A high-dimensional heterogeneous corrosion inhibitor synergistic optimization method and system

Abstract

This application provides a method and system for collaborative optimization of high-dimensional heterogeneous corrosion inhibitors. The method includes: performing dimensionality reduction, outlier removal, and feature extraction on multi-source heterogeneous data of industrial oil transportation pipelines to obtain a global state vector; combining this vector with a pre-built corrosion inhibitor prediction model to obtain a combination of corrosion inhibitor molecules; obtaining a target injection strategy based on the corrosion inhibitor molecule combination and a pre-built multi-agent collaborative optimization framework; converting the target injection strategy into a target tensor based on a multi-objective reward function using tensor decomposition; performing distributed privacy protection and multi-timescale optimization on the target tensor to obtain a target injection control sequence; and injecting the corrosion inhibitor into the industrial oil transportation pipeline according to the target injection sequence. Through the above technical solution, collaborative optimization of the entire corrosion inhibitor process is achieved.

Description

Technical Field

[0001] This application relates to the field of industrial pipeline corrosion protection technology, and in particular to a method and system for synergistic optimization of high-dimensional heterogeneous corrosion inhibitors. Background Technology

[0002] In the field of industrial oil pipeline operation, corrosion of the pipeline inner wall is a core hidden danger leading to equipment failure and crude oil leakage. It not only threatens production safety and causes environmental pollution, but also incurs huge maintenance and downtime losses. The effectiveness of corrosion prevention directly determines energy transmission efficiency and operating costs, making it one of the key factors restricting the long-term stable operation of industrial pipelines. To effectively inhibit corrosion, adding corrosion inhibitors is an economical and widely used mainstream protection technology. Although significant progress has been made in optimizing existing corrosion inhibitor application technologies, many challenges still exist in practical industrial applications.

[0003] First, the fusion of high-dimensional heterogeneous monitoring data is challenging. Multimodal data such as electrochemical impedance spectroscopy, thermal imaging, and acoustic emission are highly dimensional and exhibit significant modal differences, making it difficult for traditional methods to uncover potential correlations between data points. This leads to distorted corrosion status assessments and affects the accuracy of injection decisions. Second, the lack of a comprehensive collaborative optimization mechanism means that the design of corrosion inhibitor formulations, injection control, and supply chain scheduling are fragmented, failing to adapt to the coupled needs of dynamic operating conditions and making it difficult to balance protective effectiveness and economic efficiency. Furthermore, the adaptive capability to dynamic operating conditions is weak. Faced with extreme situations such as sudden changes in media composition and sensor failures, traditional control methods lack robustness and cannot achieve real-time iterative optimization of corrosion inhibitor injection strategies. Summary of the Invention

[0004] This application provides a high-dimensional heterogeneous corrosion inhibitor collaborative optimization method and system, which solves the problems of difficulty in multimodal spatiotemporal heterogeneous data fusion, incomplete modeling of multi-scale physical action mechanism, and high complexity of dynamic optimization decision under multi-objective constraints in the prior art, and realizes collaborative optimization of the entire corrosion inhibitor process.

[0005] To achieve the above objectives, the technical solution of this application embodiment is as follows: In a first aspect, embodiments of this application provide a method for synergistic optimization of high-dimensional heterogeneous corrosion inhibitors, the method comprising: The original data space is defined based on the multi-source heterogeneous data of the industrial oil transportation pipeline at the current moment, and the initial tensor parameters are obtained. The tensor parameters are then projected onto the low-dimensional space through a hybrid data fusion model to obtain the low-dimensional tensor parameters. The anomaly detection method based on Riemannian manifold learning removes invalid data from the low-dimensional tensor parameters and extracts features through a feature extraction model to obtain a global state vector. Based on the global state vector and combined with the pre-built corrosion inhibitor prediction model, a combination of corrosion inhibitor molecules is obtained; the corrosion inhibitor prediction model consists of a cross-scale computation framework, a graph network learning framework, and a molecular inverse design framework. Based on the corrosion inhibitor molecule combination, a target injection strategy is obtained using a pre-constructed multi-agent cooperative optimization framework; the multi-agent cooperative optimization framework includes a policy network and a value network. The multi-objective reward function based on tensor decomposition converts the target betting strategy into a target tensor. Distributed privacy protection and multi-timescale optimization are performed on the target tensor to obtain the target injection control sequence, and the corrosion inhibitor is injected into the industrial oil delivery pipeline according to the target injection sequence.

[0006] In one possible implementation, the corrosion inhibitor prediction model is optimized based on a learning-based dynamic optimization framework, a safety-constrained optimization framework, and an adaptive sample selection and active learning framework. The learning-based dynamic optimization framework embeds the optimization problem into a neural network model through a differential programming layer, supporting end-to-end backpropagation optimization; the safety-constrained optimization framework is based on stability theory and ensures the stable operation of the model by constructing a Lyapunov function; the adaptive sample selection and active learning framework is built based on the results of multi-precision evaluation and causal explanation conclusions, and selects high-value sample data by maximizing the information gain function.

[0007] In one possible implementation, the policy gradient formula of the multi-agent cooperative optimization framework is: ;in, For policy gradient, For the first Policy network of individual agents For the first The value network of individual agents For the agent to observe the state, This is a joint decision-making action.

[0008] In one possible implementation, the step of performing distributed privacy protection and multi-timescale optimization on the target tensor to obtain the target annotation control sequence includes: The target tensor is encrypted using homomorphic encryption technology, and the encrypted target tensor is then aggregated. The target annotation control sequence is obtained by adjusting the aggregated target tensor using a rolling temporal optimization method.

