Data-driven organic heat carrier deterioration risk intelligent early warning method and system

Abstract

This invention relates to the field of industrial equipment safety monitoring technology, and discloses a data-driven intelligent early warning method and system for the degradation risk of organic heat transfer fluids. The method includes: acquiring multimodal monitoring data of the organic heat transfer fluid, performing preprocessing and feature extraction to obtain entropy yield and acoustic emission activity index; inputting the entropy yield, acoustic emission activity index, and medium quality parameters into a mechanism classification neural network model to obtain the determination probability of the degradation mechanism; according to the determination probability, inputting the feature vector into the corresponding target branch neural network to obtain an instantaneous risk index; predicting the time required for the medium quality parameters and entropy yield to reach a safety threshold, taking the minimum value as the predicted failure time, and calculating the degradation resilience value in conjunction with the planned maintenance time; and determining the comprehensive risk level based on the instantaneous risk index and the degradation resilience value. This solution can achieve early detection of degradation trends and distinguish different degradation mechanisms, improving assessment accuracy and making maintenance strategies more targeted.

Description

Technical Field

[0001] This invention belongs to the field of industrial equipment safety monitoring technology, and in particular relates to a data-driven intelligent early warning method and system for the deterioration risk of organic heat transfer fluids. Background Technology

[0002] Organic heat carrier boilers are high-temperature, low-pressure heat transfer devices widely used in chemical engineering, solar thermal power generation, and other fields. Their operation relies on the transfer of heat through a closed-loop system using organic heat carriers. However, under long-term high-temperature conditions, the medium is prone to oxidation, thermal decomposition, and other chemical degradation reactions, generating acidic substances, polymers, and residual carbon. This leads to system coking, increased flow resistance, and in severe cases, can cause furnace tube overheating, tube rupture, or even fires. To ensure operational safety, current national standards stipulate that the medium must be sampled and tested offline periodically to monitor key physicochemical indicators.

[0003] Current technologies for monitoring the degradation of organic heat transfer fluids have significant limitations. First, monitoring primarily relies on offline laboratory analysis, depending on periodic testing of indicators such as acid value, viscosity, and carbon residue. This data update lags, making it difficult to capture the dynamic evolution of early-stage degradation, often issuing warnings only after significant deterioration has occurred. Second, existing systems lack the ability to identify degradation mechanisms, failing to distinguish between different degradation pathways such as oxidation and thermal decomposition. This results in a lack of targeted maintenance measures and may exacerbate degradation due to misjudgments. Finally, traditional methods rely solely on static thresholds to determine risk status, neglecting the system's resilience and safety margins under disturbances, thus failing to support predictive maintenance decisions. Furthermore, while some online sensors (such as dielectric constant and viscometers) are used, they fail to effectively integrate multi-source information such as thermodynamic parameters and acoustic signals to form a comprehensive and collaborative degradation state perception system, thus limiting the timeliness and accuracy of warnings. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a data-driven intelligent early warning method and system for the deterioration risk of organic heat transfer fluids. This method can detect deterioration trends in advance, distinguish different deterioration mechanisms, improve assessment accuracy, and make maintenance strategies more targeted.

[0005] This invention provides a data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids, the method comprising the following steps:

[0006] S1. Acquire multimodal monitoring data of the organic heat transfer fluid, including operating parameters, medium quality parameters, and acoustic emission signals;

[0007] S2. Preprocess and extract features from the multimodal monitoring data to obtain the entropy yield and acoustic emission activity index;

[0008] S3. Based on entropy yield, acoustic emission activity index and medium quality parameters, construct feature vectors, input them into a pre-trained mechanism classification neural network model, and obtain the determination probability of the organic heat carrier being in each preset degradation mechanism.

[0009] S4. Based on the preset degradation mechanism with the highest judgment probability, select the corresponding target branch neural network from the pre-trained physical constraint branch neural network, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value.

