A cathodic protection intelligent monitoring control system

Abstract

The present application relates to a kind of cathodic protection intelligent monitoring control system, specifically relates to pipeline protection field, the system significantly improves the monitoring accuracy and reliability of pipeline cathodic protection under complex electromagnetic interference environment, through high fidelity synchronous acquisition and intelligent noise reduction technology ensure original data quality, utilize the inductive voltage drop distribution generated by dynamic updating digital twin body accurate simulation stray current interference, combined with the hybrid analytical model of deep learning and physical law fusion realizes the real-time analysis of polarization potential without power-off, and based on uncertainty quantification constructs self-diagnosis and self-optimization closed loop, effectively eliminates the risk of protection state misjudgment caused by signal distortion in traditional method, significantly reduces the probability of pipeline corrosion leakage accident, while reducing artificial maintenance cost, forms the unmanned safe protection system of state perception, intelligent decision, dynamic optimization.

Description

Technical Field

[0001] This invention relates to the field of pipeline protection, and more specifically, to an intelligent monitoring and control system for cathodic protection. Background Technology

[0002] Cathodic protection systems are a key technology for protecting buried metal pipelines, storage tanks, and other infrastructure from corrosion. In practical engineering applications, these systems are often deployed in areas with exceptionally complex electromagnetic environments, such as gas pipelines crossing urban subway networks, long-distance crude oil pipelines located beneath high-voltage direct current transmission lines, or industrial pipelines near variable frequency drive equipment areas of large chemical plants. In these scenarios, the start-up and shutdown of rail transit, drastic fluctuations in power grid load, and the operation of high-power electrical equipment generate stray currents with high amplitude and rapid changes in the soil. This causes the surface potential field along the pipeline to exhibit continuous and irregular dynamic fluctuations, posing a severe challenge to the accurate measurement of cathodic protection potential.

[0003] Currently, the technical bottleneck in this field lies in how to accurately obtain the true polarization potential of the pipeline under such strong dynamic interference environment. The industry generally uses a synchronous interruptor in conjunction with a potentiostat to perform instantaneous power-off measurement in order to eliminate the ohmic drop interference generated by the current flowing in the medium, thereby measuring the power-off potential without the influence of IR drop. Another auxiliary method is to use hardware or software filters with fixed parameters to smooth the continuously measured potential signal. However, synchronous interruption technology relies on capturing the transient response of the potential in the extremely short time when the current returns to zero. When stray currents change rapidly, it is difficult to achieve accurate time synchronization. Its measurement results may still contain residual interference, and frequent switching operations may affect the system. On the other hand, conventional filtering methods introduce significant phase lag while suppressing noise. They also eliminate the dynamic characteristics of the signal while filtering out harmful noise, resulting in delayed response and an inability to reflect the real changes in the protection status in real time. These methods all have inherent limitations and cannot achieve online and accurate separation of the polarization potential component from the total measured potential containing strong ohmic drops without relying on physical power disconnection. Therefore, developing a new technology that can adapt to extreme dynamic interference environments and rely on high-frequency data acquisition and advanced intelligent algorithms to achieve real-time or near-real-time accurate estimation and restoration of polarization potential has become an urgent need to improve the monitoring reliability of cathodic protection systems and ensure the safe operation of pipelines. Summary of the Invention

[0004] This invention addresses the technical problems existing in the prior art by providing an intelligent monitoring and control system for cathodic protection, thereby resolving the issues raised in the background section.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: Specifically, it includes: a data acquisition and synchronization module, a digital twin construction module, a hybrid dynamic parsing module, and an optimization feedback management module, wherein...

[0006] Data acquisition and synchronization module: It is used to synchronously acquire potential and current data on the pipeline and raw data of the external environment through multiple data acquisition units deployed in a wide area, and perform adaptive noise reduction preprocessing on the acquired raw data to output time-synchronized multi-dimensional time series data.

[0007] Digital twin construction module: connected to the data acquisition and synchronization module, used to receive interference source data in the multidimensional time series data, construct a parameterized finite element model based on the geometric and physical parameters of the pipeline network, obtain the IR drop distribution field data of the entire pipeline by solving the electromagnetic field equation, and update and verify the finite element model parameters using measured data;

[0008] Hybrid dynamic analysis module: connected to the data acquisition synchronization module and the digital twin construction module, used to receive the multidimensional time series data and the IR drop distribution field data, establish a dynamic nonlinear mapping relationship between the interference source and the IR drop through the long short-term memory network, and use the Kalman filter algorithm to optimize and correct the output of the long short-term memory network based on physical laws, and finally output the polarization potential estimate after stripping the IR drop in real time;

[0009] Optimized Feedback Management Module: Connected to the Hybrid Dynamic Analysis Module, it is used to perform uncertainty quantification analysis on the polarization potential estimate. When the uncertainty is lower than a set threshold, it outputs the final result. When the uncertainty is higher than the set threshold, it triggers an alarm and feeds back the current data context to the digital twin construction module to drive it to generate new training samples.

[0010] In a preferred embodiment, the specific operation of synchronously acquiring raw data through multiple data acquisition units deployed over a wide area in the data acquisition synchronization module is as follows:

[0011] When the preset acquisition cycle is reached, the data acquisition units distributed at key nodes along the pipeline start synchronous acquisition. Each data acquisition unit relies on its built-in precision time protocol as a unified clock source to drive the analog-to-digital converter to synchronously sample the original AC and DC potential signals, the original AC and DC current density signals, and the original soil reference potential signals on the pipeline. It also accesses the original data of the external environment context through the Internet of Things interface and assigns a globally unified high-precision timestamp to each acquired data point, thereby forming a timestamped original potential sequence, a timestamped original current sequence, a timestamped original soil potential sequence, and a timestamped external environment data sequence.