[0009] In one possible implementation, the method further includes: Based on the dynamic game model of the corrosion inhibitor supply network, a target inventory scheduling strategy is determined, and the inventory and scheduling costs of each node in the corrosion inhibitor supply network are adjusted based on the target inventory scheduling strategy. A combinatorial decision-making framework based on quantum optimization algorithms is used to optimize the supply path in the corrosion inhibitor supply network; The objective formula for path optimization is: ;in, For Hamiltonian.

[0010] In one possible implementation, the method further includes: The Betty number in different dimensions is determined by a persistent coherence method, and the presence of corrosion weak points in the industrial oil delivery pipeline is determined based on the Betty number. If the presence of corrosion weak points is determined and the number of corrosion weak points is greater than or equal to a preset number threshold, a fault warning is triggered.

[0011] In one possible implementation, the method further includes: The contribution of each global state vector to the target betting strategy is determined by the integral gradient method, so as to evaluate the degree of influence of each global state vector on the betting decision; A digital twin model is constructed based on the fluid motion control equation and the concentration distribution equation. The transport effect and distribution characteristics of the corrosion inhibitor in the industrial oil pipeline are simulated based on the digital twin model, and the injection strategy is adjusted based on the transport effect and distribution characteristics.

[0012] In one possible implementation, the method further includes: After the corrosion inhibitor is injected into the industrial oil pipeline according to the target injection sequence, multi-source heterogeneous data and the actual amount of corrosion inhibitor consumed at the next moment are collected. Based on the aforementioned multi-source heterogeneous data and the actual consumption of corrosion inhibitor, a real-time reward value is determined through multi-objective evaluation indicators; the reward value represents the comprehensive effectiveness of the refueling decision at the next moment. Based on the target betting strategy at the current moment, the betting strategy at the next moment, and the reward value, a complete set of experience data is formed. Combined with high-value sample data selected by the adaptive sample selection and active learning framework, a sample dataset is obtained. Data is extracted from the sample dataset according to a preset sampling strategy, and the multi-agent collaborative optimization framework is iteratively optimized.

[0013] In one possible implementation, the reward value is obtained using the following formula: ; in, The measured corrosion rate at time t+1 To preset the maximum allowable corrosion rate, The cost of corrosion inhibitor consumption at time t+1. Budget for cost per unit of time. The supply chain operating efficiency at time t+1 For ideal operating efficiency, The target weight coefficients are and satisfy the following conditions: .

[0014] Secondly, this application provides a high-dimensional heterogeneous corrosion inhibitor synergistic optimization system, the system comprising: a data acquisition device, a controller, and an execution device; the data acquisition device is connected to the controller, and the controller is connected to the execution device; The data acquisition device is used to collect multi-source heterogeneous data from industrial oil transportation pipelines and send it to the controller; The controller is used to execute the high-dimensional heterogeneous corrosion inhibitor synergistic optimization method described in the first aspect and generate control commands; The execution device is used to add corrosion inhibitor into the industrial oil delivery pipeline based on the control command.

[0015] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages: The technical solution of this application transforms the optimization of industrial corrosion inhibitor filling into a cross-scale coupled problem. It constructs a hybrid optimization framework that integrates microscopic mechanism analysis, mesoscopic mass transfer process simulation and intelligent decision-making algorithm. This framework can efficiently integrate high-dimensional heterogeneous monitoring data and achieve collaborative decision-making in corrosion inhibitor formulation design, filling control and supply chain scheduling. Under the premise of ensuring that the corrosion rate meets the standard, it significantly reduces the consumption of corrosion inhibitors and improves the economy and reliability of industrial pipeline corrosion protection. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments of this application or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 A flowchart of a high-dimensional heterogeneous corrosion inhibitor synergistic optimization method provided in this application embodiment; Figure 2 This is a block diagram of a high-dimensional heterogeneous corrosion inhibitor synergistic optimization system provided in an embodiment of this application. Detailed Implementation

[0018] 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, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0019] In the description of the embodiments of this application, it should be noted that the terms "center," "upper," "lower," "left," "right," "vertical," "horizontal," "inner," and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing the embodiments of this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. The terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; a mechanical connection or an electrical connection; a direct connection or an indirect connection through an intermediate medium; or a connection within two components. Those skilled in the art can understand the specific meaning of the above terms in the embodiments of this application according to the specific circumstances.

[0020] To overcome the limitations of existing technologies, this application proposes a high-dimensional heterogeneous corrosion inhibitor synergistic optimization method and system. Based on multi-source heterogeneous data from an industrial oil pipeline at the current moment, an original data space is defined to obtain initial tensor parameters. These tensor parameters are then projected onto a low-dimensional space using a hybrid data fusion model to obtain low-dimensional tensor parameters. An anomaly detection method based on Riemannian manifold learning is used to remove invalid data from these low-dimensional tensor parameters, and feature extraction is performed using a feature extraction model to obtain a global state vector. Based on this global state vector, combined with a pre-built corrosion inhibitor prediction model, the following is obtained: A corrosion inhibitor molecular combination is proposed. The corrosion inhibitor prediction model comprises a cross-scale computational framework, a graph network learning framework, and a molecular inverse design framework. Based on this corrosion inhibitor molecular combination, a target injection strategy is obtained using a pre-constructed multi-agent collaborative optimization framework. This multi-agent collaborative optimization framework includes a policy network and a value network. A multi-objective reward function based on tensor decomposition converts the target injection strategy into a target tensor. Distributed privacy protection and multi-timescale optimization are performed on this target tensor to obtain a target injection control sequence. The corrosion inhibitor is then injected into the industrial oil pipeline according to this target injection sequence. This technical solution achieves efficient fusion of high-dimensional heterogeneous data through tensor network decomposition, constructs a cross-stage collaborative decision-making framework using multi-agent adversarial learning, and combines a closed-loop optimization mechanism to achieve precise control of the corrosion inhibitor throughout its entire lifecycle. It can accurately adapt to dynamic industrial operating conditions, significantly reducing corrosion inhibitor consumption while ensuring the corrosion rate is stably controlled below a safe threshold, effectively improving the economy and reliability of corrosion protection for industrial pipelines and equipment.

