Method and system for determining maintenance timing of oil storage tank considering resource guarantee constraint

Abstract

The present application relates to the technical field of predictive maintenance, and specifically provides a method and system for determining maintenance time of oil storage tank considering guarantee resource constraint, wherein the method for determining maintenance time of oil storage tank considering guarantee resource constraint acquires multi-source state characteristic data of the storage tank, extracts multi-scale degradation features by using a time sequence feature extraction network, and predicts future state characteristic values through a nonlinear mapping network; the predicted values are input into a confidence rule base model to output confidence distribution of the storage tank at each performance state level; expected maintenance cost and premature maintenance resource loss are determined based on the distribution, a maintainable time interval is determined according to a maintenance start threshold and a safe state threshold, and the optimal maintenance start time is solved by taking minimizing comprehensive maintenance cost as an objective and introducing guarantee resource capacity constraint, so as to help realize comprehensive balance among performance safety, maintenance economy and guarantee resource utilization efficiency of the oil storage tank.

Description

Technical Field

[0001] This invention belongs to the field of predictive maintenance technology, specifically relating to a method and system for determining the timing of maintenance for oil storage tanks while taking into account resource constraints. Background Technology

[0002] Oil storage tanks are critical infrastructure in the storage of crude oil, refined oil products, and chemical media, and are widely used in oil fields, refining enterprises, storage and transportation areas, and port oil depots. During long-term service, oil storage tanks are subject to corrosion from the stored media, bottom plates, tank walls, top, floating roof seals, inlet and outlet pipes, and foundation structures. These factors can lead to degradation phenomena such as wall thinning, pitting corrosion, coating damage, foundation settlement, seal leakage, and localized deformation. If these degradation phenomena are not detected and maintained in a timely manner, they may result in leaks, fires and explosions, environmental pollution, and production stoppages.

[0003] Traditional oil storage tank maintenance methods mainly include periodic inspection and maintenance and post-failure repair: Although periodic inspection and maintenance is simple to implement, it is easy to overlook the degradation differences between different tanks and different parts, which may lead to tanks in good condition being shut down for maintenance prematurely, resulting in waste of resources such as tank cleaning, replacement, opening for inspection and production stoppage; Post-failure repair can make full use of the equipment's lifespan, but it has problems such as high leakage risk, high emergency response costs and significant safety and environmental risks.

[0004] In other words, existing technologies make it difficult to accurately determine the optimal maintenance time for oil storage tanks, making it difficult to balance efficiency, safety, and low cost when maintaining oil storage tanks. Summary of the Invention

[0005] This invention aims to at least address the technical problem of accurately determining the optimal maintenance timing for oil storage tanks in related technologies. Therefore, the objective of this invention is to provide a method, system, electronic device, and storage medium for determining the maintenance timing of oil storage tanks while considering resource constraints.

[0006] In a first aspect, the present invention proposes a method for determining the maintenance timing of oil storage tanks while considering resource constraints, comprising the following steps: acquiring a multi-source state feature data sequence of the oil storage tank; extracting multi-scale temporal degradation features from the multi-source state feature data sequence based on a temporal feature extraction network; establishing a mapping relationship between the multi-scale temporal degradation features and the predicted state feature values ​​at future times based on a nonlinear mapping network, thereby obtaining the predicted multi-source state feature values ​​at future times; inputting the predicted multi-source state feature values ​​at future times into a performance evaluation model based on a confidence rule base, and outputting the oil storage tank's future performance state levels at various preset performance state levels. Confidence distribution; based on the confidence distribution, the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time are determined, as well as the performance state utility value of the oil storage tank at each discrete prediction time. According to the preset maintenance start threshold and the preset safety state threshold, the maintainable time interval of each oil storage tank is determined, with the goal of minimizing the comprehensive maintenance cost. At the same time, the guaranteed resource capacity constraint is introduced into the comprehensive maintenance cost. The optimal maintenance start time is solved within the maintainable time interval. The comprehensive maintenance cost is at least composed of the expected maintenance cost and the premature maintenance resource loss weighted together.

[0007] According to an embodiment of the present invention, a method for determining the maintenance timing of oil storage tanks considering resource constraints first acquires a multi-source state feature data sequence of the oil storage tank. Multi-scale temporal degradation features are extracted from the multi-source state feature data sequence using a temporal feature extraction network. A mapping relationship is established between the multi-scale temporal degradation features and the predicted state feature values ​​at future times using a nonlinear mapping network, yielding the predicted multi-source state feature values ​​at future times. Then, the predicted multi-source state feature values ​​at future times are input into a performance evaluation model based on a confidence rule base, outputting the confidence distribution of the oil storage tank at each preset performance state level at future times. Finally, based on the confidence distribution, the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time are determined, as well as the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each discrete pre-set maintenance start time are determined. The system measures the performance state utility value at a given time and determines the maintainable time interval for each oil storage tank based on preset maintenance initiation and safety state thresholds. The goal is to minimize the overall maintenance cost, while incorporating resource capacity constraints into the overall maintenance cost. The optimal maintenance start time is then determined within the maintainable time interval. This collaborative architecture of time-series prediction and interpretable performance evaluation addresses the issues of insufficient interpretability in predictive maintenance and difficulties in modeling multi-source nonlinear coupling. Furthermore, the joint optimization of maintenance cost and premature maintenance resource depletion terms, along with the introduction of resource capacity constraints, solves the problem of neglecting resource depletion and resource capacity constraints in determining maintenance timing. This contributes to achieving a comprehensive balance between performance safety, maintenance economy, and resource utilization efficiency for oil storage tanks.

[0008] In addition, the method for determining the maintenance timing of oil storage tanks considering resource constraints according to embodiments of the present invention may also have the following additional technical features: Furthermore, the temporal feature extraction network is a temporal convolutional network; wherein, the temporal convolutional network is a one-dimensional causal convolutional structure, so that the prediction of future states depends only on current and historical information; the temporal convolutional network expands the receptive field through progressively increasing dilation coefficients to simultaneously extract short-term fluctuation features and long-term degradation trend features of performance states; the temporal convolutional network uses residual connections in each residual block, adding the input of the residual block to its output after convolution transformation to form a cross-layer addition structure.

[0009] Furthermore, the nonlinear mapping network is a Kolmogrove-Arnold network, and the final output of the Kolmogrove-Arnold network is the multi-source state feature prediction value at the future time. The Kolmogrove-Arnold network is composed of multiple stacked Kolmogrove-Arnold layers, and the edge function connecting the input and output nodes in each Kolmogrove-Arnold layer is composed of a base function and a B-spline function superimposed. The B-spline function achieves the nonlinear mapping from input to output through learnable spline coefficients. The base function and the B-spline function are each set with learnable scaling parameters, which are used to adjust the contribution ratio of the base function and the B-spline function.

[0010] Furthermore, in the performance evaluation model based on the confidence rule base, each confidence rule includes a reference level, rule weight, attribute weight, and output confidence distribution corresponding to each performance state level for each input state feature indicator. The step of inputting the predicted multi-source state features at future times into the performance evaluation model based on the confidence rule base and outputting the confidence distribution of the oil storage tank at each preset performance state level at future times includes: matching each performance degradation indicator in the predicted multi-source state features at future times with the reference level in each confidence rule, calculating the matching degree of each performance degradation indicator under each confidence rule; calculating the total matching degree of each confidence rule based on the attribute weight of each performance degradation indicator, and determining the activation weight of each confidence rule in combination with the rule weight; obtaining the preset output confidence distribution in the consequent of each confidence rule as a confidence vector, and weighting and fusing each confidence vector based on the evidence reasoning parsing algorithm and the activation weight to obtain the comprehensive confidence distribution of the oil storage tank.

[0011] Further, determining the maintainable time interval for each oil storage tank based on a preset maintenance initiation threshold and a preset safety state threshold includes: comparing the performance state utility value at each discrete prediction time with the maintenance initiation threshold, and setting the moment when the performance state utility value first falls below the maintenance initiation threshold as the earliest maintenance start time for the oil storage tank; comparing the performance state utility value at each discrete prediction time with the safety state threshold, and setting the moment when the performance state utility value first falls below the safety state threshold as the safety critical moment; shifting the safety critical moment forward by a preset oil storage tank maintenance duration to obtain the latest maintenance start time for the oil storage tank; and jointly defining the maintainable time interval based on the earliest maintenance start time and the latest maintenance start time.