[0012] In a preferred embodiment, the specific process of adaptive noise reduction preprocessing for the collected raw data is as follows:

[0013] For each signal channel in the original potential sequence, original current sequence, and original soil potential sequence with timestamps, the original signal of the channel is first decomposed into multiple different sub-bands using wavelet packet transform, thus obtaining a set of wavelet packet coefficients corresponding to each sub-band. Next, the Shannon entropy value of the wavelet packet coefficients in each sub-band is calculated. Then, the noise reduction threshold of the sub-band is dynamically calculated based on the entropy value. Finally, the wavelet packet coefficients of each sub-band after the above thresholding are used to perform inverse wavelet packet transform to reconstruct the noise-reduced signal, thereby obtaining time-synchronized multidimensional time series data, including preprocessed pipeline potential signals, preprocessed pipeline current signals, and preprocessed external environment data.

[0014] In a preferred embodiment, the specific operation of constructing a parametric finite element model based on the geometric and physical parameters of the pipeline network in the digital twin construction module is as follows:

[0015] The digital twin construction module reads the geometric and physical parameters of the pipeline, including the pipeline route, burial depth, coating leakage resistance distribution, and initial soil stratification structure from GIS, and automatically constructs a three-dimensional finite element geometric model of the pipeline network and surrounding soil medium. The soil is treated as a non-uniform medium, and its electrical conductivity distribution field is defined as the core parameterized variable to be inverted and updated. This variable is set with an initial value based on historical geological survey report data during model initialization, thereby completing the construction of a parameterized finite element model whose soil electrical conductivity properties can be dynamically updated.

[0016] In a preferred embodiment, the specific operation of updating and verifying the finite element model parameters using measured data is as follows:

[0017] First, interference source data is extracted from the multidimensional time-series data transmitted from the data acquisition and synchronization module, and injected as a current boundary condition into the parameterized finite element model. The Laplace equation for the electromagnetic field is solved numerically to simulate and calculate the IR drop distribution field data of the pipeline in the entire soil space. The interference source data and the induced voltage drop distribution field data are then constructed as an "input-output" sample pair. Subsequently, a Bayesian inversion algorithm based on a Gaussian process surrogate model is used to optimize and update the soil conductivity distribution parameters. This algorithm first constructs a Gaussian process surrogate model, which learns the aforementioned parameters a finite number of times. The "input-output" sample pairs generated by the simulation calculation establish a nonlinear mapping relationship from the conductivity distribution parameters to the simulated potential value, replacing the computationally expensive forward finite element calculation. Then, the measured pipeline potential signal data obtained after preprocessing by the data acquisition synchronization module is compared with the simulation results of the current model. With the goal of minimizing the difference between the two, the optimal conductivity distribution parameters under the maximum a posteriori probability are solved within the Bayesian optimization framework, incorporating the prior knowledge distribution of soil conductivity as a constraint. Finally, the optimal parameters are updated back into the parameterized finite element model to complete the calibration and verification of the model parameters.

[0018] In a preferred embodiment, the specific process of establishing a dynamic nonlinear mapping relationship between the interference source and the IR drop through the long short-term memory network in the hybrid dynamic analysis module is as follows:

[0019] First, an IR decrease sequence corresponding to the pipeline monitoring point location from the IR decrease distribution field data received from the digital twin construction module, and interference source data with time stamps extracted from the multidimensional time series data received from the data acquisition synchronization module, are combined to form a hybrid training dataset. Then, this hybrid training dataset is used to train a Long Short-Term Memory (LSTM) network model. This network, through its internal forgetting gate, input gate, cell state update mechanism, and output gate gating structure, learns and memorizes the complex, long-term dependent, dynamic nonlinear mapping between interference source data and IR decrease data. Specifically, the forgetting gate determines which information to discard from the cell state, the input gate determines which new information will be stored in the cell state, the cell state is updated based on the outputs of the forgetting and input gates, and the output gate determines the final hidden state output based on the current cell state. After training, the network is deployed for online inference.

[0020] In a preferred embodiment, the specific process of using the Kalman filter algorithm to optimize and correct the output of the Long Short-Term Memory network based on physical laws is as follows:

[0021] The IR drop predicted by the Long Short-Term Memory (LSTM) network based on real-time interference source data is used as the observation input for the Kalman filtering process. Simultaneously, the true polarization potential of the pipeline corrosion protection state is defined as the state variable of the Kalman filter, and a state equation is constructed based on its slowly changing physical laws. This equation includes the estimated value of the polarization potential at the previous moment and a process noise term. An observation equation is established based on the physical relationship between the preprocessed measured pipeline potential signal data in the multidimensional time series data and the sum of the true polarization potential, the IR drop predicted by the LSTM network, and an observation noise term. Finally, the output of the LSTM network is optimized and corrected by recursively executing the prediction and update steps of the Kalman filter. The prediction step is used to calculate the prior estimate of the polarization potential at the current moment and the prior estimate error covariance. The update step is used to calculate the Kalman gain, update the posterior state estimate and the posterior error covariance, and finally output the optimal estimated value of the polarization potential after IR drop removal in real time.

[0022] In a preferred embodiment, the specific process of performing uncertainty quantification analysis on the polarization potential estimate in the optimization feedback management module is as follows:

[0023] The system receives polarization potential estimates from the hybrid dynamic analysis module and employs a Bayesian neural network inference method based on Monte Carlo random sampling. It performs multiple forward propagations on each polarization potential estimate to obtain a set of output distributions. The discrete variance of this output distribution is calculated to quantify the cognitive uncertainty caused by insufficient cognition in the model. At the same time, the variance term of the inherent output of the neural network is extracted to quantify the inherent random uncertainty of the data itself. Finally, the two uncertainties mentioned above are weighted and fused according to preset weights to calculate the total uncertainty quantification value of each polarization potential estimate.

[0024] In a preferred embodiment, the step of outputting the final result when the total uncertainty quantification value is lower than a set threshold and triggering an alarm when the total uncertainty quantification value is higher than the set threshold specifically includes: comparing the calculated total uncertainty quantification value with a dynamically adjusted decision threshold, which is calculated and updated on a rolling basis according to the statistical characteristics of recent historical uncertainty data;

[0025] If the total uncertainty quantization value is less than or equal to the dynamic threshold, the current polarization potential estimate is determined to be reliable, and it is output as the final result to the monitoring terminal.