[0010] S5. Based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid, predict the time required for each indicator to reach the safety threshold, take the minimum value among the times required for each indicator to reach the safety threshold as the predicted failure time, and calculate the deterioration toughness value in combination with the planned maintenance time.

[0011] S6. Based on the instantaneous risk index prediction value and the deterioration resilience value, determine the comprehensive risk level and generate the corresponding resilience maintenance strategy.

[0012] Furthermore, in S1, the operating parameters include temperature, pressure, and flow rate;

[0013] Medium quality parameters include acid value, kinematic viscosity, and carbon residue.

[0014] Furthermore, in S2, the formula for calculating the entropy production rate is as follows:

[0015] ;

[0016] in, This represents the entropy yield of the organic heat transfer fluid. This represents the total thermal power of the input system. This represents the effective thermal power output by the system. This indicates the average temperature of the furnace. Indicates ambient temperature.

[0017] Furthermore, in S2, the formula for calculating the acoustic emission activity index is as follows:

[0018] ;

[0019] in, The acoustic emission activity index represents the acoustic emission activity of organic heat transfer fluids. This represents the root mean square value of the measured acoustic emission signal. The critical root-mean-square value represents the acoustic emission signal. This represents the root mean square value of the background noise of the system under normal conditions.

[0020] Furthermore, in S3, the preset degradation mechanism includes at least the oxidation degradation mechanism and the thermal cracking-coking degradation mechanism.

[0021] Furthermore, in S4, based on the preset degradation mechanism with the highest judgment probability, the corresponding target branch neural network is selected from the pre-trained physical constraint branch neural network, including:

[0022] When the probability of determining the oxidative degradation mechanism is at its maximum, the target branch neural network is determined to be the first physical constraint branch neural network; the first physical constraint loss function used by the first physical constraint branch neural network during training is constructed based on the calculation formula of the first degradation immunity index;

[0023] When the probability of determining the thermal cracking-coking degradation mechanism is at its maximum, the target branch neural network is determined to be the second physical constraint branch neural network; the second physical constraint loss function used by the second physical constraint branch neural network during training is constructed based on the calculation formula of the second degradation immunity index.

[0024] Furthermore, the formula for calculating the first physical constraint loss function is as follows:

[0025] ;

[0026] ;

[0027] ;

[0028] in, Represents the first physical constraint loss function. This indicates task loss due to oxidative degradation mechanism. This indicates the physical constraint loss in the oxidative degradation mechanism. This represents the hyperparameter of the oxidative degradation mechanism, where i represents the i-th sample and N represents the total number of samples. This represents the predicted value of the immune index. This indicates the true value of the immune index representing the oxidative degradation mechanism. This represents the predicted value of the instantaneous risk index. This indicates the true value of the instantaneous risk index for the oxidative degradation mechanism;

[0029] The formula for calculating the true value of the immune index in the oxidative degradation mechanism is as follows:

[0030] ;

[0031] in, , , These represent the acid value weighting coefficient, entropy yield weighting coefficient, and nitrogen sealing weighting coefficient, respectively, where A represents the current acid value. Indicates the acid value safety threshold. This represents the entropy yield of the organic heat transfer fluid. The critical threshold representing the entropy productivity. This indicates the effectiveness of the nitrogen blanketing system.

[0032] Furthermore, the formula for calculating the second physical constraint loss function is as follows:

[0033] ;

[0034] ;

[0035] ;

[0036] in, This represents the second physical constraint loss function. This indicates the task loss due to the thermal cracking-coking degradation mechanism. This indicates the physical constraint loss in the thermal cracking-coking degradation mechanism. This represents the hyperparameter of the thermal cracking-coking degradation mechanism, where i represents the i-th sample and N represents the total number of samples. This represents the predicted value of the immune index. This indicates the true value of the immune index for the thermal decomposition-coking degradation mechanism. This represents the predicted value of the instantaneous risk index. This represents the true value of the instantaneous risk index for the thermal cracking-coking degradation mechanism;

[0037] The formula for calculating the true value of the immune index of the thermal decomposition-coking degradation mechanism is as follows:

[0038] ;

[0039] in, , , , These represent the viscosity weighting coefficient, residual carbon weighting coefficient, acoustic emission activity index weighting coefficient, and entropy yield weighting coefficient, respectively. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value. Indicates acoustic emission activity index, This represents the entropy yield of the organic heat transfer fluid. The critical threshold representing the entropy production rate.