[0012] Furthermore, the determination of the expected maintenance cost includes: pre-establishing a correspondence between the performance status level and maintenance level of the oil storage tank, wherein each performance status level corresponds to a maintenance level, and each maintenance level has a corresponding maintenance cost coefficient; using the confidence level of each performance status level in the confidence distribution output by the performance evaluation model as a weight, weighting the maintenance cost coefficient corresponding to each performance status level, and using the weighted result as the expected maintenance cost of the oil storage tank at the start time of the corresponding maintenance.

[0013] Furthermore, the determination of premature maintenance resource loss includes: using the earliest maintenance start time of the oil storage tank as a benchmark, constructing a function that decreases non-linearly as the candidate maintenance start time is postponed, and using the function value as the premature maintenance resource loss of the oil storage tank at the candidate maintenance start time; wherein, the function is used to quantify the waste of remaining life resources caused by premature shutdown for maintenance before the performance status of the oil storage tank has completely degraded; setting independent resource loss weight coefficients for different oil storage tanks, multiplying the resource loss weight coefficients by the function value, and using the product as the final value of premature maintenance resource loss for the corresponding oil storage tank, so as to characterize the difference in the degree of premature maintenance resource waste among the oil storage tanks due to differences in importance or maintenance urgency.

[0014] Furthermore, the introduction of resource capacity constraints in the comprehensive maintenance cost includes: determining the set of oil storage tanks that need to perform maintenance tasks, and limiting each oil storage tank in the set to perform a maintenance task at most once within a preset prediction time domain; setting an available job window capacity sequence, and for any discrete time, limiting the total number of all ongoing maintenance tasks covering that time to not exceed the available job window capacity corresponding to that time, wherein the available job window capacity sequence is used to characterize the maximum number of maintenance tasks that can be accommodated simultaneously at each discrete time.

[0015] Further, solving for the optimal maintenance start time within the maintainable time interval includes: under the constraint of guaranteed resource capacity, minimizing the weighted sum of the expected maintenance cost and the premature maintenance resource loss of all oil storage tanks requiring maintenance tasks as the optimization objective, and solving the problem using mixed integer linear programming under the following constraints: the maintenance start time of each oil storage tank is no earlier than its earliest maintenance start time and no later than its latest maintenance start time; each oil storage tank is scheduled for maintenance only once within a preset prediction time domain; the total number of work windows occupied by maintenance tasks at any discrete time does not exceed the available work window capacity at that time; the decision variable for the maintenance start time is binary; and the optimal maintenance start time and its corresponding maintenance plan for each oil storage tank are output after solving.

[0016] Secondly, based on the same inventive concept, this invention also proposes a system for determining the maintenance timing of oil storage tanks while considering resource constraints, comprising: a data acquisition module for acquiring multi-source state feature data sequences of oil storage tanks; a prediction module, which incorporates a time-series feature extraction network and a nonlinear mapping network, wherein the prediction module uses the time-series feature extraction network to extract multi-scale time-series degradation features from the multi-source state feature data sequences, and uses the nonlinear mapping network to establish a mapping relationship between the time-series degradation features and the predicted state feature values ​​at future times, and outputs the predicted multi-source state feature values ​​at future times; and a performance evaluation module, which incorporates a performance evaluation model based on a confidence rule base, wherein the performance evaluation module is used to evaluate the predicted multi-source state feature values ​​at future times. The input is given to the performance evaluation model, which outputs a confidence distribution of the oil storage tank at each preset performance state level at future times. The optimization decision module is used to determine the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time based on the confidence distribution, as well as the performance state utility value of the oil storage tank at each discrete prediction time. Based on the preset maintenance start threshold and the preset safety state threshold, the maintainable time interval of each oil storage tank is determined to minimize the comprehensive maintenance cost. At the same time, the guaranteed resource capacity constraint is introduced into the comprehensive maintenance cost. The optimal maintenance start time is solved within the maintainable time interval. The comprehensive maintenance cost is at least composed of the expected maintenance cost and the premature maintenance resource loss weighted average.

[0017] The system for determining the maintenance timing of oil storage tanks considering resource constraints provided in the second aspect has the same technical features as the method for determining the maintenance timing of oil storage tanks considering resource constraints provided in the first aspect. Therefore, it also has similar technical effects as the first aspect, which will not be described in detail here.

[0018] Thirdly, based on the same inventive concept, the present invention also proposes an electronic device, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store a computer program; and the processor, when executing the computer program stored in the memory, implements the steps of the method for determining the maintenance timing of oil storage tanks considering resource constraints as described in the above embodiments of the present invention.

[0019] The electronic equipment provided in the third aspect has the same technical features as the method for determining the maintenance timing of oil storage tanks considering resource constraints provided in the first aspect. Therefore, it also has similar technical effects as the first aspect, which will not be described in detail here.

[0020] Fourthly, based on the same inventive concept, the present invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the method for determining the maintenance timing of oil storage tanks considering resource constraints as described in the above embodiments of the present invention.

[0021] The computer-readable storage medium provided in the fourth aspect has the same technical features as the method for determining the maintenance timing of oil storage tanks considering resource constraints provided in the first aspect, and therefore also has similar technical effects as the first aspect, which will not be described in detail here. Attached Figure Description

[0022] Figure 1 This is a flowchart of a method for determining the maintenance timing of oil storage tanks considering resource constraints according to an embodiment of the present invention; Figure 2 This is a comparison chart of the predicted results of the settlement performance degradation index according to a specific embodiment of the present invention; Figure 3 This is a comparison chart of predicted results of deformation performance degradation index according to a specific embodiment of the present invention; Figure 4 This is a schematic diagram of performance utility output values ​​according to a specific embodiment of the present invention; Figure 5 This is a schematic diagram illustrating the changes in the original expected maintenance cost and premature maintenance resource depletion according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram of the solution results of the normalized component costs, the comprehensive maintenance objective function, and the optimal maintenance time according to a specific embodiment of the present invention. Figure 7 This is a structural block diagram of a system for determining the maintenance timing of oil storage tanks considering resource constraints, according to an embodiment of the present invention. Figure 8This is a structural block diagram of an electronic device according to an embodiment of the present invention.

[0023] Figure label: 100 - System for determining the maintenance timing of oil storage tanks considering resource constraints; 110 - Data acquisition module; 120 - Prediction module; 130 - Performance evaluation module; 140 - Optimization decision-making module. Detailed Implementation

[0024] The embodiments of the present invention are described in detail below, and the embodiments described with reference to the accompanying drawings are exemplary.

[0025] To address the problem of accurately determining the optimal maintenance timing for oil storage tanks in related technologies, this invention provides a method, system, electronic equipment, and storage medium for determining the maintenance timing of oil storage tanks while considering resource constraints. The following references... Figures 1-8 This invention describes a method, system, electronic device, and storage medium for determining the maintenance timing of oil storage tanks while considering resource constraints, according to embodiments of the present invention.

[0026] Figure 1 This is a flowchart of a method for determining the maintenance timing of oil storage tanks considering resource constraints, according to an embodiment of the present invention. Figure 1 As shown, a method for determining the maintenance timing of oil storage tanks considering resource constraints according to an embodiment of the present invention includes the following steps: Step S1: Obtain the multi-source state feature data sequence of the oil storage tank, extract the multi-scale temporal degradation features from the multi-source state feature data sequence based on the temporal feature extraction network, and establish the mapping relationship between the multi-scale temporal degradation features and the predicted values ​​of the state features at future times based on the nonlinear mapping network to obtain the predicted values ​​of the multi-source state features at future times.

[0027] In a specific embodiment, taking a single oil storage tank as the maintenance object, the maximum uneven settlement and the maximum radial deformation of the tank wall can be selected as indicators of tank performance degradation. Among them, the maximum uneven settlement is used to characterize the degree of settlement degradation of the tank foundation during long-term service; the maximum radial deformation of the tank wall is used to characterize structural deformations such as bulging, denting, or out-of-roundness of the tank wall. Considering that the tank degradation process is affected by operating conditions and environmental factors, the medium temperature can be further selected as an environmental operating condition parameter, which together constitute a multi-source state characteristic input.

[0028] Let the multi-source state characteristic sequence monitoring vector of oil storage tanks be defined. for: In the formula, The monitoring time for the characteristic sequence, The number of multi-source state characteristic sequences of oil storage tanks. express A multi-source state characteristic index in Real-time monitoring values, such as maximum uneven settlement, maximum radial deformation of the tank wall, medium temperature, and liquid level.