[0026] If the total uncertainty quantification value exceeds the dynamic threshold, a high uncertainty alarm signal is immediately triggered, and the complete data context at the current moment is locked and packaged.

[0027] In a preferred embodiment, the step of feeding back the current data context to the digital twin construction module to drive it to generate new training samples specifically includes: sending the data context packet locked when the alarm is triggered, which contains a timestamp, interference source data, and preprocessed measured data of the pipeline potential signal from the multidimensional time series data, to the digital twin construction module; after receiving the data, the digital twin construction module uses the interference source data in the data packet as boundary conditions to drive its internal high-fidelity parametric finite element model to perform a forward simulation calculation, outputting the accurate IR drop distribution data corresponding to this moment, thereby generating a new set of "interference source-IR drop" paired training samples; the new samples are automatically added to the training dataset to drive the incremental learning and optimization of the hybrid dynamic analysis module.

[0028] The beneficial effects of this invention are: the system significantly improves the monitoring accuracy and reliability of pipeline cathodic protection under complex electromagnetic interference environments; it ensures the quality of raw data through high-fidelity synchronous acquisition and intelligent noise reduction technology; it accurately simulates the distribution of induced voltage drop caused by stray current interference using dynamically updated digital twins; it achieves real-time analysis of polarization potential without power interruption by combining a hybrid analytical model that integrates deep learning and physical laws; and it constructs a self-diagnosis and self-optimization closed loop based on uncertainty quantification, effectively eliminating the risk of misjudgment of protection status caused by signal distortion in traditional methods, greatly reducing the probability of pipeline corrosion and leakage accidents, while reducing manual maintenance costs, and forming an unattended safety protection system that integrates status perception, intelligent decision-making, and dynamic optimization. Attached Figure Description

[0029] Figure 1 This is a flowchart of the method of the present invention;

[0030] Figure 2 This is a block diagram of the system structure of the present invention. Detailed Implementation

[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0032] In the description of this application, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the stated features. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0033] In the description of this application, the term "for example" is used to mean "used as an example, illustration, or description." Any embodiment described as "for example" in this application is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is provided to enable any person skilled in the art to make and use the invention. Details are set forth in the following description for purposes of explanation. It should be understood that those skilled in the art will recognize that the invention can be made without using these specific details. In other instances, well-known structures and processes will not be described in detail to avoid obscuring the description of the invention with unnecessary detail. Therefore, the invention is not intended to be limited to the embodiments shown, but is consistent with the broadest scope of the principles and features disclosed in this application.

[0034] Example 1

[0035] This embodiment provides, for example Figure 1-2 The cathode protection intelligent monitoring and control system shown includes: a data acquisition and synchronization module, a digital twin construction module, a hybrid dynamic analysis module, and an optimization feedback management module.

[0036] Data Acquisition Synchronization Module: This module is used to synchronously acquire potential and current data from the pipeline, as well as raw data from the external environment, through multiple data acquisition units deployed over a wide area. It also performs adaptive noise reduction preprocessing on the acquired raw data to output high-precision, time-synchronized multidimensional time series data. Traditional data acquisition methods are limited by insufficient synchronization accuracy and weak anti-interference capabilities, leading to distortion of input for subsequent analysis. Therefore, this module aims to fundamentally solve this problem through wide-area synchronous acquisition and adaptive spectral clarity noise reduction technology based on information entropy, providing downstream modules with high-fidelity, highly synchronized clean data.

[0037] Digital Twin Construction Module: Connected to the data acquisition and synchronization module, this module receives interference source data from multi-dimensional time series data and constructs a parameterized finite element model based on the geometric and physical parameters of the pipeline network. It then simulates and calculates the IR drop distribution field data across the entire pipeline area by solving the electromagnetic field equations. The module updates and verifies the finite element model parameters using measured data. Since traditional static models cannot cope with the dynamic changes in the soil environment and interference sources, leading to a mismatch between simulation and measured data, this module constructs a parameterized finite element digital twin that evolves synchronously with the physical world and is dynamically updated. Its core lies in employing a Bayesian inversion framework and surrogate model technology to achieve online and accurate estimation of the spatial distribution of soil conductivity. This allows for high-fidelity simulation and calculation of the IR drop distribution field across the entire pipeline area under dynamic interference, providing a crucial prior knowledge base and training data source for the subsequent intelligent analysis module.

[0038] Hybrid Dynamic Analysis Module: Connected to the data acquisition synchronization module and the digital twin construction module, this module receives multi-dimensional time-series data and IR drop distribution field data. It establishes a dynamic nonlinear mapping relationship between the interference source and the IR drop through a Long Short-Term Memory (LSTM) network, and uses a Kalman filter algorithm to optimize and correct the LSM network output based on physical laws. Finally, it outputs the estimated polarization potential after IR drop removal in real time. Traditional methods either rely on physical power-off (synchronization interruption method) or fixed filtering, which are ineffective under complex dynamic interference. Therefore, this module utilizes the LSM network to capture the highly nonlinear and dynamic temporal relationship between interference and IR drop, and introduces Kalman filtering based on the physical evolution of polarization potential to constrain and optimize the network output, ultimately achieving high-precision and high-reliability online analysis without the need for physical power-off.

[0039] The optimized feedback management module connects to the hybrid dynamic analysis module and is used to perform uncertainty quantification analysis on the polarization potential estimate. When the uncertainty is below a set threshold, the final result is output. When the uncertainty is above the set threshold, an alarm is triggered and the current data context is fed back to the digital twin construction module to drive it to generate new training samples. Since traditional systems lack an evaluation and closed-loop optimization mechanism for the credibility of their own output, this module aims to build an intelligent management unit with "self-awareness" capabilities. By quantifying and evaluating the uncertainty of the estimation results, it achieves a closed-loop control of reliable result output, real-time risk alarm, and system self-optimization, ultimately ensuring the system's reliability and continuous evolution capability in complex environments.