[0040] Furthermore, in S4, the formula for calculating the true value of the instantaneous risk index is as follows:

[0041] ;

[0042] ;

[0043] in, R represents the true value of the instantaneous risk index for degradation mechanism of type X, and R represents the base risk index. , , These represent the acid value weighting coefficient, viscosity weighting coefficient, and residual carbon weighting coefficient, respectively, with A representing the current acid value. Indicates the acid value safety threshold. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value. This represents the true value of the immune index for the degradation mechanism of type X. This represents the weighting coefficient of acoustic emission activities in the risk. This indicates the acoustic emission activity index.

[0044] Furthermore, in S5, the formula for calculating the degradation toughness value is as follows:

[0045] ;

[0046] in, Indicates the deterioration toughness value. Indicates the predicted failure time. Indicates the planned maintenance time.

[0047] This invention also provides a data-driven intelligent early warning system for the degradation risk of organic heat transfer fluids, used to execute the above-mentioned data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids, characterized in that the system includes the following modules:

[0048] The data acquisition module is used to acquire multimodal monitoring data of the organic heat transfer fluid, including operating parameters, medium quality parameters, and acoustic emission signals.

[0049] The feature extraction module is used to preprocess and extract features from multimodal monitoring data to obtain entropy yield and acoustic emission activity index;

[0050] The degradation mechanism classification module is used to construct feature vectors based on entropy yield, acoustic emission activity index and medium quality parameters, input them into a pre-trained mechanism classification neural network model, and obtain the determination probability of the organic heat carrier being in each preset degradation mechanism.

[0051] The instantaneous risk index calculation module is used to select the corresponding target branch neural network from the pre-trained physical constraint branch neural network based on the preset degradation mechanism with the highest judgment probability, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value.

[0052] The degradation toughness value calculation module is used to predict the time required for each indicator to reach the safety threshold based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid. The minimum value among the times required for each indicator to reach the safety threshold is used as the predicted failure time, and the degradation toughness value is calculated in combination with the planned maintenance time.

[0053] The early warning module is used to determine the comprehensive risk level and generate corresponding resilience maintenance strategies based on the instantaneous risk index prediction value and the deterioration resilience value.

[0054] The embodiments of the present invention have the following technical effects:

[0055] This invention integrates multi-source data such as operating parameters, medium quality parameters, and acoustic emission signals, and innovatively uses system entropy yield as the core evaluation indicator. It can sensitively reflect the irreversible loss of overall boiler system energy efficiency. Compared to traditional threshold judgments relying solely on offline indicators such as acid value and viscosity, entropy yield can detect the trend of system efficiency decline caused by coking on heat transfer surfaces and overall medium deterioration earlier, thus achieving macroscopic early warning. Simultaneously, by constructing a mechanism classification neural network model, the system can automatically and accurately identify different dominant mechanisms such as oxidation deterioration and thermal cracking-coking deterioration, overcoming the rigidity and incomplete coverage problems that may exist in traditional fixed-rule judgments. This provides a more intelligent basis for subsequent targeted maintenance strategies. This approach provides a more adaptive basis for mechanism diagnosis. It innovatively proposes a quantitative indicator called "deterioration resilience," which compares the failure time predicted by a time-series prediction model with the planned maintenance time. This shifts the assessment focus from static "whether the standard is exceeded" to dynamic "how much safety margin remains," achieving a leap from simple risk warning to risk-resilience collaborative assessment. Furthermore, the proposed "physical constraint loss function" embeds the degradation immunity index of different mechanisms as a strong constraint into the branch training process of the neural network. This enables the entire model to possess both data-driven complex pattern learning capabilities and strictly ensures that the output results conform to physical laws, significantly improving the model's interpretability, generalization ability, and reliability under small sample conditions. This allows decision-makers not only to understand the level of risk but also to grasp the system's ability to maintain function and resist failure even after degradation has occurred. It provides more intelligent and reliable key data support for shifting from "passive maintenance" to "proactive early warning and resilience maintenance" based on safety margins, enhancing the scientific and forward-looking nature of maintenance decisions and providing comprehensive technical support for achieving safer, more economical, and intelligent operation of organic heat carrier boiler systems. Attached Figure Description