[0029] To predict future performance degradation trends using historical state information, a sliding window is used to construct the input samples, with a window length of [missing information]. The predicted step size is Let the input of the temporal feature extraction network (e.g., TCN (Temporal Convolutional Network)) - nonlinear mapping network (e.g., KAN (Kolmogorov-Arnold Network)) model (hereinafter referred to as the TCN-KAN model for ease of description) be... The output is : In the formula, This represents the historical monitoring times of the feature sequence. To represent the future prediction times of the feature sequence, the TCN-KAN model needs to learn the following mapping relationship: In the formula, for Time length The true historical observation sequence, For the future Step-by-step prediction sequence, for Predicted value at time, This represents a nonlinear prediction function composed of TCN-KAN.

[0030] Before training the TCN-KAN model, it is usually necessary to normalize the raw monitoring data to reduce the impact of differences in the units of different indicators on the training process. In the formula, For monitoring indicators The standardized value, and These represent the mean and standard deviation of each indicator, respectively; in this specific embodiment, the input indicators of the TCN-KAN model have all been normalized.

[0031] Specifically, this embodiment of the invention first acquires a multi-source state feature data sequence of oil storage tanks. Based on a time-series feature extraction network, it extracts multi-scale time-series degradation features from the multi-source state feature data sequence. Then, based on a nonlinear mapping network, it establishes a mapping relationship between the multi-scale time-series degradation features and the predicted state feature values ​​for future times, obtaining the predicted multi-source state feature values ​​for future times. Using this, the time-series feature extraction network performs time-series modeling on historical data sequences, extracting multi-scale degradation features such as short-term fluctuations and long-term trends. The nonlinear mapping network maps the time-series features to predicted state feature values ​​for future times, achieving effective fitting of the complex correlations between multi-source state features. A sliding window is used to construct input samples, enabling the model to learn the mapping pattern from history to the future using limited historical monitoring data. Normalization eliminates the impact of differences in the dimensions of different indicators on the training process. Ultimately, this achieves high-precision prediction of multi-source state features for future times, providing an accurate and reliable data foundation for subsequent performance status evaluation.

[0032] Step S2: Input the predicted values ​​of multi-source state features for future time moments into the performance evaluation model based on the confidence rule base, and output the confidence distribution of the oil storage tank at each preset performance state level in the future time moment.

[0033] In a specific embodiment, the TCN-KAN model takes as input a multi-source state feature sequence of the oil storage tank and outputs predicted values ​​of multi-source state features for future time periods. It uses the performance degradation index from the predicted state feature sequence as input to a performance evaluation model based on a BRB (Belief Rule Base) (hereinafter referred to as the BRB model for ease of description). Let the... The predicted performance degradation index of oil storage tanks at any given time is : In the formula, The quantity of indicators for the performance degradation of oil storage tanks. For the first Individual performance degradation indicators The predicted value at any given time.

[0034] The first in BRB Rules Represented as: In the formula, For the number of rules, for time Predicted values ​​for each performance degradation index. For reference, the number of evaluation levels, For the first The reference evaluation level corresponding to each input indicator in the rule is as follows: For logical AND operation, The performance evaluation results output by the BRB model. A set of performance status reference levels, The total number of performance evaluation reference levels, This is the corresponding output confidence vector. For the first Evaluation level under the rules The corresponding confidence level, for The attribute weights corresponding to each performance degradation metric For the first The rule weight of each rule.

[0035] First, BRB rules are activated, converting the input metric information into a matching degree: In the formula, , Indicates the first The first indicator is relative to the first Matching degree of each reference level It is the first The first indicator The specific numerical values ​​corresponding to each reference level. It is the first The first indicator in the The specific numerical value corresponding to each reference level is used to calculate the matching degree of the input indicator information using this formula. Then, the matching degree of all input indicator information with the first reference level is further calculated. Total matching degree of the rules : In the formula, For the first The first indicator is relative to the first Matching degree of each reference level For the first The original attribute weights of each indicator For the first The relative attribute weights of each indicator after normalization show that different input information has different effects on each rule, and this difference can be reflected by the activation weights of the rules: In the formula, , For the total number of rules, For the first The activation weights after rule normalization For the first The rule weight of each BRB rule. For all input metrics, the first The total matching degree of the rules.

[0036] After rule fusion and activation, the confidence vectors generated by all rules are fused using the analytical ER (Evidential Reasoning) algorithm to obtain the comprehensive assessment confidence vector of the oil storage tank. The ER analytical algorithm expression is as follows: In the formula, For all rules on the first Reference level for each assessment result confidence level As the normalization factor, For the first Rule number 1 Reference level for each assessment result confidence level For the first Rule number 1 Reference level for each assessment result confidence level Given the total number of rules, after the above ER analysis and fusion process, the comprehensive confidence vector can be obtained. : , This represents the total number of performance evaluation reference levels.

[0037] By matching the output confidence level with the expected utility value, a performance status evaluation value can be obtained. : In the formula, For assessment level The pre-set expected utility value.

[0038] Specifically, in this embodiment of the invention, the predicted values ​​of multi-source state features at future time points are then input into a performance evaluation model based on a confidence rule base, outputting the confidence distribution of the oil storage tank at each preset performance state level at future time points. This allows the performance evaluation process to have clear physical meaning and a traceable reasoning path through the reference level, rule weight, attribute weight, and output confidence distribution contained in each rule in the confidence rule base. By matching each performance degradation index with the reference level to calculate the matching degree, the numerical input is transformed into a semantic matching degree. Activation weights are calculated through attribute weights and rule weights to achieve a reasonable distinction in the importance of different expert rules and different input indicators. A weighted fusion of each confidence vector is performed using an evidence reasoning parsing algorithm to ensure the consistency and rationality of multi-source information fusion. This ensures that the final output of the comprehensive confidence distribution of the oil storage tank can provide a highly interpretable and information-rich health status input for subsequent maintenance decisions.

[0039] Step S3: Based on the confidence distribution, determine the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time, and determine the performance state utility value of the oil storage tank at each discrete prediction time. According to the preset maintenance start threshold and the preset safety state threshold, determine the maintainable time interval of each oil storage tank. With the goal of minimizing the comprehensive maintenance cost, the guarantee resource capacity constraint is introduced into the comprehensive maintenance cost. Solve for the optimal maintenance start time within the maintainable time interval. The comprehensive maintenance cost is at least composed of the expected maintenance cost and the premature maintenance resource loss weighted together.

[0040] In a specific embodiment, in order to generate a maintenance time window, the first... The earliest maintenance time for an oil storage tank Critical moment of safety They are respectively: In the formula, Indicates oil storage tanks, Indicates the first The oil storage tank was in the first The performance state prediction utility value at time t, where, Indicates the discrete prediction time. Indicates the length of the prediction time domain. The maintenance start threshold represents the health status boundary of an oil storage tank that is considered for preventive maintenance. When the health index drops below this threshold, it indicates that the oil storage tank has undergone significant degradation and maintenance should be considered. The safety threshold represents the minimum health condition boundary that allows an oil storage tank to continue operating. When the health index is close to or below this threshold, continuing operation will pose a high risk, so maintenance should be completed before then.

[0041] If the first The maintenance duration for each oil storage tank is as follows: Therefore, in order to ensure that the maintenance completion time is no later than the safety critical moment, the latest maintenance start time is... With the Maintenance time window for each oil storage tank They are respectively: This time window indicates that maintenance should not be performed earlier than [date missing]. Otherwise, it will result in a significant waste of remaining lifespan; maintenance should not be later than [date missing]. Otherwise, the oil storage tank will not meet the minimum safety threshold before maintenance is completed.

[0042] To describe the start time of an oil storage tank maintenance task, binary decision variables are introduced. : In the formula, , For the number of oil storage tanks, if , indicating the first Oil storage tanks from time Start maintenance and continue. One time unit; during maintenance, this task will occupy one job window.

[0043] The BRB model can not only output the performance utility value of oil storage tanks, but also give the final confidence level of the oil storage tank belonging to each performance status reference level. Different performance status reference levels correspond to different maintenance levels. For example, when the performance status of an oil storage tank is good, only inspection or light maintenance is required, while when the performance status is poor, deep repair may be required.

[0044] Based on this, a correspondence between performance status reference levels and maintenance levels is established: ,in, Indicates the performance status level The corresponding maintenance levels are set as follows: Several oil storage tanks Expected maintenance cost when maintenance begins for: In the formula, For the first Confidence level of individual oil storage tanks at various performance condition levels For the first The oil storage tanks are at the maintenance level The maintenance costs are reduced.