[0021] Figure 1 This is a flowchart illustrating a synergistic optimization method for high-dimensional heterogeneous corrosion inhibitors provided in an embodiment of this application. Figure 1 As shown, the method may include the following steps.

[0022] S101. Define the original data space based on the multi-source heterogeneous data of the industrial oil transportation pipeline at the current moment, obtain the initial tensor parameters, and project the tensor parameters to the low-dimensional space through the hybrid data fusion model to obtain the low-dimensional tensor parameters.

[0023] For example, this multi-source heterogeneous data can be collected by the distributed edge computing node acquisition device of the industrial oil pipeline control system; this multi-source heterogeneous data can include pipeline length, pH value, etc. concentration, Factors such as partial pressure, pipeline temperature, pipeline pressure, and corrosion inhibitor dosage are not specified here.

[0024] S102. An anomaly detection method based on Riemannian manifold learning removes invalid data from the low-dimensional tensor parameters and extracts features through a feature extraction model to obtain the global state vector.

[0025] S103. Based on the global state vector and combined with the pre-built corrosion inhibitor prediction model, the corrosion inhibitor molecule combination is obtained.

[0026] The corrosion inhibitor prediction model consists of a cross-scale computational framework, a graph network learning framework, and a molecular reverse design framework.

[0027] S104. Based on the combination of corrosion inhibitor molecules, a target injection strategy is obtained using a pre-constructed multi-agent collaborative optimization framework.

[0028] This multi-agent collaborative optimization framework includes a policy network and a value network.

[0029] S105. The multi-objective reward function based on tensor decomposition converts the target betting strategy into a target tensor.

[0030] S106. Perform distributed privacy protection and multi-timescale optimization on the target tensor to obtain the target injection control sequence, and inject the corrosion inhibitor into the industrial oil delivery pipeline according to the target injection sequence.

[0031] The technical solution of this application transforms the optimization of industrial corrosion inhibitor filling into a cross-scale coupled problem. It constructs a hybrid optimization framework that integrates microscopic mechanism analysis, mesoscopic mass transfer process simulation and intelligent decision-making algorithm. This framework can efficiently integrate high-dimensional heterogeneous monitoring data and achieve collaborative decision-making in corrosion inhibitor formulation design, filling control and supply chain scheduling. Under the premise of ensuring that the corrosion rate meets the standard, it significantly reduces the consumption of corrosion inhibitors and improves the economy and reliability of industrial pipeline corrosion protection.

[0032] In some embodiments, S101 may include the following steps.

[0033] (1) Multi-source heterogeneous data of industrial oil transportation pipelines are collected based on distributed edge computing nodes, and modeling is carried out on the multi-source heterogeneous data. The original data space is defined to obtain the initial tensor parameters. The original data space is: , Indicates the first i Each monitoring point is t Moment K The first-order characteristic tensor, including the electrochemical impedance characteristic tensor. Multispectral thermal imaging feature tensor Acoustic emission event characteristic tensor Complex modal data, etc.

[0034] (2) Establish a hybrid data fusion model based on Tucker decomposition and tensor chain decomposition, and uniformly project the initial tensor parameters onto a low-dimensional subspace to obtain low-dimensional tensor parameters, thereby realizing dimensionality reduction and feature extraction of high-dimensional data, following the formula as follows: ;in, For the core tensor, A,B,C Given the factor matrices for each dimension, the tensor reconstruction error is minimized using an alternating least squares algorithm to solve the model. The tensor reconstruction error follows the formula below: ;in, The Frobenius norm is used to quantify the degree of difference between the original tensor and the decomposed and reconstructed tensor. The optimal decomposition parameters are obtained through iterative optimization and used for subsequent anomaly detection and feature analysis.

[0035] In some embodiments, S102 may include the following steps.

[0036] (1) An anomaly detection framework based on Riemannian manifold learning maps low-dimensional tensor parameters to symmetric positive definite matrix manifolds. Above, abnormal patterns are identified by calculating the Fraser distance between sample points and the center of the manifold. The Fraser distance follows the formula as follows: ;in, Let be the covariance matrix of the tensor mode expansion. The central matrix of the manifold, This is a matrix trace operation used to measure the degree to which a sample deviates from the normal data distribution, to filter and remove invalid monitoring data, and to ensure the reliability of the data in the modeling.

[0037] (2) Construct a multi-scale spatiotemporal feature extraction model. Through the fusion architecture of three-dimensional convolutional temporal memory units and graph node attention mechanism, the spatiotemporal correlation features of the data are mined. The multi-scale spatiotemporal feature extraction model is represented as follows: ;in, A The spatial adjacency matrix between monitoring points is used, and the attention coefficient is used to quantify the association weights between nodes.