[0056] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0057] Figure 1 This is a flowchart of a data-driven intelligent early warning method for the degradation risk of organic heat carriers provided in an embodiment of the present invention;

[0058] Figure 2 This is a schematic diagram of the comprehensive assessment matrix for the degradation risk of organic heat transfer fluid provided in an embodiment of the present invention;

[0059] Figure 3 This is a schematic diagram of the structure of the data-driven intelligent early warning system for the degradation risk of organic heat carriers provided in an embodiment of the present invention. Detailed Implementation

[0060] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0061] Figure 1 This is a flowchart of a data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids provided in an embodiment of the present invention. See also... Figure 1 The method includes the following steps:

[0062] S1. Acquire multimodal monitoring data of the organic heat transfer fluid, including operating parameters, medium quality parameters, and acoustic emission signals.

[0063] The system acquires operating parameters, media quality parameters, and acoustic emission signals through the boiler control system and sensors. Operating parameters include temperature, pressure, and flow rate; media quality parameters include acid value, kinematic viscosity, and carbon residue. Acoustic emission signals are collected by installing acoustic emission sensors on the boiler body, circulating pump, and key pipelines.

[0064] S2. Preprocess and extract features from the multimodal monitoring data to obtain the entropy yield and acoustic emission activity index.

[0065] In some embodiments, preprocessing may include data cleaning and alignment, processing outliers and missing values ​​from data from different sources, and synchronizing discrete oil analysis data (i.e., medium quality parameters) with continuous operating parameters and high-frequency acoustic emission signals acquired by the boiler control system using a unified timestamp.

[0066] In some embodiments, feature extraction includes calculating entropy productivity and acoustic emission activity index. Entropy productivity characterizes the degree of energy dissipation caused by irreversible processes within the system (such as combustion, heat transfer, and friction). The entropy productivity of a system is a sensitive macroscopic indicator measuring the overall thermodynamic integrity and degree of degradation of the boiler system. Compared to monitoring only the chemical parameters of the medium itself (such as acid value), entropy productivity can detect system-level performance degradation trends caused by coking on heat transfer surfaces and increased flow resistance earlier, achieving early macroscopic warning. The formula for calculating entropy productivity is as follows:

[0067] ;

[0068] in, This represents the entropy yield of the organic heat transfer fluid. An increase in this value indicates a greater overall irreversible loss in the system, decreased energy efficiency, and a higher risk of degradation. This represents the total thermal power input to the system. For gas / oil boilers, it is calculated using a fuel flow meter and the lower heating value of the fuel. For electric boilers, it is the electrical power measured by the electricity meter. The effective thermal power output of the system is calculated by measuring the total mass flow rate, inlet and outlet temperatures, and specific heat capacity of the organic heat transfer fluid; that is, the net heat actually absorbed by the medium and used in the process. This indicates the average temperature of the furnace. This indicates the ambient temperature. A sustained and significant increase in entropy production rate indicates a decrease in the overall thermal efficiency of the system, which may be caused by factors such as overall medium deterioration, severe coking inside the boiler, or reduced combustion efficiency.