[0040] In this embodiment, it is specifically necessary to explain the specific operation of the data acquisition synchronization module in which multiple data acquisition units deployed in a wide area synchronously acquire raw data as follows:

[0041] When the system initializes or reaches the preset acquisition cycle, the data acquisition units distributed at key nodes along the pipeline start synchronous acquisition. Each data acquisition unit relies on its built-in precision time protocol or high-precision GPS timing module as a unified clock source to drive the analog-to-digital converter to process the raw AC and DC potential signals on the pipeline. AC and DC current density raw signals The raw soil reference potential signal is sampled synchronously and accessed via an IoT interface to obtain raw data of the external environmental context. (Such as subway traction power supply curves and substation load data, accessed via IoT interfaces); the synchronization process ensures that the sampling clock deviation of all data acquisition units is less than one microsecond, and assigns a globally unified high-precision timestamp to each acquired data point. This forms a timestamped original potential sequence. Original current sequence with timestamps Original soil potential sequences with timestamps and external environmental data sequences with timestamps. ;

[0042] The specific process of adaptive noise reduction preprocessing for the collected raw data is as follows:

[0043] For each signal channel in the original potential sequence, original current sequence, and original soil potential sequence with timestamps, wavelet packet transform is first applied to the original signal of this channel (e.g., ...). Decomposed into multiple different sub-band frequencies This allows us to obtain a set of wavelet packet coefficients corresponding to each sub-band frequency band. Next, calculate the frequency band for each sub-band. Shannon entropy of inner wavelet packet coefficients This entropy value is used to quantify the uncertainty or complexity of the signal within that sub-band. The formula for calculating the Shannon entropy value is:

[0044] ;

[0045] in, Indicates the first The Shannon entropy value of the wavelet packet coefficients of each frequency band; the higher the Shannon entropy value ( The larger the entropy value, the more complex and uncertain the signal in that sub-band is. It may contain a mixture of strong noise and useful signals, requiring careful handling to avoid filtering out useful signals as noise. The lower the entropy value, the more complex and uncertain the signal in that sub-band is. The smaller the value, the simpler the signal in this sub-band is, and it is very likely to be stable and simple background noise, which can be safely and strongly filtered. The index number represents the sub-band frequency, indicating which frequency band it is. For example, Represents the low-frequency band. Represents the high-frequency band. The index number of the wavelet coefficient indicates which coefficient is in a certain subband, and the summation symbol is used. It means adding up the contributions of all the coefficients in this sub-band. Indicates the first The first sub-band frequency band The energy efficiency of each coefficient is calculated as follows: ,in The total energy of this subband is represented by the following calculation method: Then, based on that entropy value Dynamically calculate the noise reduction threshold for this sub-band frequency band. The calculation process incorporates an S-shaped adjustment function, when... Very high (complex signal): The function value approaches 1, at which point... Approaching the classic universal threshold This is a relatively low threshold, the purpose of which is to retain more coefficients and prevent valid signals from being mistakenly deleted;

[0046] when Very low (simple signal, mostly noise): the function value approaches 0, which makes the entire The value becomes very large, thus forcefully filtering out most of the coefficients of the subband, achieving the purpose of efficient noise reduction;

[0047] The formula for calculating the noise reduction threshold is:

[0048] ;

[0049] in, Indicates the noise reduction threshold, used to determine the first... Wavelet coefficients of each frequency band It should be retained. Should it be discarded? The critical value of ) The estimated value of the noise standard deviation is calculated as follows: , This represents the number of sub-band coefficients, i.e., the number of... How many wavelet coefficients are there in total in each sub-band? This represents the general threshold value, which serves as the base value for the threshold in this formula. This represents the adjustment factor (size parameter), which controls the steepness of the function curve as it changes from low to high. The larger the value, the steeper the curve, meaning the more sensitive the system is to changes in entropy. This represents the adjustment factor (offset parameter), which controls the center position of the function curve. It can be understood as the "baseline" of the entropy value. At that time, the value of the sigmoid function was 0.5. The adjustment function represents the S-shape; finally, the wavelet packet coefficients of each sub-band after the above thresholding process are used. Perform inverse wavelet packet transform to reconstruct the denoised signal (e.g.) This yields high-precision, time-synchronized multidimensional time-series data, including preprocessed pipeline potential signals, preprocessed pipeline current signals, and preprocessed external environmental data. .

[0050] In this embodiment, it is specifically necessary to explain the following operation in the digital twin construction module: the construction of a parameterized finite element model based on the geometric and physical parameters of the pipeline network is as follows:

[0051] The digital twin construction module reads the geometric and physical parameters of the pipeline, including pipeline routing, burial depth, coating leakage resistance distribution, and initial soil stratification structure from GIS, and automatically constructs a three-dimensional finite element geometric model of the pipeline network and surrounding soil medium; and treats the soil as a non-uniform medium, its conductivity distribution field ( represents the electrical conductivity of soil, a physical quantity describing the soil's ability to conduct electricity. The higher the conductivity, the better the soil's conductivity, the larger the range of influence of stray currents, and the greater the IR drop. It is a core model parameter that needs to be inverted and updated in this step. This represents a spatial coordinate vector, signifying any point in three-dimensional space. In this model, it is used to locate every position within the pipes and soil. , Representing the specific components of spatial coordinates, for spatial coordinate vectors In a more concrete way, in a three-dimensional finite element model, each point needs to be determined by three coordinates to pinpoint its precise location:

[0052] x: represents the east-west direction or longitude coordinate;

[0053] y: represents the north-south direction or latitude coordinate;

[0054] z: represents the vertical depth direction or elevation coordinate;

[0055] therefore, In reality It means at a certain point in space. The soil electrical conductivity at the location is defined as the core parameter variable to be inverted and updated. This variable is initialized with an initial value based on historical geological survey report data during model initialization. In the subsequent "update verification" process, this initial value will be updated and replaced by the new value calculated by the optimization algorithm, thereby completing the construction of a parameterized finite element model whose soil electrical conductivity properties can be dynamically updated.

