gauge based on the Internet of Things according to claim 1, characterized in that, A secondary smoothing of the signal is performed using a moving average algorithm with weights assigned based on a Gaussian distribution function, specifically including: Set a length of Sliding window, The value of is odd; Based on the time distance between each sampling point in the window and the center point of the window, the corresponding weight is calculated using a Gaussian function, so that the data closer to the center of the window has a greater weight. The original pressure values ​​of each sampling point within the window are multiplied by their corresponding weights, summed, and then divided by the sum of all weights to obtain the filtered output value at the current moment.

3. The remote calibration method for a pressure level gauge based on the Internet of Things according to claim 1, characterized in that, After the calibration control command is sent to the field terminal, the process also includes a closed-loop verification step performed by the field terminal: The field terminal receives and decrypts the calibration control command, and writes the new calibration parameters into the non-volatile memory; Enter self-test mode, collect the current pressure signal, medium temperature and ambient atmospheric pressure, and apply new calibration parameters to correct the uncalibrated liquid level value, obtain the calibrated measured liquid level value and upload it to the cloud platform; The cloud platform receives multiple calibrated measured liquid level values ​​within a preset time period, compares them with the corresponding reference liquid level values, and calculates the sum of squared residuals. The residual sum of squares is compared with a preset residual threshold. If the residual sum of squares is less than the threshold, the calibration is considered successful; if it is greater than or equal to the threshold, the calibration is considered unsuccessful, and a fault alarm is triggered when the number of consecutive failures reaches a preset value.

4. The remote calibration method for a pressure level gauge based on the Internet of Things according to claim 1, characterized in that, The method also includes predictive maintenance steps: Extract historical calibration records of the level gauge terminal from the cloud database. Each record includes at least the calibration timestamp, zero offset, full-scale gain error, and number of ambient temperature cycles since the last calibration. By using multiple consecutive calibration records as input features, training samples are constructed and fed into a long short-term memory neural network for training to obtain a predictive maintenance model. Using a trained predictive maintenance model, based on the most recent calibration records of the current terminal, the time interval for the next recommended calibration is predicted, and an early warning message is generated when the remaining time is less than a preset lead time.

5. The remote calibration method for a pressure level gauge based on the Internet of Things according to claim 4, characterized in that, The number of ambient temperature cycles is defined as the number of times the ambient temperature crosses a preset threshold range within a statistical period.

6. The remote calibration method for a pressure level gauge based on the Internet of Things according to claim 1, characterized in that, Before receiving the timestamped pressure value, medium temperature value, and ambient atmospheric pressure value uploaded by the field terminal, the process also includes: The liquid level gauge terminal is equipped with a high-precision pressure sensing unit, a platinum resistance temperature sensor, and a MEMS absolute pressure sensor to simultaneously collect pressure, medium temperature, and ambient atmospheric pressure. Each frame of data collected is timestamped using a GPS or BeiDou timing module. The timestamped data is encapsulated and a checksum is generated using the CRC-16 cyclic redundancy check algorithm. The data is then transmitted to the cloud platform via a narrowband IoT protocol.

7. A remote calibration system for a pressure level gauge based on the Internet of Things (IoT), applied to the remote calibration method for a pressure level gauge based on the IoT described in any one of claims 1-6, characterized in that: This includes on-site terminals and cloud platforms; The field terminal includes: A high-precision pressure sensing unit is used to collect pressure values; Platinum resistance temperature sensor, used to acquire medium temperature values; MEMS absolute pressure sensor, used to collect ambient atmospheric pressure values; The time synchronization module is used to add a high-precision timestamp to each frame of data collected; The first processing unit is used to encapsulate the collected raw data and upload it to the cloud platform through the narrowband IoT communication module; and to receive and execute the calibration control instructions issued by the cloud platform and write the new calibration parameters into the non-volatile memory. The cloud platform includes: The data receiving module is used to receive pressure, temperature and atmospheric pressure data with timestamps uploaded by the field terminal; The data preprocessing module is used to perform adaptive median filtering and Gaussian weighted moving average filtering on the original time series. The model building module is used to construct a nonlinear environmental compensation model based on preprocessed data using the recursive least squares method with a forgetting factor, and output real-time correction coefficients. The calibration decision module is used to calculate the zero-point offset, full-scale gain error and linearity deviation based on real-time correction coefficients, and generate calibration control instructions that include zero-point correction values, slope compensation coefficients and linearized polynomial parameters. The command issuance module is used to encrypt calibration control commands and then issue them to the corresponding field terminals. The closed-loop verification module is used to receive the calibrated measurement values ​​fed back by the field terminal, calculate the sum of squares of the residuals between the measured values ​​and the reference liquid level values, and determine whether the calibration is successful based on the comparison results with the preset threshold. The predictive maintenance module is used to predict the next recommended calibration interval and generate early warning information based on the zero offset, full-scale gain error and ambient temperature cycle number in the historical calibration records, using a long short-term memory neural network.

Images