[0069] In some embodiments, the acquired acoustic emission signals are processed to extract an acoustic emission activity index. This index is used to quantify the activity level of localized physical degradation (such as coking and spalling), and its introduction compensates for the limitations of chemical and thermodynamic parameters in monitoring localized microscopic physical failures. The formula for calculating the acoustic emission activity index is as follows:

[0070] ;

[0071] in, The acoustic emission activity index represents the acoustic emission activity of organic heat transfer fluids. This represents the root mean square value of the measured acoustic emission signal, reflecting the average energy level of the signal. The critical root mean square value representing the acoustic emission signal can be set based on historical fault data or engineering experience, and represents the intensity of acoustic emission activity that may cause serious problems. This represents the root mean square value of the background noise of the system under normal conditions.

[0072] Furthermore, the entropy production rate and acoustic emission activity index are normalized.

[0073] S3. Construct feature vectors based on entropy yield, acoustic emission activity index and medium quality parameters, input them into a pre-trained mechanism classification neural network model, and obtain the probability of the organic heat carrier being in each preset degradation mechanism.

[0074] In this embodiment, the preset degradation mechanism includes at least the oxidation degradation mechanism and the thermal cracking-coking degradation mechanism.

[0075] The pre-trained mechanistic classification neural network model can employ a multilayer perceptron, using multiple hidden layers to perform nonlinear transformations and combinations on the input features, learning the complex patterns of different degradation mechanisms in the feature space. The model's output layer uses the Softmax activation function, and its output is a probability distribution vector. For example, if the preset degradation mechanisms are "oxidative degradation" and "thermal cracking-coking degradation," the output includes [P]. 氧化 , P 热裂解 ]. Among them, P 氧化 P represents the probability that the current state of the organic heat transfer fluid is determined to be oxidatively degraded. 热裂解 This represents the probability that the current state of the organic heat transfer fluid is determined to be thermal decomposition-coking degradation, and the sum of the two is 1.

[0076] S4. Based on the preset degradation mechanism with the highest judgment probability, select the corresponding target branch neural network from the pre-trained physical constraint branch neural network, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value.

[0077] Among them, the target branch neural network selected from the pre-trained physical constraint branch neural network according to the preset degradation mechanism with the highest judgment probability includes:

[0078] When the probability of determining the oxidative degradation mechanism is at its maximum, the target branch neural network is determined to be the first physical constraint branch neural network; the first physical constraint loss function used by the first physical constraint branch neural network during training is constructed based on the calculation formula of the first degradation immunity index;

[0079] When the probability of determining the thermal cracking-coking degradation mechanism is at its maximum, the target branch neural network is determined to be the second physical constraint branch neural network; the second physical constraint loss function used by the second physical constraint branch neural network during training is constructed based on the calculation formula of the second degradation immunity index.

[0080] The pre-trained physical constraint branch neural network employs a Physical Information Neural Network (PINN). For example, PINN includes:

[0081] Input layer: Receives a feature vector consisting of entropy yield, acoustic emission activity index, and medium quality parameters (acid value, viscosity, carbon residue, etc.);

[0082] Hidden layers: These contain multiple fully connected layers, and the number of layers and neurons can be set according to the complexity of the problem; smooth activation functions such as tanh and sigmoid can be used to model continuous changes in physical fields.

[0083] Output layer: Directly outputs the predicted value of the instantaneous risk index;

[0084] The physical constraint loss function consists of data-driven loss and physical residual loss.

[0085] In some embodiments, the first physical constraint loss function is calculated using the following formula:

[0086] ;

[0087] ;

[0088] ;

[0089] in, Represents the first physical constraint loss function. This indicates task loss due to oxidative degradation mechanism. This indicates the physical constraint loss in the oxidative degradation mechanism. This represents the hyperparameter of the oxidative degradation mechanism, where i represents the i-th sample and N represents the total number of samples. This represents the predicted value of the immune index. This represents the true value of the immune index indicating the mechanism of oxidative degradation; the closer the value is to 1, the stronger the resistance. This represents the predicted value of the instantaneous risk index. This indicates the true value of the instantaneous risk index for the oxidative degradation mechanism;