[0038] Attention coefficient Follow the formula as follows: ;in, For attention weight vectors, The characteristic transformation matrix, For node feature vectors, For nodes i The neighborhood set is used to assign the feature contribution of different monitoring nodes, which is used to enhance the feature expression of key nodes and improve the accuracy of subsequent corrosion status assessment.

[0039] In some embodiments, the cross-scale computational framework in this corrosion inhibitor prediction model is based on the coupling of microscopic action mechanisms and mesoscopic processes. The core action parameters of the corrosion inhibitor molecules are obtained by solving the electronic structure evolution equation. The simplified form of the electronic structure evolution equation is as follows: Among them, effective potential From electron density ρDetermined in conjunction with the external environment, It is used to characterize the interaction between corrosion inhibitor molecules and metal surfaces, and to guide the design of high-performance corrosion inhibitor molecular structures.

[0040] The graph network learning framework in this corrosion inhibitor prediction model characterizes the correlation between the molecular structure and performance of the corrosion inhibitor. It mines the mapping relationship between molecular structural features and effectiveness through a message passing mechanism. The prediction formula for molecular orbital energy levels is as follows: ;in, These are the edge features between atoms in a molecular diagram. This refers to the message vector passed between nodes. This is a global feature readout function used to establish a quantitative correlation between molecular structure and orbital energy levels, and to screen for corrosion inhibitor molecular structures with high adsorption performance.

[0041] The molecular inverse design framework in this corrosion inhibitor prediction model is based on generative adversarial and policy optimization. Candidate molecular structures are generated through adversarial training between the generator and the discriminator. The adversarial loss formula is as follows: .

[0042] Furthermore, a policy gradient optimization mechanism is introduced to strengthen the performance constraints of generated molecules. The reinforcement learning loss formula is as follows: ;in, R(x) This is a reward function for evaluating molecular performance, used to guide the generator to iterate towards high-performance molecular structures, and to quickly obtain corrosion inhibitor molecular solutions that meet industrial needs.

[0043] In one possible implementation, the corrosion inhibitor prediction model is optimized based on a learning-based dynamic optimization framework, a safety-constrained optimization framework, and an adaptive sample selection and active learning framework. The learning-based dynamic optimization framework embeds the optimization problem into a neural network model through a differential programming layer, supporting end-to-end backpropagation optimization. The safety-constrained optimization framework is based on stability theory and ensures the stable operation of the model by constructing a Lyapunov function. The adaptive sample selection and active learning framework is constructed based on the results of multi-precision evaluation and causal interpretation conclusions, and selects high-value sample data by maximizing the information gain function.

[0044] In some embodiments, the end-to-end optimization loss function of this learning-based dynamic optimization framework is as follows: .and, MPC The layer's output is used to calculate gradients via implicit function differentiation, supporting end-to-end backpropagation optimization. This enables deep integration of the optimization framework and the corrosion inhibitor prediction model, improving the real-time performance of dynamic decision-making.

[0045] Lyapunov functions in this safety-constrained optimization framework V(x)The stability constraint formula is as follows: ;in, α>0 For system attenuation rate, The upper bound of the allowable deviation is used to constrain the safety boundary of decision-making actions and ensure the stable operation of the corrosion inhibitor injection system.

[0046] The adaptive sample selection and decision interpretation framework in the active learning framework are based on causal reasoning. It quantifies the causal relationship between decision variables and outputs through intervention analysis. The formula for the causal intervention effect is as follows: .

[0047] Furthermore, by identifying instrumental variables in causal structural equations, the influence of confounding factors can be eliminated, revealing the intrinsic causal relationship between decision-making actions and the effectiveness of corrosion control, thereby enhancing the scientific rigor and interpretability of decision-making.

[0048] The multi-precision evaluation results in this adaptive sample selection and active learning framework are obtained based on a multi-precision evaluation framework that integrates the advantages of high-precision simulation analysis and field measurement data. The fusion formula for multi-precision evaluation is as follows: ;in, This is the precision scaling factor. This is a difference correction function for high- and low-precision data, used to achieve complementary fusion of data with different precision levels and improve the accuracy of system performance evaluation.

[0049] This adaptive sample selection and active learning framework is built upon the results of multi-precision evaluation and causal explanation conclusions. It selects the sample points most valuable for model iteration by maximizing the information gain function. The information gain function for sample selection is as follows: ;in, As an information entropy function, it balances the exploration and utilization needs of model training. The selected high-value samples will guide the targeted data collection of edge monitoring nodes, improve the efficiency of data collection and model training, and provide high-quality data support for subsequent strategy iterations.

[0050] In one possible implementation, the policy gradient formula for this multi-agent cooperative optimization framework is: ;in, For policy gradient, For the first Policy network of individual agents For the first The value network of individual agents For the agent to observe the state, This is a joint decision-making action.

[0051] In some embodiments, a meta-learning framework for continuous-time evolution can be established, which describes the dynamic update law of policy parameters in the target injection policy output by the multi-agent cooperative optimization framework through neural differential equations. The parameter evolution equation formula is as follows: ;in, For a parameterized neural network model, the backpropagation solution of the differential equation is achieved through the adjoint method.

[0052] In other embodiments, this multi-objective reward function is used for tensor decomposition, which can represent optimization objectives such as corrosion control effect, refueling cost, and supply efficiency in tensor form. The multi-objective reward function is as follows: ;in, For the target weight tensor, Let be the feature tensor formed by the reward vectors of each target. The tensor Hadamard product is used to quantify the comprehensive benefits of multi-objective collaborative optimization, thereby guiding the policy iteration direction of the agent.