[0045] Within the maintenance time window, earlier maintenance is not always better. While immediate maintenance at the moment the maintenance initiation threshold is reached ensures safety, it may result in significant waste of remaining performance. Therefore, the resource loss caused by premature tank shutdown for maintenance can be considered a crucial component of the optimization objective. Let's define a nonlinear function representing the resource loss that gradually decreases as the maintenance start time is delayed. for: In the formula, For the first The resource depletion weight of each oil storage tank is used to characterize the degree of resource waste caused by premature maintenance of the oil storage tank.

[0046] To minimize resource depletion and reduce maintenance costs caused by premature maintenance, and considering the varying importance, repair costs, and resource consumption levels of different oil storage tanks, a MILP (Mixed-Integer Linear Programming) weighted objective function is established, incorporating the importance weights of oil storage tanks: In the formula, Indicates the first The weighting coefficient of resource loss for each oil storage tank. Indicates the first The weighting coefficient for the maintenance cost item of each oil storage tank, and When more attention is paid to the economic efficiency of maintenance, the amount can be appropriately increased. When greater emphasis is placed on utilizing remaining lifespan and ensuring resource conservation, the amount can be appropriately increased. .

[0047] Considering that maintenance costs and resource depletion may have different dimensions, the two types of costs can be normalized in practical applications.

[0048] When maintaining oil storage tanks, numerous resource and maintenance constraints must be considered; therefore, constraints need to be established for the objective function. Establish a unique maintenance constraint: For oil storage tanks that have already entered the maintenance demand set, maintenance should be scheduled once within their maintenance time window. In the formula, This represents the set of oil storage tanks that need maintenance within the prediction time domain; this constraint ensures that each oil storage tank requiring maintenance will only have its maintenance task scheduled once within the prediction time domain.

[0049] Establish maintenance time window constraints: Maintenance tasks can only begin within the corresponding maintenance time window. This constraint ensures that the maintenance start time is neither earlier than the time corresponding to the maintenance start threshold, nor later than the latest maintenance start time allowed by the safety state threshold.

[0050] Establish maintenance and repair resource capacity constraints: Set time The number of available job windows is If other tasks are already occupying the window at some point, then This can represent the number of remaining available windows after deducting those already occupied.

[0051] Due to the The maintenance duration for each oil storage tank is If it is at time Maintenance begins in the interval. Each task occupies one work window, therefore, at any given time... The total number of maintenance tasks currently in progress cannot exceed the number of available job windows. This constraint is the core of the resource constraint guarantee in this invention. It ensures that maintenance tasks can only be scheduled within the time period allowed by the job window capacity. When a job window becomes unavailable at a certain moment, the following can be set: At this point, the model will not schedule any maintenance tasks that cover that moment.

[0052] Establish decision variable constraints: Maintain the initial variable as a binary variable. This constraint ensures that the maintenance arrangement has a clear discrete decision meaning.

[0053] In summary, the model for determining the optimal maintenance timing, considering both resource constraints and maintenance costs, can be expressed as: In the formula, the objective function is composed of maintenance cost and resource depletion terms. The maintenance cost term is obtained by weighting the final confidence level of the oil storage tank at each reference performance state level with the maintenance cost of the corresponding maintenance level. The resource depletion term is used to characterize the waste of remaining life caused by premature maintenance.

[0054] After solving the MILP model, if the first... Each oil storage tank at any time Optimal maintenance decision variables The optimal maintenance plan set for this oil storage tank for: In the formula, For the first Optimal maintenance start time for each oil storage tank For the first Optimal maintenance start time for each oil storage tank For the duration of maintenance of this oil storage tank, The expected maintenance cost at this maintenance time. This refers to the premature maintenance resource consumption corresponding to this maintenance time.

[0055] The maintenance plan simultaneously meets the following requirements: the maintenance timing is within the maintenance time window generated by the maintenance initiation threshold and the safety status threshold; the number of operation windows occupied at any given time does not exceed the available window capacity; maintenance tasks within the same oil storage tank meet the necessary mutual exclusion requirements; and, under the premise of meeting the safety status requirements, the maintenance cost calculated based on the final confidence level is kept as low as possible, while the resource loss caused by premature maintenance is kept as low as possible. Thus, the model can achieve a comprehensive balance between performance safety, maintenance economy, and ensuring resource utilization efficiency.

[0056] Specifically, the embodiments of the present invention, based on the confidence distribution, determine the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time, and determine the performance state utility value of the oil storage tank at each discrete prediction time. Based on a preset maintenance initiation threshold and a preset safety state threshold, the maintainable time interval for each oil storage tank is determined, with the goal of minimizing the overall maintenance cost. Simultaneously, a guaranteed resource capacity constraint is introduced into the overall maintenance cost. The optimal maintenance start time is then solved within the maintainable time interval, wherein the overall maintenance cost is at least weighted by the expected maintenance cost and the premature maintenance resource loss. The system is structured as follows: The expected maintenance cost is obtained by weighting each maintenance cost coefficient using the confidence level of each performance state level in the confidence distribution as a weight; the resource loss due to premature maintenance is quantified by constructing a function that decreases non-linearly with the postponement of the candidate maintenance start time, and independent resource loss weight coefficients are set for different storage tanks to characterize their individual differences; by introducing a guaranteed resource capacity constraint, maintenance tasks can only be scheduled within the time period allowed by the work window capacity, making the output maintenance plan practically executable; and a comprehensive balance is achieved between performance safety, maintenance economy, and guaranteed resource utilization efficiency through a weighted objective function.

[0057] Therefore, according to the embodiment of the present invention, the method for determining the maintenance timing of oil storage tanks considering resource constraints firstly acquires a multi-source state feature data sequence of the oil storage tank, extracts multi-scale temporal degradation features from the multi-source state feature data sequence based on a temporal feature extraction network, and establishes a mapping relationship between the multi-scale temporal degradation features and the predicted state feature values ​​at future times based on a nonlinear mapping network to obtain the predicted multi-source state feature values ​​at future times; then, the predicted multi-source state feature values ​​at future times are input into a performance evaluation model based on a confidence rule base, outputting the confidence distribution of the oil storage tank at each preset performance state level at future times; finally, based on the confidence distribution, the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time are determined, and the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each discrete time are determined. The system predicts the performance state utility value at a given moment and determines the maintainable time interval for each oil storage tank based on preset maintenance initiation and safety state thresholds. The goal is to minimize the overall maintenance cost, while incorporating resource capacity constraints into the overall maintenance cost. The optimal maintenance start time is then determined within the maintainable time interval. This collaborative architecture of time-series prediction and interpretable performance evaluation addresses the issues of insufficient interpretability in predictive maintenance and the difficulty in modeling multi-source nonlinear coupling. Furthermore, the joint optimization of maintenance cost and premature maintenance resource depletion terms, along with the introduction of resource capacity constraints, solves the problem of neglecting resource depletion and resource capacity constraints in determining maintenance timing. This contributes to achieving a comprehensive balance between performance safety, maintenance economy, and resource utilization efficiency for oil storage tanks.

[0058] In one embodiment of the present invention, the temporal feature extraction network is a temporal convolutional network; wherein, the temporal convolutional network is a one-dimensional causal convolutional structure, so that the prediction of future states depends only on the information of the current and historical moments; the temporal convolutional network expands the receptive field by increasing the dilation coefficient layer by layer, so as to simultaneously extract the short-term fluctuation features and long-term degradation trend features of the performance state; the temporal convolutional network uses residual connections in each residual block, and adds the input of the residual block to its output after convolution transformation to form a cross-layer addition structure.

[0059] In a specific implementation, TCN employs a one-dimensional causal convolution structure, enabling the model to use only current and historical information when predicting future states, thus avoiding the leakage of future information. To capture the long-term degradation trend of oil storage tank performance, TCN introduces dilated convolution, specifically dilated causal convolution. Represented as: In the formula, For convolution kernel parameters, The feature representation after convolution. The coefficient of thermal expansion is 1 / 3. To find the upper bound of the summation of the expansion, when The time-series model focuses on information from adjacent time points, when The time-based model can extract historical features across longer time intervals.

[0060] For the Layer TCN, its expansion coefficient Recorded as It is usually set as follows: .

[0061] As the number of network layers increases, the receptive field of the model expands, enabling it to simultaneously capture both short-term fluctuations and long-term degradation trends in the performance of oil storage tanks.

[0062] To alleviate the vanishing gradient problem during deep network training, TCN typically employs residual connections, where the first... The input of each residual block is Convolution transformation is The output will be: In the formula, It typically consists of structures such as dilated causal convolution, nonlinear activation functions, and Dropout.

[0063] After multi-layer TCN processing, the time-series characteristic representation of the historical performance sequence of oil storage tanks can be obtained: In the formula, It includes the degradation trend, fluctuation characteristics and time-series information of the performance status of oil storage tanks within a historical time window.