[0090] When the diagnosis is dominated by "oxidative degradation", the calculation formula should emphasize the system's antioxidant capacity and de-emphasize non-primary factors such as viscosity and carbon residue. The calculation formula for the true value of the oxidative degradation mechanism immune index is as follows:

[0091] ;

[0092] in, , , These represent the acid value weighting coefficient, entropy yield weighting coefficient, and nitrogen sealing weighting coefficient, respectively, where A represents the current acid value. The acid value represents the safe threshold; the value can be determined by referring to national regulations, etc. This represents the entropy yield of the organic heat transfer fluid. The critical threshold representing the entropy production rate can be calibrated using historical failure data. Indicates the effectiveness of the nitrogen blanketing system. ∈[0,1], where 1 represents a completely sealed system with nitrogen protection, which can be determined through evaluation by pressure sensors, leak detection, etc.

[0093] In some embodiments, the second physical constraint loss function is calculated using the following formula:

[0094] ;

[0095] ;

[0096] ;

[0097] in, This represents the second physical constraint loss function. This indicates the task loss due to the thermal cracking-coking degradation mechanism. This indicates the physical constraint loss in the thermal cracking-coking degradation mechanism. The hyperparameters representing the thermal decomposition-coking degradation mechanism are used to balance the strength of data fitting and adherence to physical laws. i represents the i-th sample, and N represents the total number of samples. This represents the predicted value of the immune index. This indicates the true value of the immune index for the thermal decomposition-coking degradation mechanism. This represents the predicted value of the instantaneous risk index. This represents the true value of the instantaneous risk index for the thermal cracking-coking degradation mechanism;

[0098] When the diagnosis is dominated by "thermal decomposition-coking", the calculation formula should focus on the ability to inhibit decomposition and coking. The calculation formula for the true value of the immune index of the thermal decomposition-coking degradation mechanism is as follows:

[0099] ;

[0100] in, , , , These represent the viscosity weighting coefficient, residual carbon weighting coefficient, acoustic emission activity index weighting coefficient, and entropy yield weighting coefficient, respectively. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value, which can be referenced from national standard limits. Indicates acoustic emission activity index, This represents the entropy yield of the organic heat transfer fluid. This represents the critical threshold for entropy production rate. It emphasizes the suppression of pyrolysis product accumulation and localized thermal stress, highlighting the importance of acoustic signals. The acoustic emission activity index is incorporated into the immunity calculation to comprehensively assess the system's overall ability to resist pyrolysis product accumulation and coking physical damage.

[0101] In some embodiments, the formula for calculating the true value of the instantaneous risk index is as follows:

[0102] ;

[0103] ;

[0104] in, R represents the true value of the instantaneous risk index for degradation mechanism X, where X can be O (oxidative degradation) or T (thermal cracking-coking degradation). R represents the basic risk index, which is a normalized weighted sum based on traditional media quality parameters (acid value, viscosity, carbon residue), reflecting the degree of chemical degradation of the media. , , These represent the acid value weighting coefficient, viscosity weighting coefficient, and residual carbon weighting coefficient, respectively, with A representing the current acid value. Indicates the acid value safety threshold. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value. This indicates that, based on the current dominant mechanism X and the selected deterioration immune index, the lower the immunity, the greater the risk. The weighting coefficient of acoustic emission activities in the risk is determined based on system characteristics. The acoustic emission activity index reflects the level of activity in local structural deterioration.

[0105] This formula combines chemical degradation (basic risk R), systemic resistance (immune attenuation factor) and other factors. ) and physical instantaneous threats (local activity amplification factor) The combination of these three factors transforms the calculated instantaneous risk index from an isolated numerical value into an intelligent indicator that comprehensively, dynamically, and hierarchically reflects the overall risk status of the system. It not only determines whether the system is currently failing but also suggests how soon it might fail and whether an immediate threat exists. During the training phase, using the instantaneous risk index as a label for the neural network forces it to learn to extract complex patterns related to these three dimensions (chemical degradation, system resilience, and physical instantaneous threat) from the input features, ultimately outputting a predicted value that closely approximates this "standard answer" in both numerical and physical meaning. Through this fitting formula, the model learns "how these features interact to lead to high risk" (causal / logical relationship). This enables the final model to predict... It better reflects the internal mechanisms of risk formation, improving its generalization ability and the reliability of judgment under boundary conditions.