[0053] In one possible implementation, S105 may include: encrypting the target tensor using homomorphic encryption technology and aggregating the encrypted target tensor; adjusting the aggregated target tensor using a rolling temporal optimization method to obtain the target annotation control sequence.

[0054] In some embodiments, the encryption formula is as follows: ;in, For public key generator, A random encryption factor. For encryption modulus, For node parameters; the server-side aggregation encryption gradient formula is as follows: .

[0055] In other embodiments, parameters in the encrypted state can be aggregated to ensure data privacy and security during the distributed optimization process. A multi-timescale dynamic optimization framework is established, and the injection strategy is adjusted in real time through a rolling time-domain optimization method. The objective function of the rolling time-domain optimization is as follows: ; ; ;in, Let be the system state vector. To add decision vectors, H To optimize the time domain length, this method is used to solve the optimal injection decision at different time scales, and to achieve dynamic and precise control of corrosion inhibitor injection.

[0056] In one possible implementation, the method may further include: determining a target inventory scheduling strategy based on a dynamic game model of the corrosion inhibitor supply network; adjusting the inventory and scheduling costs of each node in the corrosion inhibitor supply network based on the target inventory scheduling strategy; and optimizing the supply paths in the corrosion inhibitor supply network based on a combinatorial decision framework using a quantum optimization algorithm; the path optimization objective formula is: ;in, For Hamiltonians. Example, H The Hamiltonian for path optimization is used to find the optimal solution corresponding to the ground state energy through a quantum variational characteristic solver, which is used to achieve rapid optimization of the supply path and improve the efficiency and economy of corrosion inhibitor supply.

[0057] In some embodiments, this dynamic game optimization model can solve for the game equilibrium to obtain the optimal inventory scheduling strategy. The conditions for solving the Nash equilibrium are as follows: ;in, For total supply, Given the supply price function, an equilibrium solution set is obtained through a fixed-point iterative algorithm to balance the inventory and scheduling costs of each node in the supply network, thereby achieving the globally optimal configuration of corrosion inhibitor supply.

[0058] In one possible implementation, the method may further include: determining the Betti number in different dimensions using a persistent coherence method; determining whether a corrosion weak point exists in the industrial oil delivery pipeline based on the Betti number; and triggering a fault warning when a corrosion weak point is determined to exist and the number of corrosion weak points is greater than or equal to a preset quantity threshold. The preset quantity threshold can be set by the user, for example, 1, 2, 3, or 5, and is not limited here.

[0059] For example, a topological evaluation model of health status is established using the persistent cohomology method to determine the Betti number in different dimensions and quantify the evolutionary characteristics of the topological structure. The Betti number formula is as follows: ;in, for k Closed-chain group for k A dimensional boundary chain group is used to characterize the topological features of a healthy state. The evolution of the topological structure is visualized through a persistent graph, which can be used to achieve early warning of corrosion weak points.

[0060] In one possible implementation, the method may further include: determining the contribution of each global state vector to the target refueling strategy using an integral gradient method to assess the influence of each global state vector on the refueling decision; constructing a digital twin model based on the fluid motion control equation and the concentration distribution equation; simulating the transport effect and distribution characteristics of the corrosion inhibitor in the industrial oil pipeline based on the digital twin model; and adjusting the refueling strategy based on the transport effect and distribution characteristics.

[0061] In some embodiments, an interpretability analysis framework for annotation decisions is constructed using the integral gradient method to determine the contribution of each input feature to the decision outcome. The feature importance formula is as follows: Furthermore, counterfactual interpretation generation technology is introduced to construct virtual feature samples to analyze the changing patterns of decision-making results, thereby quantifying the impact of each monitoring feature on decision-making and improving the transparency and credibility of intelligent decision-making.

[0062] In other embodiments, the flow and mass transfer process of the medium is described by fluid motion control equations, which are as follows: ; Furthermore, the concentration distribution law of the corrosion inhibitor in the medium is described by a concentration distribution equation, the formula of which is as follows: ;in, D The diffusion coefficient of the corrosion inhibitor is... S This is the source term, used to simulate the transport and distribution characteristics of corrosion inhibitors in pipelines, and to guide the optimization and adjustment of the injection scheme.

[0063] In one possible implementation, the method may further include: after adding corrosion inhibitor to the industrial oil pipeline according to the target injection sequence, collecting multi-source heterogeneous data and the actual consumption of corrosion inhibitor at the next moment; determining a real-time reward value based on the multi-source heterogeneous data and the actual consumption of corrosion inhibitor through a multi-objective evaluation index; the reward value characterizing the comprehensive effectiveness of the injection decision at the next moment; constructing a complete empirical data set according to the target injection strategy at the current moment, the injection strategy at the next moment, and the reward value, and obtaining a sample dataset by combining high-value sample data selected by an adaptive sample selection and active learning framework; extracting data from the sample dataset according to a preset sampling strategy, and iteratively optimizing the multi-agent collaborative optimization framework. For example, the multi-source heterogeneous data may include the measured state of the industrial oil pipeline; the actual consumption of corrosion inhibitor may include the measured corrosion rate and the cost of corrosion inhibitor consumption.

[0064] In one possible implementation, the reward value is obtained using the following formula: ;in, The measured corrosion rate at time t+1 To preset the maximum allowable corrosion rate, The cost of corrosion inhibitor consumption at time t+1. Budget for cost per unit of time. The supply chain operating efficiency at time t+1 For ideal operating efficiency, The target weight coefficients are and satisfy the following conditions: .