[0064] Specifically, in this embodiment of the invention, the temporal feature extraction network is a one-dimensional causal convolutional structure, which effectively avoids prediction distortion caused by the leakage of future information; by expanding the receptive field through progressively increasing dilation coefficients, it can simultaneously extract short-term fluctuation features and long-term degradation trend features of performance status; by using residual connections in each residual block, it effectively alleviates the gradient vanishing problem in deep network training, allowing the network to stack more layers to obtain a larger receptive field without sacrificing training stability.

[0065] In one embodiment of the present invention, the nonlinear mapping network is a Kolmogrove-Arnold network, and the final output of the Kolmogrove-Arnold network is the predicted value of multi-source state features at future time steps. The Kolmogrove-Arnold network is composed of multiple stacked Kolmogrove-Arnold layers, and the edge function connecting the input and output nodes in each Kolmogrove-Arnold layer is composed of a base function and a B-spline function superimposed. The B-spline function achieves the nonlinear mapping from input to output through learnable spline coefficients. The base function and the B-spline function are each set with learnable scaling parameters, which are used to adjust the contribution ratio of the base function and the B-spline function.

[0066] In a specific embodiment, for a layer KAN, the first Each output node is represented as: In the formula, Indicates connection to input node With output node Learnable univariate functions.

[0067] For a multi-layered KAN, its recursive form is: In the formula, Indicates the first Layer input, Indicates the first The final output of the layered KAN mapping is: .

[0068] The edge functions in KAN are composed of both the fundamental functions and B-spline functions: In the formula, As a basic function, its common form is: , For B-spline basis functions, For B-spline coefficients, and For learnable scaling parameters, denoted as the number of spline basis functions.

[0069] Specifically, in this embodiment of the invention, the nonlinear mapping network is composed of multiple Kolmogrove-Arnold layers stacked together. Through the progressive abstraction and transformation of feature representations via the multi-layer stacked structure, it can fit complex multivariate nonlinear function relationships with fewer parameters. The edge functions connecting the input and output nodes in each layer are composed of a base function and a B-spline function superimposed. The base function ensures the global stability and smoothness of the mapping, while the B-spline function achieves flexible local nonlinear mapping from input to output through learnable spline coefficients. The base function and the B-spline function are each set with learnable scaling parameters to adjust the contribution ratio of the base function and the B-spline function, so that the network can adaptively balance the weights of the global linear part and the local nonlinear part during training, thereby achieving nonlinear expression capabilities and fitting accuracy far exceeding those of traditional multilayer perceptrons with the same number of parameters on a logarithmic scale.

[0070] In one embodiment of the present invention, the method for determining the maintenance timing of oil storage tanks considering resource constraints further includes: training a model composed of a time-series feature extraction network and a nonlinear mapping network.

[0071] In a specific embodiment, the TCN-KAN model is trained, and the training set is set as follows: Given the sample size, the basic prediction loss function uses the mean squared error. Represented as: In the formula, the first The true future performance degradation index for each sample is: , No. The model prediction value for each sample is Considering that different performance degradation indicators may have varying degrees of impact on the health status and maintenance requirements of oil storage tanks, the following is introduced: For the first Weighting coefficients for each performance degradation index.

[0072] The TCN-KAN model employs an end-to-end joint training approach. During training, the convolution kernel parameters and bias parameters of the TCN, as well as the B-spline coefficients and scaling parameters of the KAN, are simultaneously optimized. The model parameter optimization objective is... for: In the formula, This represents all learnable parameters in the TCN-KAN model, including the convolution kernel parameters of TCN and the B-spline coefficients and scaling parameters of KAN.

[0073] In actual training, automatic differentiation and backpropagation can be implemented based on the PyTorch framework, and the Adam (Adaptive Moment Estimation) optimizer is used to iteratively update the model parameters. The first-order moment estimates and second-order moment estimates of the Adam optimizer are as follows: In the formula, Indicates the current gradient. This represents the first moment estimate of the gradient. This represents the second-order moment estimate of the squared gradient. and The parameter update process is as follows: (This refers to the attenuation coefficient.) In the formula, For learning rate, To prevent extremely small constants with a denominator of zero, and These are the first-order moment estimates and second-order moment estimates after bias correction, respectively. The bias correction process is as follows: In the formula, and They are respectively and of Power of 1.

[0074] Through continuous iterative training, the TCN-KAN model can gradually learn the mapping relationship between the historical performance state and the future performance state of oil storage tanks, thereby obtaining more accurate performance prediction results.

[0075] Specifically, this embodiment of the invention adopts an end-to-end joint training method for the TCN-KAN model, so that the convolution kernel parameters and bias parameters of TCN and the B-spline coefficients and scaling parameters of KAN participate in the optimization simultaneously, avoiding the error accumulation and local optima problems that may be caused by staged training; a weighted mean square error is introduced as a loss function, allowing different weight coefficients to be assigned to different performance degradation indicators; and the Adam optimizer is used for parameter updates, which accelerates model convergence while ensuring training stability.

[0076] In one embodiment of the present invention, the method for determining the maintenance timing of oil storage tanks considering resource constraints further includes: optimizing the parameters of a performance evaluation model based on a confidence rule base.

[0077] In a specific embodiment, the attribute weights of each input performance degradation index in the BRB model are... Weights of each confidence rule Each confidence level The model parameters are all derived from expert experience and are easily affected by disturbances and environmental changes, leading to a decrease in adaptability. Therefore, it is necessary to optimize these parameters based on actual sample data. The optimized model is represented by the mean squared error. In the formula, S This indicates the amount of data used in the performance status assessment. and These represent the actual output utility and the performance evaluation utility value, respectively. This represents the mean square error.

[0078] Specifically, in this embodiment of the invention, the attribute weights, confidence rule weights, and confidence levels of each input performance degradation index in the BRB model are optimized based on actual sample data, with the goal of minimizing the mean square error. This enables the BRB model to adaptively adjust model parameters to adapt to changes in performance degradation characteristics under different storage tanks and different operating conditions while maintaining the interpretability of expert knowledge.

[0079] In one embodiment of the present invention, in the performance evaluation model based on a confidence rule base, each confidence rule includes a reference level, rule weight, attribute weight, and output confidence distribution corresponding to each performance state level for each input state feature indicator. The model inputs multi-source state feature predictions for future times into the performance evaluation model based on the confidence rule base, outputting the confidence distribution of the oil storage tank at each preset performance state level in the future. This includes: matching each performance degradation indicator in the multi-source state feature predictions for future times with the reference level in each confidence rule, calculating the matching degree of each performance degradation indicator under each confidence rule; calculating the total matching degree of each confidence rule based on the attribute weight of each performance degradation indicator, and determining the activation weight of each confidence rule based on the rule weight; obtaining the preset output confidence distribution in the consequent of each confidence rule as a confidence vector, and weighting and fusing each confidence vector based on an evidence reasoning parsing algorithm and the activation weight to obtain the comprehensive confidence distribution of the oil storage tank.

[0080] In specific embodiments, the formulas for matching degree, total matching degree, activation weight, and ER parsing fusion algorithm can be found in the relevant descriptions and formulas in step S2 above, and will not be repeated here.

[0081] Specifically, in this embodiment of the invention, each confidence rule includes a reference level, rule weight, attribute weight, and output confidence distribution corresponding to each performance state level for each input state feature indicator. This makes the knowledge representation semantically clear, structurally sound, easy to inject expert knowledge, and easy to maintain later. The numerical input is transformed into semantic matching degree through a Gaussian similarity function. The activation weight is determined through a dual weighting mechanism of attribute weight and rule weight, which considers both the matching degree between the current input and the rule and the prior judgment of the expert on the importance of the rule. The confidence vectors are weighted and fused based on the evidence reasoning parsing algorithm to ensure the completeness, consistency, and rationality of multi-source information fusion.

[0082] In one embodiment of the present invention, the maintainable time interval for each oil storage tank is determined based on a preset maintenance initiation threshold and a preset safety state threshold, including: comparing the performance state utility value at each discrete prediction time with the maintenance initiation threshold, and setting the moment when the performance state utility value first falls below the maintenance initiation threshold as the earliest maintenance start time for the oil storage tank; comparing the performance state utility value at each discrete prediction time with the safety state threshold, and setting the moment when the performance state utility value first falls below the safety state threshold as the safety critical moment; shifting the safety critical moment forward by a preset oil storage tank maintenance duration to obtain the latest maintenance start time for the oil storage tank; and jointly defining the maintainable time interval based on the earliest maintenance start time and the latest maintenance start time.