[0056] The specific steps for updating and verifying the parameters of the finite element model using measured data are as follows:

[0057] First, the multidimensional time series data transmitted from the data acquisition and synchronization module. Extracting interference source data ( (Stray current interference source) from (The current signal after cleaning) was analyzed and used as a current boundary condition in the parameterized finite element model. The Laplace equation for the electromagnetic field was solved numerically. (The partial differential equation describing the potential distribution under a quasi-static electromagnetic field is used when the soil conductivity distribution is known.) and boundary conditions (injected interference current) The potential within the entire field can be solved using numerical methods (such as the finite element method). Simulation calculations yielded IR drop distribution field data for the pipeline across the entire soil space. The interference source data and the induced voltage drop distribution field data were then used to construct an "input-output" sample pair. Subsequently, a Bayesian inversion algorithm based on a Gaussian process surrogate model was used to optimize and update the soil electrical conductivity distribution parameters. This algorithm first constructs a Gaussian process surrogate model. The model establishes a model based on the conductivity distribution parameters by learning from a finite number of input-output sample pairs generated by the aforementioned simulation calculations. to simulated potential value The nonlinear mapping relationship is This is used to replace the expensive finite element forward calculation; then, the measured data of the pipeline potential signal obtained after preprocessing by the data acquisition synchronization module ( The simulation results of the model are compared with those of the current model. With the goal of minimizing the difference between the two, the optimal conductivity distribution parameters in the sense of maximum a posteriori probability are solved within the Bayesian optimization framework, incorporating the prior knowledge distribution of soil conductivity as a constraint. The formula is:

[0058] ;

[0059] in, This represents the optimal conductivity distribution parameters obtained from the maximum a posteriori probability estimation. This represents the parameterized soil electrical conductivity distribution vector to be optimized. Its purpose is to be the direct objective of the inversion solution, and its optimal solution... Used to update digital twin models This represents a prior probability distribution, used to incorporate prior knowledge from geological exploration and other fields to constrain the solution space of inversion problems and avoid unrealistic solutions. This represents the likelihood function, used to quantize the current parameters. Below are the simulation results. Compared with measured data The degree of matching is usually defined based on a Gaussian distribution of residuals. Indicates proportional to, This represents the maximization operator; ultimately, the optimal parameters are updated back into the parameterized finite element model, completing the calibration and verification of the model parameters.

[0060] In this embodiment, it is specifically necessary to explain the process by which the dynamic nonlinear mapping relationship between the interference source and the IR drop is established through the long short-term memory network in the hybrid dynamic analysis module as follows:

[0061] First, the IR downdistribution field data received from the digital twin building module will be... IR drop sequence corresponding to the location of pipeline monitoring point and multidimensional time series data received from the data acquisition and synchronization module. The interference source data extracted from the time stamp. and the interference source data As input features, IR-decreasing sequences As target labels, they together constitute a mixed training dataset. Subsequently, this mixed training dataset was used to train a long short-term memory network model. During training, the network learns and memorizes interference source data through its internal forget gate, input gate, cell state update mechanism, and output gate gating structure. The complex, long-term dependent dynamic nonlinear mapping relationship between IR reduction data and IR reduction data; among which, the forget gate is used to determine which information to discard from the cell state, and its expression is:

[0062] ;

[0063] in, express The input vector at time 10:00, in practical applications, is the interference source data at that time. This is the latest information on network processing. The output of the forget gate is a function of the sigmoid function (i.e., ... The value generated is in the range (0,1), and its purpose is to view the hidden state at the previous time step. and current input and the cell state at the previous moment. Each element in the input determines a "retention ratio" (1 for complete retention, 0 for complete forgetting). The input gate determines which new information will be stored in the cell state, and its expression is:

[0064] ;

[0065] ;

[0066] in, This represents the output of the input gate, with a value range of (0,1). Its purpose is to determine the proportion of new information updated in the cell state. Representing candidate cell states, this is a new candidate value vector generated by the tanh function, ranging from (-1, 1). Its purpose is to indicate what new information might be added to the cell state. The cell state is updated based on the outputs of the forget gate and the input gate. The cell state update formula is: The output gate determines the final hidden state output based on the current cell state, and the expression is:

[0067] ;

[0068] ;

[0069] in, express The hidden state at any given moment contains information learned and extracted from historical sequences that is relevant to the current task. In practical applications, it carries information about how historical interference sources affect the IR drop. express The cell state at any given moment is the core "memory unit" of the LSTM network. It acts like a conveyor belt, passing information from the beginning of the sequence to the end. The gating mechanism's role is to carefully control which information is added to or removed from this conveyor belt. Representing the weight matrix and bias vector, these are the trainable parameters of the LSTM network. (representing f, i, C, o), during training, the network learns how to extract useful mapping patterns from the data by adjusting these tens of thousands of W and b values. This represents the Sigmoid function, used to compress any real number to the range (0,1), thereby generating a gating signal to simulate a "gating ratio". This represents the hyperbolic tangent activation function, used to compress any real number to the range (-1, 1). It is often used to standardize data to near the zero mean, as a candidate state or output. This represents the Hadamard product (element-wise multiplication), used to implement the "regulation" function of gating, such as the output of a forget gate. Compared to the previous cell state Element-by-element multiplication is used to achieve selective forgetting of historical memories. This represents the output of the output gate, with a value range of (0,1). Its purpose is to determine the current cell state based on the current input and historical information. Which parts of the output represent the current hidden state? After training, deploy the network for online inference.