[0065] For example, on-site refueling operations can be performed based on the optimized target refueling strategy, and operating condition data and corrosion rate data at time t+1 can be re-collected. The actual consumption of corrosion inhibitor can be simultaneously calculated, and a real-time reward value can be determined by combining multi-objective evaluation indicators. This reward value represents the comprehensive effectiveness of this refueling decision, providing a quantitative basis for model iteration. Furthermore, [the following text appears to be incomplete and requires further context: "and will..."] t A complete empirical dataset consisting of the observed state at time t, decision-making actions, the measured state at time t+1, and the real-time reward value. By combining high-value sample data selected through active learning, the data is fed back into the model's experience replay pool. Data is extracted according to a preset sampling strategy to iteratively train the policy network and value network in the multi-agent collaborative optimization framework, forming a closed-loop optimization circuit of perception, decision-making, execution, feedback, and re-optimization, thereby achieving continuous iterative optimization of the corrosion inhibitor application strategy.

[0066] The workflow of this application is described below with reference to an embodiment. This embodiment uses an oil pipeline corrosion inhibitor injection system as an application scenario. It is assumed that the oil pipeline system contains 5 key monitoring nodes, each equipped with a multimodal sensor array consisting of an electrochemical impedance spectroscopy sensor, a multispectral thermal imager, and an acoustic emission monitor, with a sampling frequency of 5Hz. Multi-source heterogeneous data from the oil pipeline system from January to June 2025 are used as the training set, and July data is used as the test set. The training set sample size is 1.08 × 10⁻⁶. 6 The test set has 1.8 × 10⁻⁶ samples. 5 strip.

[0067] (1) Construct the original data space and initialize the tensor parameters.

[0068] Define the original data space as ,in i To monitor the node index, t For sampling time index, k Indexed by feature dimension. t Taking the 3rd monitoring node at time 36000 as an example, its electrochemical impedance characteristic tensor The real part of the matrix takes values ​​in the range [450, 5200]Ω, and the multispectral thermal imaging feature tensor... The grayscale value range is [50, 220], and the acoustic emission feature tensor The average event count is 28 times per minute, and this original tensor is used as input for subsequent mixed data fusion models.

[0069] (2) Perform hybrid data fusion model calculation to achieve dimensionality reduction of high-dimensional heterogeneous data.

[0070] Assume a hybrid model of Tucker decomposition and tensor chain decomposition with an objective rank r = (15, 10, 8) and a core tensor... Factor matrix , , The tensor reconstruction error formula is obtained by iteratively solving the problem using the alternating least squares algorithm as follows: ;in, For the Frobenius norm, for n Modular tensor product. After 12 iterations, the reconstruction error decreased from the initial value of 215.6 to 9.8, satisfying the condition. E The convergence condition is <10, and the fused low-dimensional tensor features are used for anomaly detection and spatiotemporal feature extraction.

[0071] (3) Conduct anomaly detection calculations based on Riemannian manifold learning to eliminate invalid data.

[0072] Expand the fused tensor into a symmetric positive definite matrix. Define the manifold center matrix , Given the number of samples in the test set, the Fraser distance formula is as follows: ;in, Perform matrix trace operation; set distance threshold. A total of 3,247 abnormal samples were detected, with an abnormality rate of 1.8%. The dataset after removing abnormal data was used for training the spatiotemporal feature extraction model.

[0073] (4) Implement multi-scale spatiotemporal feature extraction calculations to explore the spatiotemporal correlation patterns of corrosion in industrial oil pipelines.

[0074] The structure contains 3 layers Conv3D-LSTM A feature extraction model using a 1-layer graph attention network is used, with an adjacency matrix A∈R for 5 monitoring nodes. 5×5 The multi-scale spatiotemporal feature fusion formula is as follows: .

[0075] Attention coefficient as follows: ;in, α =0.7 is Conv3D-LSTM Feature weights a For attention weight vectors, WThe characteristic transformation matrix yields the global state vector. This vector is used to characterize the real-time corrosion status of the pipeline system, providing input for subsequent molecular performance prediction and decision-making.

[0076] (5) Perform molecular performance prediction and reverse design of corrosion inhibitors to adapt to current working conditions.

[0077] Four imidazoline candidate molecules were screened from a corrosion inhibitor molecule library based on electronic structure evolution equations. Calculate the molecular orbital energy levels, including the effective potential. electron density The molecular adsorption properties are predicted using the message passing mechanism of molecular graph networks. The prediction formula for molecular orbital energy levels is as follows: .

[0078] In response to the current working conditions Cl - The concentration is 7800 ppm abnormal conditions, start GAN-RL Molecular reverse design engine to combat loss as follows: ; Reinforcement learning loss as follows: After 200 iterations, an optimized formula was generated, with a predicted corrosion inhibition efficiency of 95.6%.

[0079] (6) Conduct multi-agent collaborative decision-making computation and output target injection strategy.

[0080] Construct a MADDPG framework comprising four agents: formulation decision-making, main pump control, auxiliary pump control, and inventory scheduling, with agent policy gradients. as follows: Design a multi-objective reward tensor Among them, the weight tensor Corresponding to corrosion control, cost, and efficiency objectives, the Hadamard product achieves the allocation of objective weights; through iterative training, the joint actions... Q The value reached 22.3, an improvement of 38.7% compared to the baseline strategy.

[0081] (7) Implement distributed privacy protection and multi-timescale optimization to ensure decision security and real-time performance.