[0106] S5. Based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid, predict the time required for each indicator to reach the safety threshold. Take the minimum value among the times required for each indicator to reach the safety threshold as the predicted failure time, and calculate the deterioration toughness value in combination with the planned maintenance time.

[0107] In some embodiments, historical time series are established for each indicator to be predicted (acid value, viscosity, residual carbon value, entropy yield), and a pre-trained prediction model (such as LSTM) is used for extrapolation prediction to predict the future change curve of each indicator. The predicted curve of each indicator is compared with its corresponding safety threshold to determine the time point when its predicted curve first crosses its own safety threshold line. The time length from the current point to that point is calculated, which is the time required for the indicator to reach the safety threshold. Finally, the minimum value of the time required for each indicator to reach the safety threshold is taken as the predicted failure time of the system. The overall safe operating time of the system depends on the first critical parameter to fail.

[0108] For example, suppose the system monitors three key metrics: viscosity, acid value, and entropy yield. The model predicts that viscosity will exceed its safety threshold in 60 days, acid value in 90 days, and entropy yield in 45 days. Therefore, the predicted failure time is 45 days. This approach ensures both the safety and forward-looking nature of the early warning system, forcing maintenance decisions to focus on the parameters that deteriorate first, thus allowing action to be taken before the earliest potential risks materialize.

[0109] In some embodiments, the degradation resilience value represents the system's ability to maintain core functions without failure and to withstand degradation until the planned maintenance time. The degradation resilience value is calculated using the following formula:

[0110] ;

[0111] in, This represents the system's resilience to degradation; a value greater than 1 indicates sufficient system safety margin, while a value less than 1 indicates the system requires early maintenance. Indicates the predicted failure time. This indicates the planned maintenance time, which is the planned time for the next downtime maintenance starting from today, for example, 60 days.

[0112] By introducing the concept of degradation resilience, the assessment focus is extended from the current state to the future time dimension, providing a direct basis for predictive maintenance decisions.

[0113] S6. Based on the instantaneous risk index prediction value and the deterioration resilience value, determine the comprehensive risk level and generate the corresponding resilience maintenance strategy.

[0114] In some embodiments, a two-dimensional decision matrix is ​​established with the instantaneous risk index prediction value on the horizontal axis and the degradation resilience value on the vertical axis. The matrix is ​​divided into multiple regions, each corresponding to a specific comprehensive risk level and action instructions. For example, by setting thresholds, the instantaneous risk index prediction value is divided into high risk, medium risk, and low risk, and the degradation resilience value is divided into high resilience, medium resilience, and low resilience, thereby constructing nine combined states. Figure 2 This is a schematic diagram of the comprehensive assessment matrix for the degradation risk of organic heat transfer fluid provided in an embodiment of the present invention. See also: Figure 2 Nine combined states can be categorized into different comprehensive risk levels. For example, low risk + high resilience and low risk + low resilience both belong to the comprehensive low risk level, while high risk + low resilience belongs to the comprehensive high risk level. Combining the comprehensive risk level with the degradation-dominant type determined by the mechanistic classification neural network, targeted resilience maintenance strategies are generated, such as adding high-temperature anti-cracking agents, adjusting boiler outlet temperature, and adjusting maintenance time. By constructing a two-dimensional decision matrix that integrates risk severity (the instantaneous risk index prediction value assessed by the neural network model) and time urgency (deterioration resilience value), a scientific and precise risk level classification is achieved. This decision-making process, based on the model's intelligent predictive output, improves the accuracy and action guidance of early warning information, realizing a fundamental shift from passive threshold alarm responses to proactive, predictive, and resilience maintenance based on model predictions.