[0083] In specific embodiments, the methods for determining the earliest maintenance time, the safety critical time, and the latest maintenance start time can be found in the relevant descriptions and formulas in step S3 above, and will not be repeated here.

[0084] Specifically, the embodiments of the present invention avoid the waste of resources caused by premature tank shutdown for maintenance when the tank is still in good condition by using the earliest maintenance start time; avoid the risk of major safety accidents caused by late maintenance by using the safety critical time; ensure that the maintenance task can be completed before the tank reaches the safety critical state by shifting the safety critical time forward by a preset maintenance duration; and define the maintainable time interval by the earliest and latest maintenance start times, providing a reasonable feasible domain for subsequent optimization search.

[0085] In one embodiment of the present invention, the determination of the expected maintenance cost includes: pre-establishing a correspondence between the performance status level and the maintenance level of the oil storage tank, wherein each performance status level corresponds to a maintenance level and each maintenance level has a corresponding maintenance cost coefficient; using the confidence level of each performance status level in the confidence distribution output by the performance evaluation model as a weight, weighting the maintenance cost coefficient corresponding to each performance status level, and using the weighted result as the expected maintenance cost of the oil storage tank at the start time of the corresponding maintenance.

[0086] In a specific embodiment, the formula for the expected maintenance cost can be found in the relevant formula in step S3 above, and will not be repeated here.

[0087] Specifically, the embodiments of the present invention pre-establish a correspondence between performance status levels and maintenance levels, mapping continuous performance degradation levels to discrete maintenance strategy selections, which conforms to the actual decision-making logic in engineering practice of determining the repair depth based on the degree of tank damage; by using the confidence level of each performance status level in the confidence distribution as a weight to weight each maintenance cost coefficient, and taking into account the uncertainty that the tank may be in multiple performance status levels, a risk-neutral estimate of maintenance costs is given in a probabilistic weighted manner.

[0088] In one embodiment of the present invention, the determination of premature maintenance resource loss includes: constructing a function that decreases nonlinearly as the candidate maintenance start time is delayed, based on the earliest maintenance start time of the oil storage tank; and using the function value as the premature maintenance resource loss of the oil storage tank at the candidate maintenance start time. The function is used to quantify the waste of remaining life resources caused by premature shutdown for maintenance before the performance of the oil storage tank has fully degraded. Independent resource loss weighting coefficients are set for different oil storage tanks, and the resource loss weighting coefficients are multiplied by the function value. The resulting product is used as the final value of the premature maintenance resource loss of the corresponding oil storage tank, thus characterizing the difference in the degree of premature maintenance resource waste among the oil storage tanks due to differences in importance or maintenance urgency.

[0089] In a specific embodiment, the formula for the resource loss function can be found in the relevant formula in step S3 above, and will not be repeated here.

[0090] Specifically, the embodiments of the present invention construct a non-linear decreasing function starting from the earliest maintenance time, which describes the engineering law that the closer the maintenance time is to the safety critical moment, the higher the remaining life utilization rate and the less resource waste. By setting independent resource loss weight coefficients for different oil storage tanks, the optimization model can make differentiated maintenance decisions based on the engineering importance and maintenance urgency of the storage tank.

[0091] In one embodiment of the present invention, a resource capacity constraint is introduced into the comprehensive maintenance cost, including: determining a set of oil storage tanks that need to perform maintenance tasks, and limiting each oil storage tank in the set to perform a maintenance task at most once within a preset prediction time domain; setting an available job window capacity sequence, and for any discrete time, limiting the total number of all maintenance tasks currently being performed that time to not exceed the available job window capacity corresponding to that time, wherein the available job window capacity sequence is used to characterize the maximum number of maintenance tasks that can be accommodated simultaneously at each discrete time.

[0092] In a specific embodiment, the formulas for maintaining uniqueness constraints and maintenance resource capacity constraints can be found in the relevant formulas in step S3 above, and will not be repeated here.

[0093] Specifically, the embodiments of the present invention ensure that each storage tank is scheduled for maintenance only once within the optimized time domain by maintaining uniqueness constraints; and ensure that maintenance tasks can only be scheduled within the time period allowed by the capacity of the work window by maintenance resource constraints. When the work window is unavailable at a certain moment, maintenance tasks are completely prohibited from covering that moment. Thus, the formulation of the maintenance plan is precisely matched with the actual available support resources, and the output maintenance plan has practical engineering feasibility.

[0094] In one embodiment of the present invention, solving for the optimal maintenance start time within the maintainable time interval includes: under the constraint of guaranteed resource capacity, minimizing the weighted sum of the expected maintenance cost and premature maintenance resource loss of all oil storage tanks that need to perform maintenance tasks as the optimization objective, and solving the problem using mixed integer linear programming under the following constraints: the maintenance start time of each oil storage tank is no earlier than its earliest maintenance start time and no later than its latest maintenance start time; each oil storage tank is scheduled for maintenance only once within a preset prediction time domain; the total number of work windows occupied by maintenance tasks being performed at any discrete time does not exceed the available work window capacity at that time; the decision variable for the maintenance start time is binary; and the optimal maintenance start time of each oil storage tank and its corresponding maintenance plan are output after solving.

[0095] In a specific embodiment, the objective function and constraints of the MILP optimization model can be found in the relevant descriptions and formulas in step S3 above, and will not be repeated here.

[0096] Specifically, the embodiments of the present invention achieve a flexible balance between maintenance economy and resource conservation by adjusting the weight coefficients; and ensure the engineering executability of the optimized solution from three dimensions—time feasibility, task uniqueness, and resource feasibility—by constraining maintenance time window, maintenance uniqueness, maintenance resource capacity, and decision variable constraints; thereby ensuring that the maintenance plan output after the solution can simultaneously meet the comprehensive balance requirements of performance safety, maintenance economy, and ensuring resource utilization efficiency.

[0097] The advantages and benefits of the method for determining the maintenance timing of oil storage tanks that takes into account resource constraints, as described in the above embodiments of the present invention, will be explained below with reference to a specific embodiment. In this specific embodiment, a case study will be conducted using the condition monitoring data of a certain type of oil storage tank.

[0098] In this specific embodiment, a certain type of oil storage tank is first selected as the experimental object. Each oil storage tank contains two performance degradation indicators and one environmental auxiliary parameter. Therefore, each sampling time contains a total of three multi-source state features.

[0099] Based on the engineering variation range of oil storage tank settlement, tank wall deformation, wall thickness reduction, and operating condition parameters, continuous monitoring sample data is set up to illustrate the implementation process of the TCN-KAN-BRB-MILP method (that is, the oil storage tank maintenance timing determination method considering resource constraints in the above embodiments of the present invention); a total of 100 sets of status monitoring data are collected. According to the data division method, the data from the first to the 90th set are used as historical monitoring data, and the data from the 91st to the 100th set are used as the future prediction interval.

[0100] The input samples for the TCN-KAN model are constructed using a sliding window method. The KAN module uses B-spline basis function expansion to establish a nonlinear mapping relationship between TCN temporal features and future degradation indicators. The model is trained using the Adam optimizer with an initial learning rate of 0.001, a maximum number of training epochs of 180, and a batch size of 16. The main parameters of the TCN-KAN model are shown in Tables 1 and 2, and the main parameters for model training are shown in Table 3.

[0101] Table 1. Main Parameter Settings for TCN Network

[0102] Table 2. Main Parameter Settings for KAN Network

[0103] Table 3. Main Parameter Settings for TCN-KAN Training

[0104] Since different degradation indicators may have different impacts on health status assessment and maintenance decisions, weighted mean square error is used as the training loss function for the TCN-KAN model. The input of the TCN-KAN model is all the state characteristics of the oil storage tank, and the output is two performance degradation indicators of the oil storage tank at future time.

[0105] Then, based on the training samples constructed in the above steps, the TCN-KAN model is trained using historical monitoring data from groups 1 to 90. Before model training, the state features are first standardized to eliminate the influence of different dimensions on the model training process. The standardized multi-source state features are input into the TCN module, and the TCN extracts short-term fluctuation features and long-term degradation trend features within the historical window through causal convolution and dilated convolution. Subsequently, the time-series features output by the TCN are input into the KAN module, and the predicted values ​​of degradation indicators for future times are obtained by expanding the B-spline basis function and mapping it with learnable coefficients.