[0082] Homomorphic encryption technology is used to encrypt the gradient and parameters of the node model. as follows: ; Set public key generator random numbers Modulus The server-side aggregation encryption gradient formula is: Simultaneously, rolling time-domain optimization is initiated, and the prediction time domain is set.H =8, control time domain M =4, the objective function for rolling time-domain optimization is as follows: ; The target refueling control sequence is obtained by solving the problem, enabling dynamic tracking of the operating conditions.

[0083] (8) Conduct supply chain game optimization and system health assessment to ensure full-process collaboration.

[0084] Construct a dynamic game model of the supply network, using the Nash equilibrium condition. The optimal inventory level was determined, resulting in a 12.5% ​​reduction in raw material inventory costs. This was achieved by calculating the Betti number based on persistent coherence. ,get This indicates that there are two potential corrosion weak points in the pipeline system, enabling early warning.

[0085] (9) Perform interpretability analysis and digital twin simulation to improve the credibility of decision-making.

[0086] By integrating the gradient formula Calculate the feature contribution. Concentration, temperature, and flow rate are the key influencing factors, contributing a total of 78.3%.

[0087] A digital twin model was constructed based on the fluid motion control equation and the concentration distribution equation. The simulation showed that the diffusion time of the corrosion inhibitor in the industrial oil delivery pipeline was 12.3 minutes, with an error of less than 5% compared with the measured value.

[0088] (10) Execute closed-loop optimization loop to achieve continuous strategy iteration.

[0089] Based on the measured data at time t+1 after the betting strategy is executed, the real-time reward value is calculated using the following formula: ; set up ; Empirical data group The data is stored in the experience replay pool, and 256 experiences are sampled to iteratively update the model, completing the closed loop of perception, decision-making, execution, feedback, and re-optimization.

[0090] (11) Verify the system's operational effectiveness.

[0091] Table 1 shows the comparison of key performance indicators between this application and traditional control methods (PID control, single-agent RL system) after 31 consecutive days of testing in July. It can be seen that the performance indicators of this application are significantly better than those of traditional control methods.

[0092] Table 1: (12) Determine the cost-benefit ratio and quantify the economic value.

[0093] The total cost of ownership involved in this application includes equipment investment, reagent consumption, maintenance costs, and energy consumption. The determination method is as follows: ;in, For equipment investment, For drug consumption, For maintenance costs, Energy consumption. From this formula, the annual energy consumption of the method in this application can be obtained. The amount is 268,000 yuan. Due to the traditional PID control... The annual cost of implementing the control method described in this application is 352,000 yuan. This can save 84,000 yuan. The cost reduction rate of the control method in this application is determined by the following formula. : Furthermore, given that the additional investment in equipment is 150,000 yuan, the payback period is... : .

[0094] (13) Evaluate the robustness of the scheme and verify the operational stability under extreme conditions.

[0095] Four sensor failures were simulated during testing: two for the electrochemical sensor, two for the thermal imaging sensor, and three for communication interruptions. Missing data was compensated for through multi-source data fusion and digital twin simulation, and the system robustness was evaluated. as follows: ;in, For normal operating conditions, corrosion inhibition efficiency The corrosion inhibition efficiency under fault conditions is shown in the test results. The maximum decrease in corrosion inhibition efficiency under fault conditions is 3.2%, and the system still maintains stable operation, meeting the reliability requirements of industrial sites.

[0096] In summary, this application, through multimodal data fusion, cross-scale modeling, multi-agent collaborative decision-making, and closed-loop optimization, significantly reduces corrosion inhibitor consumption and maintenance costs while ensuring pipeline corrosion protection, improves response speed to sudden changes in operating conditions and system robustness, solves the technical bottlenecks of traditional filling systems, and has good industrial application prospects and promotion value.

[0097] Figure 2 This is a block diagram of a high-dimensional heterogeneous corrosion inhibitor synergistic optimization system 200 provided in an embodiment of this application. Figure 2 As shown, the system 200 may include: a data acquisition device 210, a controller 220, and an execution device 230; the data acquisition device 210 is connected to the controller 220, and the controller 220 is connected to the execution device 230.

[0098] The data acquisition device 210 is used to collect multi-source heterogeneous data of industrial oil transportation pipelines and send it to the controller 220; the controller 220 is used to execute the above-mentioned high-dimensional heterogeneous corrosion inhibitor synergistic optimization method and generate control commands; the execution device 230 is used to add corrosion inhibitors into industrial oil transportation pipelines based on the control commands.

[0099] For example, the data acquisition device can be a temperature sensor, pressure sensor, pH meter, flow meter, chloride ion sensor, etc.; the controller can be a PLC controller, computer, server cluster, etc.; the execution device can be a metering pump, regulating valve, etc., and there are no restrictions here.

[0100] The technical solution of this application transforms the optimization of industrial corrosion inhibitor filling into a cross-scale coupled problem. It constructs a hybrid optimization framework that integrates microscopic mechanism analysis, mesoscopic mass transfer process simulation and intelligent decision-making algorithm. This framework can efficiently integrate high-dimensional heterogeneous monitoring data and achieve collaborative decision-making in corrosion inhibitor formulation design, filling control and supply chain scheduling. Under the premise of ensuring that the corrosion rate meets the standard, it significantly reduces the consumption of corrosion inhibitors and improves the economy and reliability of industrial pipeline corrosion protection.

[0101] The various embodiments in this specification are described in a progressive manner. For the same or similar parts between the various embodiments, please refer to each other. Each embodiment focuses on describing the differences from other embodiments.