[0070] The specific process of using the Kalman filter algorithm to optimize and correct the output of the Long Short-Term Memory network based on physical laws is as follows:

[0071] The IR degradation predicted by the Long Short-Term Memory network based on real-time interference source data. The observed input is used as the Kalman filter process; simultaneously, the true polarization potential of the pipeline corrosion protection state is defined as the state variable of the Kalman filter, and a state equation is constructed based on its slowly changing physical laws. This equation includes the estimated value of the polarization potential at the previous moment and a process noise term. The expression of the state equation is as follows:

[0072] ;

[0073] in, express The true polarization potential at any given moment, which is the state variable that needs to be optimally estimated, directly reflects the corrosion protection status of the pipeline. It is the true signal that needs to be separated from the total potential disturbed by stray currents. This represents the state transition matrix, describing how the state variable (polarization potential) evolves from the previous time step to the current time step. In this system, based on the physical law of "gradual change in polarization potential," this matrix is ​​usually set as an identity matrix (or close to an identity matrix), indicating that the algorithm assumes the polarization potential at the next time step is mainly the same as that at the previous time step. This represents the optimal estimate of the polarization potential at the previous moment. express The process noise term at time step 1 follows a normal distribution with zero mean and covariance Ql, i.e. Its purpose is to characterize the uncertainty of the modeling state equation, acknowledging that the understanding of the polarization potential evolution is not perfect and that there are unmodeled dynamics or small fluctuations. The larger the Ql, the greater the model uncertainty. It establishes an observation equation by establishing the physical relationship between the preprocessed pipe potential signal measured data in multidimensional time series data and the sum of the actual polarization potential, the IR reduction predicted by the long short-term memory network, and an observation noise term. The expression of the observation equation is:

[0074] ;

[0075] in, express Preprocessed measured data of pipeline potential signals from multidimensional time series data at various times. Indicates after integrating new data The optimal estimate (posterior estimate) at time. express The IR degradation predicted by the LSTM at time step is the IR degradation predicted by the Long Short-Term Memory network based on real-time interference source data. In the KF framework, it is compared with... Together they serve as input to the observation model. express The observed noise term at time t follows a normal distribution with a mean of zero and a covariance of Rl, i.e. Applications: Modeling the uncertainty of the observation equation, mainly reflecting the prediction error of the LSTM model and the residual measurement noise of the sensor; finally, the output of the Long Short-Term Memory network is optimized and corrected by recursively executing the prediction and update steps of the Kalman filter. The prediction step is used to calculate the prior estimate of the polarization potential at the current time and the covariance of the prior estimate error. The prediction formula for the prior estimate of the polarization potential at the current time is:

[0076] ;

[0077] in, Indicates the current (i.e.) The predicted value (prior estimate) of the polarization potential at time ). This represents the posterior state estimate from the previous time step, i.e., in The time is the optimal polarization potential estimate obtained after integrating all available information; it is the starting point for prediction at the current time.

[0078] The formula for predicting the prior estimate error covariance of the polarization potential at the current moment is:

[0079] ;

[0080] in, This represents the covariance of the prior estimation error, compared to the prior state estimation error. The corresponding uncertainty, This represents the covariance of the posterior estimation error at the previous time step (the uncertainty of the final estimate at the previous time step). This represents the process noise covariance matrix, which is the process noise... The covariance matrix is ​​used in the algorithm to adjust the degree of confidence the filter has in the predicted values ​​of the state equation. A larger value indicates greater model uncertainty, and the filter will become more dependent on the observed data. Represents the state transition matrix. This represents the transpose of the state transition matrix. The update step is used to calculate the Kalman gain, update the posterior state estimate, and the posterior error covariance. The update formula is:

[0081] ;

[0082] ;

[0083] ;

[0084] in, Represents the identity matrix. Representing the Kalman gain, it is a dynamically changing coefficient used to determine the extent to which current observation data should be trusted to correct prior predictions during the update step. It is calculated in real time based on P, Ql, and Rl. express The estimation error covariance matrix at time 1000 measures the posterior state estimate. The magnitude of uncertainty or error (P) is determined by the value of P. A smaller P value indicates that the current estimate is more certain and reliable. This represents the posterior state estimate, which is the optimal estimate of the current true polarization potential obtained by integrating prior predictions and current observation information. In other words, it is the final output, pure polarization potential stripped of IR drop. The inverse of the matrix is ​​used to solve for the Kalman gain. Finally, the optimal estimated value of the polarization potential after IR stripping is output in real time. .

[0085] In this embodiment, it is necessary to specifically explain the process of uncertainty quantification analysis of the polarization potential estimate in the optimization feedback management module as follows:

[0086] Receive polarization potential estimates from the hybrid dynamic analysis module. A Bayesian neural network inference method based on Monte Carlo random sampling is used to estimate each polarization potential. Execute multiple times (like Its purpose is to perform in the inference phase of a Bayesian neural network. (In each forward pass, some neurons are randomly shut down (discarded) to simulate Bayesian inference, thereby revealing the cognitive uncertainty within the model.) The forward pass yields a set of output distributions. The model quantifies cognitive uncertainty caused by insufficient cognition by calculating the discrete variance of the output distribution. Simultaneously, it extracts the variance term of the inherent output of the neural network to quantify the inherent random uncertainty of the data itself. Finally, the two types of uncertainty are weighted and fused according to preset weights to calculate the total uncertainty quantification value for each polarization potential estimate. The formula for calculating the total uncertainty quantification value is:

[0087] ;

[0088] in, express The total uncertainty at any given moment is a scalar value obtained by weighted fusion of cognitive and accidental uncertainties. Its purpose is to... Time-based polarization potential estimate A comprehensive and quantitative assessment of reliability will be directly incorporated into subsequent steps as the basis for intelligent decision-making. This represents the weighted fusion coefficient, with a value range of [0,1]. Its purpose is to balance the contribution ratios of cognitive uncertainty and random uncertainty to the final total uncertainty. For example, if there is greater concern about insufficient model cognition, it can be... Set it close to 1; if you are more concerned about data noise, you can set it close to 0. Representing random uncertainty, the Bayesian neural network model itself is the current estimate. The predicted variance value is used to characterize the noise inherent in the data that cannot be eliminated by adding training data. Examples include measurement noise from the sensor itself or unpredictable, transient, and severe disturbances in the environment. A high value indicates that even with a well-developed model, the output still has high uncertainty. This represents the variance calculation function, used to calculate the variance of a set of values. Its purpose is as follows: In this formula, its calculation object is... The output set of sub-Monte Carlo sampling The calculated variance value quantifies the degree of "confusion" of the model with respect to the current input. The larger the variance, the more inconsistent the model's internal judgment of this input, that is, the higher the cognitive uncertainty, usually because there is insufficient such patterns in the training data.