[0106] This specific embodiment employs a global TCN-KAN prediction structure, using the multi-source state characteristics of the oil storage tank as input, which can simultaneously consider the coupling relationship between performance degradation indicators and environmental auxiliary parameters. This approach aligns with the engineering reality where the working environment and degradation process may be interrelated; after model training, the degradation indicators from group 91 to group 100 are predicted.

[0107] Figure 2 This is a comparison chart of predicted results of settlement performance degradation index according to a specific embodiment of the present invention. Figure 3 This is a comparison chart of predicted deformation performance degradation index results according to a specific embodiment of the present invention. The horizontal axis represents time, and the vertical axis represents the performance degradation index. For example... Figure 2 and Figure 3 The prediction results show that the TCN-KAN model can predict future degradation indicators using historical multi-source state features. The TCN module can extract temporal degradation features from the historical state sequence, and the KAN module can enhance the nonlinear mapping capability through B-spline basis functions, thereby realizing the mapping between multi-source state features and future performance degradation indicators. As can be seen from the experiment, the predicted curve and the actual curve have basically the same trend, indicating that the model can provide effective input for subsequent BRB health status assessment.

[0108] After obtaining the TCN-KAN prediction results, the performance degradation index is input into the BRB model to evaluate the performance status of the oil storage tank: for the data in groups 1 to 90, the actual degradation index data is used as the BRB input; for the data in groups 91 to 100, the TCN-KAN predicted degradation index data is used as the BRB input; thus, the continuous performance status evaluation results for groups 1 to 100 are obtained.

[0109] In this specific embodiment, the utility values ​​corresponding to the BRB output performance status levels are set as shown in Table 4: the larger the utility value, the better the performance status of the oil storage tank; the smaller the utility value, the more severe the degradation of the oil storage tank.

[0110] Figure 4 This is a schematic diagram of the performance utility output value according to a specific embodiment of the present invention. As shown in Figure 4, in order to improve the adaptability of the BRB model to sample data, the BRB parameters are optimized. The optimized parameters include input attribute weights, rule weights, and rule consequent confidence. A target health utility is constructed based on the normalized degradation degree of the training segment, and the BRB parameters are optimized with the goal of minimizing the mean square error between the BRB output health utility and the target health utility. The model is optimized according to the proposed optimization objective function and constraints. The performance utility output value is as follows: Figure 4 As shown, from Figure 4It can be seen that the performance utility value output by BRB decreases steadily over time, fully reflecting the continuous degradation process of the storage tank. The utility curve has clear boundaries and can quickly locate the time intervals corresponding to the two threshold levels, thus defining the legal maintenance window.

[0111] Table 4 Performance Utility Values

[0112] Figure 5 This is a schematic diagram illustrating the changes in the original expected maintenance cost and premature maintenance resource depletion according to a specific embodiment of the present invention. Figure 6 This is a schematic diagram illustrating the solution results of the normalized component costs, the comprehensive maintenance objective function, and the optimal maintenance time according to a specific embodiment of the present invention. Figure 5 and Figure 6 As shown, next, based on the final confidence level output by BRB, an expected relative cost coefficient is established to represent the economic cost difference of maintenance levels corresponding to different performance states. In practical applications, it can be corrected according to tank volume, medium type, construction environment, downtime losses, and maintenance quotations. The maintenance costs corresponding to the performance evaluation levels are shown in Table 5. Since there are differences in the dimensions of resource loss and maintenance costs in actual engineering, the expected maintenance function and resource loss function outputs are normalized to establish a maintenance objective function. Then, considering the maintenance resource capacity constraint, a MILP optimal maintenance timing determination model is established. The model uses the comprehensive maintenance objective function value of each oil tank at the candidate maintenance start time as the objective function coefficient, and uses the maintenance uniqueness constraint and maintenance resource capacity constraint as the main constraints to solve for the maintenance plan with the minimum comprehensive maintenance cost. The curves of the maintenance objective function and the optimal maintenance timing determination are shown in Table 5. Figure 5 and Figure 6 As shown, Figure 5 It is evident that the initial maintenance cost increases with delayed maintenance, while resource depletion decreases with delayed maintenance; these two indicators conflict. After normalization, the following is obtained: Figure 6 The comprehensive cost curve is such that the moment corresponding to the minimum value of the curve is the optimal maintenance time with the lowest comprehensive cost, which takes into account safety, economy and resource utilization.

[0113] Table 5 Maintenance Cost Coefficients for Each Level

[0114] In summary, the method for determining the maintenance timing of oil storage tanks considering resource constraints provided by this invention first collects multi-source state characteristic data sequences of oil storage tanks and predicts state characteristic values ​​at multiple future moments using the TCN-KAN model. TCN is used to extract multi-scale temporal degradation features from historical multi-source state characteristic data sequences, while KAN is used to establish a nonlinear mapping relationship between temporal degradation features and future multi-source state characteristic sequences. Subsequently, the predicted degradation performance index data is input into the BRB model to obtain the future performance state of the oil storage tank. Based on this, an optimal maintenance timing decision model is established using the MILP method. A maintenance time window is generated based on the maintenance initiation threshold and safety state threshold. Considering both the resource loss constraints caused by premature maintenance and the maintenance resource capacity constraints, the optimal maintenance timing with the lowest overall maintenance cost is determined. This avoids resource waste caused by premature maintenance of oil storage tanks, reduces the failure risk caused by late maintenance, and improves the utilization efficiency of resources such as maintenance operation windows. It is suitable for predictive maintenance and optimal scheduling of resource guarantees for oil storage tanks.

[0115] Further embodiments of the present invention disclose a system for determining the maintenance timing of oil storage tanks while taking into account resource constraints. Figure 7 This is a structural block diagram of a system for determining the maintenance timing of oil storage tanks, taking into account resource constraints, according to an embodiment of the present invention. Figure 7 As shown, in one embodiment of the present invention, the oil storage tank maintenance timing determination system 100 considering resource constraints includes: a data acquisition module 110, a prediction module 120, a performance evaluation module 130, and an optimization decision module 140.

[0116] Specifically, the data acquisition module 110 is used to acquire multi-source state characteristic data sequences of oil storage tanks.

[0117] The prediction module 120 incorporates a temporal feature extraction network and a nonlinear mapping network. The prediction module 120 is used to extract multi-scale temporal degradation features from the multi-source state feature data sequence using the temporal feature extraction network, and to establish a mapping relationship between the temporal degradation features and the predicted state feature values ​​at future times using the nonlinear mapping network, thereby outputting the predicted multi-source state feature values ​​at future times.

[0118] The performance evaluation module 130 has a built-in performance evaluation model based on a confidence rule base. The performance evaluation module 130 is used to input the predicted values ​​of multi-source state features at future times into the performance evaluation model and output the confidence distribution of the oil storage tank at each preset performance state level at future times.

[0119] The optimization decision module 140 is used to determine the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time based on the confidence distribution, as well as the performance state utility value of the oil storage tank at each discrete prediction time. Based on the preset maintenance start threshold and the preset safety state threshold, it determines the maintainable time interval of each oil storage tank, with the goal of minimizing the comprehensive maintenance cost. At the same time, it introduces the guarantee resource capacity constraint into the comprehensive maintenance cost and solves the optimal maintenance start time within the maintainable time interval. The comprehensive maintenance cost is at least composed of the expected maintenance cost and the premature maintenance resource loss weighted together.

[0120] It should be noted that the specific implementation of the system 100 for determining the maintenance timing of oil storage tanks considering resource constraints in the embodiments of the present invention is similar to the specific implementation described in the method for determining the maintenance timing of oil storage tanks considering resource constraints in the above embodiments of the present invention, and therefore has similar technical effects. For details, please refer to the description of the method for determining the maintenance timing of oil storage tanks considering resource constraints. To reduce redundancy, it will not be repeated here.

[0121] Further embodiments of the present invention also disclose an electronic device, Figure 8 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Figure 8 As shown, in one embodiment of the present invention, the electronic device includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; the memory is used to store computer programs; when the processor executes the computer program stored in the memory, it implements the method for determining the maintenance timing of oil storage tanks considering resource constraints as described in any of the above embodiments of the present invention.

[0122] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0123] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0124] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0125] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0126] The method provided in this invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc. No limitation is made herein; any electronic device that can implement this invention falls within the protection scope of this invention.

[0127] It should be noted that the specific implementation of the electronic device in the embodiments of the present invention is similar to the specific implementation described in the above embodiments of the present invention regarding the method for determining the maintenance timing of oil storage tanks considering resource constraints. Therefore, it also has similar technical effects. For details, please refer to the description in the section on determining the maintenance timing of oil storage tanks considering resource constraints. To reduce redundancy, it will not be repeated here.