[0102] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit this application. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of this application.

Claims

1. A high-dimensional heterogeneous corrosion inhibitor synergistic optimization method, characterized in that, The method includes: The original data space is defined based on the multi-source heterogeneous data of the industrial oil transportation pipeline at the current moment, and the initial tensor parameters are obtained. The tensor parameters are then projected onto the low-dimensional space through a hybrid data fusion model to obtain the low-dimensional tensor parameters. The anomaly detection method based on Riemannian manifold learning removes invalid data from the low-dimensional tensor parameters and extracts features through a feature extraction model to obtain a global state vector. Based on the global state vector and combined with the pre-built corrosion inhibitor prediction model, a combination of corrosion inhibitor molecules is obtained; the corrosion inhibitor prediction model consists of a cross-scale computation framework, a graph network learning framework, and a molecular inverse design framework. Based on the corrosion inhibitor molecule combination, a target injection strategy is obtained using a pre-constructed multi-agent cooperative optimization framework; the multi-agent cooperative optimization framework includes a policy network and a value network. The multi-objective reward function based on tensor decomposition converts the target betting strategy into a target tensor. Distributed privacy protection and multi-timescale optimization are performed on the target tensor to obtain the target injection control sequence, and the corrosion inhibitor is injected into the industrial oil delivery pipeline according to the target injection sequence.

2. The method according to claim 1, characterized in that, The corrosion inhibitor prediction model is optimized based on a learning-based dynamic optimization framework, a safety-constrained optimization framework, and an adaptive sample selection and active learning framework. The learning-based dynamic optimization framework embeds the optimization problem into a neural network model through a differential programming layer, supporting end-to-end backpropagation optimization; the safety-constrained optimization framework is based on stability theory and ensures the stable operation of the model by constructing a Lyapunov function; the adaptive sample selection and active learning framework is built based on the results of multi-precision evaluation and causal explanation conclusions, and selects high-value sample data by maximizing the information gain function.

3. The method according to claim 1, characterized in that, The policy gradient formula for the multi-agent cooperative optimization framework is: ;in, For policy gradient, For the first Policy network of individual agents For the first The value network of individual agents For the agent to observe the state, This is a joint decision-making action.

4. The method according to claim 1, characterized in that, The distributed privacy protection and multi-timescale optimization performed on the target tensor to obtain the target annotation control sequence includes: The target tensor is encrypted using homomorphic encryption technology, and the encrypted target tensor is then aggregated. The target annotation control sequence is obtained by adjusting the aggregated target tensor using a rolling temporal optimization method.

5. The method according to claim 1, characterized in that, The method further includes: Based on the dynamic game model of the corrosion inhibitor supply network, a target inventory scheduling strategy is determined, and the inventory and scheduling costs of each node in the corrosion inhibitor supply network are adjusted based on the target inventory scheduling strategy. A combinatorial decision-making framework based on quantum optimization algorithms is used to optimize the supply path in the corrosion inhibitor supply network; The objective formula for path optimization is: ;in, For Hamiltonian.

6. The method according to claim 1, characterized in that, The method further includes: The Betti number in different dimensions is determined by a persistent coherence method, and the presence of corrosion weak points in the industrial oil delivery pipeline is determined based on the Betti number. If the presence of corrosion weak points is determined and the number of corrosion weak points is greater than or equal to a preset number threshold, a fault warning is triggered.

7. The method according to claim 1, characterized in that, The method further includes: The contribution of each global state vector to the target betting strategy is determined by the integral gradient method, so as to evaluate the degree of influence of each global state vector on the betting decision; A digital twin model is constructed based on the fluid motion control equation and the concentration distribution equation. The transport effect and distribution characteristics of the corrosion inhibitor in the industrial oil pipeline are simulated based on the digital twin model, and the injection strategy is adjusted based on the transport effect and distribution characteristics.

8. The method according to claim 2, characterized in that, The method further includes: After the corrosion inhibitor is injected into the industrial oil pipeline according to the target injection sequence, multi-source heterogeneous data and the actual amount of corrosion inhibitor consumed at the next moment are collected. Based on the aforementioned multi-source heterogeneous data and the actual consumption of corrosion inhibitor, a real-time reward value is determined through multi-objective evaluation indicators; the reward value represents the comprehensive effectiveness of the refueling decision at the next moment. Based on the target betting strategy at the current moment, the betting strategy at the next moment, and the reward value, a complete set of experience data is formed. Combined with high-value sample data selected by the adaptive sample selection and active learning framework, a sample dataset is obtained. Data is extracted from the sample dataset according to a preset sampling strategy, and the multi-agent collaborative optimization framework is iteratively optimized.

9. The method according to claim 1, characterized in that, The reward value is obtained using the following formula: ； in, The measured corrosion rate at time t+1 To preset the maximum allowable corrosion rate, The cost of corrosion inhibitor consumption at time t+1. Budget for cost per unit of time. The supply chain operating efficiency at time t+1 For ideal operating efficiency, The target weight coefficients are and satisfy the following conditions: .

10. A high-dimensional heterogeneous corrosion inhibitor synergistic optimization system, characterized in that, The system includes: a data acquisition device, a controller, and an execution device; the data acquisition device is connected to the controller, and the controller is connected to the execution device; The data acquisition device is used to collect multi-source heterogeneous data from industrial oil transportation pipelines and send it to the controller; The controller is configured to execute the method of any one of claims 1 to 9 and generate control instructions; The execution device is used to add corrosion inhibitor into the industrial oil delivery pipeline based on the control command.

Images