[0089] When the total uncertainty is quantified Below the set threshold The system outputs the final result in real time. When the total uncertainty quantification value is higher than the set threshold, an alarm is triggered. Specifically, the system compares the calculated total uncertainty quantification value with a dynamically adjusted decision threshold, which is calculated and updated on a rolling basis according to the statistical characteristics of recent historical uncertainty data.

[0090] If the total uncertainty quantization value is less than or equal to the dynamic threshold, the current polarization potential estimate is determined to be reliable, and it is output as the final result to the monitoring terminal.

[0091] If the total uncertainty quantification value is greater than the dynamic threshold, a high uncertainty alarm signal will be triggered immediately, and the complete data context at the current moment will be locked and packaged.

[0092] The current data context is fed back to the digital twin building module to drive it to generate new training samples. Specifically, this includes: the data context packet locked when an alarm is triggered. The data packet contains a timestamp. Interference source data Preprocessed pipe potential signal measured data from multidimensional time series data The data is sent to the digital twin construction module; upon receiving it, the digital twin construction module uses the interference source data in the data packet. As a boundary condition, it drives the high-fidelity parametric finite element model inside to perform a forward simulation calculation, outputting the accurate IR drop distribution data corresponding to this moment, thereby generating a new set of "interference source-IR drop" paired training samples. The new sample is automatically added to the training dataset to drive incremental learning and optimization of the hybrid dynamic parsing module.

[0093] It should be noted that the descriptions of each embodiment in the above embodiments have different focuses. For parts that are not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0094] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0095] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0096] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0097] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0098] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0099] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A cathodic protection intelligent monitoring and control system, characterized in that, Specifically, it includes: The module includes a data acquisition and synchronization module, a digital twin construction module, a hybrid dynamic parsing module, and an optimization feedback management module. Data acquisition and synchronization module: It is used to synchronously acquire potential and current data on the pipeline and raw data of the external environment through multiple data acquisition units deployed in a wide area, and perform adaptive noise reduction preprocessing on the acquired raw data to output time-synchronized multi-dimensional time series data. Digital twin construction module: connected to the data acquisition and synchronization module, used to receive interference source data in the multidimensional time series data, construct a parameterized finite element model based on the geometric and physical parameters of the pipeline network, obtain the IR drop distribution field data of the entire pipeline by solving the electromagnetic field equation, and update and verify the finite element model parameters using measured data; Hybrid dynamic analysis module: connected to the data acquisition synchronization module and the digital twin construction module, used to receive the multidimensional time series data and the IR drop distribution field data, establish a dynamic nonlinear mapping relationship between the interference source and the IR drop through the long short-term memory network, and use the Kalman filter algorithm to optimize and correct the output of the long short-term memory network based on physical laws, and finally output the polarization potential estimate after stripping the IR drop in real time; Optimized Feedback Management Module: Connected to the Hybrid Dynamic Analysis Module, it performs uncertainty quantification analysis on the polarization potential estimate. When the uncertainty is below a set threshold, it outputs the final result. When the uncertainty is above a set threshold, it triggers an alarm and feeds back the current data context to the digital twin construction module to drive it to generate new training samples.

2. The intelligent monitoring and control system for cathodic protection according to claim 1, characterized in that: In the data acquisition synchronization module, the specific operation of synchronously acquiring raw data through multiple data acquisition units deployed over a wide area is as follows: When the preset acquisition cycle is reached, the data acquisition units distributed at key nodes along the pipeline start synchronous acquisition. Each data acquisition unit relies on its built-in precision time protocol as a unified clock source to drive the analog-to-digital converter to synchronously sample the original AC and DC potential signals, the original AC and DC current density signals, and the original soil reference potential signals on the pipeline. It also accesses the original data of the external environment context through the Internet of Things interface and assigns a globally unified high-precision timestamp to each acquired data point, thereby forming a timestamped original potential sequence, a timestamped original current sequence, a timestamped original soil potential sequence, and a timestamped external environment data sequence.

3. The intelligent monitoring and control system for cathodic protection according to claim 2, characterized in that: The specific process of adaptive noise reduction preprocessing for the collected raw data is as follows: For each signal channel in the original potential sequence, original current sequence, and original soil potential sequence with timestamps, the original signal of the channel is first decomposed into multiple different sub-bands using wavelet packet transform, thus obtaining a set of wavelet packet coefficients corresponding to each sub-band. Next, the Shannon entropy value of the wavelet packet coefficients in each sub-band is calculated. Then, the noise reduction threshold of the sub-band is dynamically calculated based on the entropy value. Finally, the wavelet packet coefficients of each sub-band after the above thresholding are used to perform inverse wavelet packet transform to reconstruct the noise-reduced signal, thereby obtaining time-synchronized multidimensional time series data, including preprocessed pipeline potential signals, preprocessed pipeline current signals, and preprocessed external environment data.

4. The intelligent monitoring and control system for cathodic protection according to claim 3, characterized in that: In the digital twin construction module, the specific operation of constructing a parametric finite element model based on the geometric and physical parameters of the pipeline network is as follows: The digital twin construction module reads the geometric and physical parameters of the pipeline, including the pipeline route, burial depth, coating leakage resistance distribution, and initial soil stratification structure from GIS, and automatically constructs a three-dimensional finite element geometric model of the pipeline network and surrounding soil medium. The soil is treated as a non-uniform medium, and its electrical conductivity distribution field is defined as the core parameterized variable to be inverted and updated. This variable is set with an initial value based on historical geological survey report data during model initialization, thereby completing the construction of a parameterized finite element model whose soil electrical conductivity properties can be dynamically updated.