[0128] Further embodiments of the present invention disclose a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method for determining the maintenance timing of oil storage tanks taking into account resource constraints as described in any of the above embodiments of the present invention.

[0129] It should be noted that the specific implementation of the computer-readable storage medium in the embodiments of the present invention is similar to the specific implementation described in the above embodiments of the present invention regarding the method for determining the maintenance timing of oil storage tanks considering resource constraints. Therefore, it also has similar technical effects. For details, please refer to the description in the section on determining the maintenance timing of oil storage tanks considering resource constraints. To reduce redundancy, it will not be repeated here.

[0130] In the description of this specification, references to terms such as "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example.

[0131] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention.

Claims

1. A method for determining the maintenance timing of oil storage tanks while considering resource constraints, characterized in that, Includes the following steps: A multi-source state feature data sequence of an oil storage tank is obtained. Multi-scale temporal degradation features are extracted from the multi-source state feature data sequence based on a temporal feature extraction network. A mapping relationship between the multi-scale temporal degradation features and the predicted state feature values ​​at future times is established based on a nonlinear mapping network to obtain the predicted multi-source state feature values ​​at future times. The predicted values ​​of the multi-source state features at the future time are input into the performance evaluation model based on the confidence rule base, and the confidence distribution of the oil storage tank at each preset performance state level at the future time is output. Based on the confidence distribution, the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time are determined, as well as the performance state utility value of the oil storage tank at each discrete prediction time. According to the preset maintenance start threshold and the preset safety state threshold, the maintainable time interval of each oil storage tank is determined, with the goal of minimizing the comprehensive maintenance cost. At the same time, the guaranteed resource capacity constraint is introduced into the comprehensive maintenance cost. The optimal maintenance start time is solved within the maintainable time interval. The comprehensive maintenance cost is at least composed of the expected maintenance cost and the premature maintenance resource loss weighted together.

2. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The temporal feature extraction network is a temporal convolutional network; wherein... The temporal convolutional network is a one-dimensional causal convolutional structure, so that the prediction of future states depends only on information from the current and historical moments. The temporal convolutional network expands its receptive field by progressively increasing the dilation coefficient to simultaneously extract short-term fluctuation features and long-term degradation trend features of performance status. The temporal convolutional network employs residual connections within each residual block, adding the input of the residual block to its output after convolution transformation to form a cross-layer addition structure.

3. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The nonlinear mapping network is a Kolmogrove-Arnold network, and the final output of the Kolmogrove-Arnold network is the multi-source state feature prediction value for the future time step; wherein, The Kolmogrove-Arnold network is composed of multiple stacked Kolmogrove-Arnold layers. In each Kolmogrove-Arnold layer, the edge function connecting the input and output nodes is composed of a superposition of a basic function and a B-spline function. The B-spline function achieves a nonlinear mapping from input to output through learnable spline coefficients; The base function and the B-spline function are each set with learnable scaling parameters, which are used to adjust the contribution ratio of the base function and the B-spline function.

4. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, In the performance evaluation model based on the confidence rule base, each confidence rule includes a reference level, rule weight, attribute weight, and output confidence distribution for each input state feature index. The step of inputting the predicted multi-source state features of the future time into a performance evaluation model based on a confidence rule base, and outputting the confidence distribution of the oil storage tank at each preset performance state level in the future time, includes: The performance degradation index in the multi-source state feature prediction value of the future time is matched with the reference level in each confidence rule, and the matching degree of each performance degradation index under each confidence rule is calculated. The total matching degree of each confidence rule is calculated based on the attribute weight of each performance degradation metric, and the activation weight of each confidence rule is determined by combining the rule weight. The output confidence distribution preset in the consequent of each confidence rule is obtained as a confidence vector. Based on the evidence reasoning parsing algorithm, each confidence vector is weighted and fused with the activation weight to obtain the comprehensive confidence distribution of the oil storage tank.

5. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The process of determining the maintainable time interval for each oil storage tank based on a preset maintenance initiation threshold and a preset safety state threshold includes: The performance state utility value at each discrete prediction time is compared with the maintenance start threshold, and the time when the performance state utility value first falls below the maintenance start threshold is set as the earliest maintenance start time of the oil storage tank. The performance state utility value at each discrete prediction time is compared with the safety state threshold, and the moment when the performance state utility value first falls below the safety state threshold is set as the safety critical moment. By shifting the safety critical moment forward by a preset maintenance duration for the oil storage tank, the latest maintenance start time of the oil storage tank is obtained. The maintainable time interval is defined by combining the earliest maintenance start time and the latest maintenance start time.

6. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The determination of the expected maintenance cost includes: A pre-established correspondence between the performance status level and maintenance level of oil storage tanks is established, wherein each performance status level corresponds to a maintenance level, and each maintenance level has a corresponding maintenance cost coefficient; The confidence level of each performance status level in the confidence distribution output by the performance evaluation model is used as a weight to weight the maintenance cost coefficient corresponding to each performance status level, and the weighted result is used as the expected maintenance cost of the oil storage tank at the corresponding maintenance start time.

7. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The determination of premature maintenance resource loss includes: Using the earliest maintenance start time of the oil storage tank as a benchmark, a function is constructed that decreases non-linearly as the candidate maintenance start time is delayed. The function value is used as the premature maintenance resource loss of the oil storage tank at the candidate maintenance start time. The function is used to quantify the waste of remaining life resources caused by premature shutdown and maintenance of oil storage tanks before their performance status has fully degraded; An independent resource loss weighting coefficient is set for each oil storage tank. The resource loss weighting coefficient is multiplied by the function value, and the product is used as the final value of premature maintenance resource loss for the corresponding oil storage tank, so as to characterize the difference in the degree of premature maintenance resource waste caused by different importance or maintenance urgency among the oil storage tanks.

8. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, The inclusion of guaranteed resource capacity constraints in the overall maintenance cost includes: Identify the set of oil storage tanks that require maintenance tasks, and limit each oil storage tank in the set to perform a maintenance task at most once within a preset prediction time domain; A sequence of available job window capacities is defined. For any discrete time, the total number of all ongoing maintenance tasks covering that time is limited to no more than the available job window capacity corresponding to that time. The sequence of available job window capacities is used to characterize the maximum number of maintenance tasks that can be accommodated simultaneously at each discrete time.

9. The method for determining the maintenance timing of oil storage tanks considering resource constraints according to claim 1, characterized in that, Solving for the optimal maintenance start time within the maintainable time interval includes: Under the constraint of guaranteed resource capacity, the optimization objective is to minimize the weighted sum of the expected maintenance cost and the premature maintenance resource loss of all oil storage tanks that require maintenance tasks, and to solve this problem using mixed-integer linear programming under the following constraints: The maintenance start time for each oil storage tank shall not be earlier than its earliest maintenance start time and not later than its latest maintenance start time; Each oil storage tank will be scheduled for maintenance only once within the pre-defined forecast timeframe; The total number of job windows occupied by maintenance tasks at any discrete moment shall not exceed the available job window capacity at that moment; The decision variables at the initial time are maintained to have binary values; After solving, the optimal maintenance start time and corresponding maintenance plan for each oil storage tank are output.

10. A system for determining the maintenance timing of oil storage tanks while considering resource constraints, characterized in that, include: The data acquisition module is used to acquire multi-source state characteristic data sequences of oil storage tanks; The prediction module has a built-in temporal feature extraction network and a nonlinear mapping network. The prediction module is used to extract multi-scale temporal degradation features from the multi-source state feature data sequence using the temporal feature extraction network, and to establish a mapping relationship between the temporal degradation features and the predicted state feature values ​​at future times using the nonlinear mapping network, and output the predicted multi-source state feature values ​​at future times. The performance evaluation module has a built-in performance evaluation model based on a confidence rule base. The performance evaluation module is used to input the predicted values ​​of the multi-source state features at the future time into the performance evaluation model and output the confidence distribution of the oil storage tank at each preset performance state level at the future time. The optimization decision module is used to determine the expected maintenance cost and premature maintenance resource loss of the oil storage tank at each candidate maintenance start time based on the confidence distribution, and to determine the performance state utility value of the oil storage tank at each discrete prediction time. Based on a preset maintenance start threshold and a preset safety state threshold, it determines the maintainable time interval for each oil storage tank, aiming to minimize the overall maintenance cost. Simultaneously, a guaranteed resource capacity constraint is introduced into the overall maintenance cost. The optimal maintenance start time is then solved within the maintainable time interval. The overall maintenance cost is at least a weighted average of the expected maintenance cost and the premature maintenance resource loss.

Images