5. The intelligent monitoring and control system for cathodic protection according to claim 4, characterized in that: The specific operation of updating and verifying the finite element model parameters using measured data is as follows: First, interference source data is extracted from the multidimensional time-series data transmitted from the data acquisition and synchronization module, and injected as a current boundary condition into the parameterized finite element model. The IR drop distribution field data of the pipeline in the entire soil space is obtained by numerically solving the Laplace equation for the electromagnetic field, and the interference source data and the IR drop distribution field data are constructed as an "input-output" sample pair. Subsequently, a Bayesian inversion algorithm based on a Gaussian process surrogate model is used to optimize and update the soil conductivity distribution parameters. This algorithm first constructs a Gaussian process surrogate model, which learns from a finite number of the aforementioned simulation calculations. The generated "input-output" sample pairs establish a nonlinear mapping relationship from conductivity distribution parameters to simulated potential values, replacing the computationally expensive forward finite element calculations. Then, the measured pipeline potential signal data obtained after preprocessing by the data acquisition synchronization module is compared with the simulation results of the current model. With the goal of minimizing the difference between the two, and within a Bayesian optimization framework, prior knowledge of soil conductivity distribution is incorporated as a constraint to solve for the optimal conductivity distribution parameters in the sense of maximum a posteriori probability. Finally, these optimal conductivity distribution parameters are updated back into the parameterized finite element model, completing the calibration and verification of the model parameters.

6. The intelligent monitoring and control system for cathodic protection according to claim 5, characterized in that: In the hybrid dynamic analysis module, the specific process of establishing a dynamic nonlinear mapping relationship between the interference source and the IR drop through a long short-term memory network is as follows: First, an IR decrease sequence corresponding to the pipeline monitoring point location from the IR decrease distribution field data received from the digital twin construction module, and interference source data with time stamps extracted from the multidimensional time series data received from the data acquisition synchronization module, are combined to form a hybrid training dataset. Then, this hybrid training dataset is used to train a Long Short-Term Memory (LSTM) network model. This network, through its internal forgetting gate, input gate, cell state update mechanism, and output gate gating structure, learns and memorizes the complex, long-term dependent, dynamic nonlinear mapping between interference source data and IR decrease data. Specifically, the forgetting gate determines which information to discard from the cell state, the input gate determines which new information will be stored in the cell state, the cell state is updated based on the outputs of the forgetting and input gates, and the output gate determines the final hidden state output based on the current cell state. After training, the network is deployed for online inference.

7. The intelligent monitoring and control system for cathodic protection according to claim 6, characterized in that: The specific process of using the Kalman filter algorithm to optimize and correct the output of the Long Short-Term Memory network based on physical laws is as follows: The IR drop predicted by the Long Short-Term Memory (LSTM) network based on real-time interference source data is used as the observation input for the Kalman filtering process. Simultaneously, the true polarization potential of the pipeline corrosion protection state is defined as the state variable of the Kalman filter, and a state equation is constructed based on its slowly changing physical laws. This equation includes the estimated value of the polarization potential at the previous moment and a process noise term. An observation equation is established based on the physical relationship between the preprocessed measured pipeline potential signal data in the multidimensional time series data and the sum of the true polarization potential, the IR drop predicted by the LSTM network, and an observation noise term. Finally, the output of the LSTM network is optimized and corrected by recursively executing the prediction and update steps of the Kalman filter. The prediction step is used to calculate the prior estimate of the polarization potential at the current moment and the prior estimate error covariance. The update step is used to calculate the Kalman gain, update the posterior state estimate and the posterior error covariance, and finally output the optimal estimated value of the polarization potential after IR drop removal in real time.

8. The intelligent monitoring and control system for cathodic protection according to claim 7, characterized in that: The specific process of performing uncertainty quantification analysis on the polarization potential estimate in the optimization feedback management module is as follows: The system receives polarization potential estimates from the hybrid dynamic analysis module and employs a Bayesian neural network inference method based on Monte Carlo random sampling. It performs multiple forward propagations on each polarization potential estimate to obtain a set of output distributions. The discrete variance of this output distribution is calculated to quantify the cognitive uncertainty caused by insufficient cognition in the model. At the same time, the variance term of the inherent output of the neural network is extracted to quantify the inherent random uncertainty of the data itself. Finally, the two uncertainties mentioned above are weighted and fused according to preset weights to calculate the total uncertainty quantification value of each polarization potential estimate.

9. The intelligent monitoring and control system for cathodic protection according to claim 8, characterized in that: The process of outputting the final result when the total uncertainty quantification value is lower than the set threshold and triggering an alarm when the total uncertainty quantification value is higher than the set threshold specifically includes: comparing the calculated total uncertainty quantification value with a dynamically adjusted decision threshold, which is calculated and updated on a rolling basis according to the statistical characteristics of recent historical uncertainty data; If the total uncertainty quantification value is less than or equal to the dynamically adjusted decision threshold, the current polarization potential estimate is determined to be reliable, and it is output as the final result to the monitoring terminal. If the total uncertainty quantification value is greater than the dynamically adjusted decision threshold, a high uncertainty alarm signal is immediately triggered, and the complete data context at the current moment is locked and packaged.

10. The intelligent monitoring and control system for cathodic protection according to claim 9, characterized in that: The step of feeding back the current data context to the digital twin construction module to drive it to generate new training samples specifically includes: sending the data context packet locked when the alarm is triggered, which contains timestamps, interference source data, and preprocessed measured data of the pipeline potential signal from the multidimensional time series data, to the digital twin construction module; after receiving it, the digital twin construction module uses the interference source data in the data context packet as boundary conditions to drive its internal high-fidelity parametric finite element model to perform a forward simulation calculation, outputting the accurate IR drop distribution data corresponding to this moment, thereby generating a new set of "interference source-IR drop" paired training samples; the new training samples are automatically added to the training dataset to drive the incremental learning and optimization of the hybrid dynamic analysis module.

Images