[0115] This invention introduces system entropy yield as a key evaluation indicator. Entropy yield directly characterizes the degree of irreversible loss during the system's heat-work conversion process. When the organic heat carrier deteriorates (e.g., oxidation, coking) leading to a decrease in heat transfer efficiency or an increase in flow resistance, the entropy yield will significantly increase. By integrating this macroscopic thermodynamic efficiency indicator with traditional medium quality parameters (acid value, viscosity) and the acoustic emission activity index reflecting local physical states, this invention overcomes the limitations of traditional reliance on offline oil analysis. It achieves earlier detection of deterioration trends and can automatically and accurately identify the dominant deterioration mechanism (e.g., oxidation or thermal cracking) through a pre-trained mechanistic classification neural network model. This overcomes the potential for incomplete coverage and rigidity in traditional fixed rules, providing a basis for targeted maintenance. Furthermore, it innovatively quantifies the system's "resilience." By predicting the time when key indicators reach safety thresholds using a time series forecasting model and comparing it with planned maintenance time, a degradation resilience value is calculated. This value assesses the system's time margin to maintain function and resist failure after degradation has occurred. This indicator shifts the early warning perspective from static "whether it exceeds the standard" to dynamic "how long can it still operate safely?", enabling decision-makers to grasp the system's safety buffer capacity. This achieves a shift in decision-making mode from "passive emergency response" to "proactive early warning and resilience-based maintenance planning," effectively avoiding unplanned downtime. Furthermore, by constructing a risk assessment model that integrates physical information neural networks, the system dynamically integrates the basic degradation degree (medium quality parameters), system immunity (degradation immunity index output by the physical constraint branch network), and local activity (acoustic emission activity index). This model deeply embeds the calculation methods of degradation immunity indices for different degradation mechanisms into the learning process of the neural network through a "physical constraint loss function." This allows the evaluation results to possess both data-driven adaptive and complex pattern recognition capabilities, while strictly ensuring consistency with physical laws. This significantly improves the model's interpretability and generalization reliability under unknown operating conditions. Finally, by combining the risk index and resilience value for comprehensive judgment, it generates graded resilience maintenance strategies (such as optimized operation, adding agents, and early maintenance) corresponding to the degradation mechanism, rather than simply issuing alarms. This significantly improves the refinement and intelligence of risk management, which is of great significance for ensuring safety and reducing costs and increasing efficiency.

[0116] This invention also provides a data-driven intelligent early warning system for the degradation risk of organic heat transfer fluids, used to execute the aforementioned data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids. Figure 3 This is a schematic diagram of the structure of the data-driven intelligent early warning system for the degradation risk of organic heat carriers provided in this embodiment of the invention. See also... Figure 3 The system includes the following modules:

[0117] The data acquisition module is used to acquire multimodal monitoring data of the organic heat transfer fluid, including operating parameters, medium quality parameters, and acoustic emission signals.

[0118] The feature extraction module is used to preprocess and extract features from multimodal monitoring data to obtain entropy yield and acoustic emission activity index;

[0119] The degradation mechanism classification module is used to construct feature vectors based on entropy yield, acoustic emission activity index and medium quality parameters, input them into a pre-trained mechanism classification neural network model, and obtain the determination probability of the organic heat carrier being in each preset degradation mechanism.

[0120] The instantaneous risk index prediction module is used to select the corresponding target branch neural network from the pre-trained physical constraint branch neural network based on the preset degradation mechanism with the highest judgment probability, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value.

[0121] The degradation toughness value calculation module is used to predict the time required for each indicator to reach the safety threshold based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid. The minimum value among the times required for each indicator to reach the safety threshold is used as the predicted failure time, and the degradation toughness value is calculated in combination with the planned maintenance time.

[0122] The early warning module is used to determine the comprehensive risk level and generate corresponding resilience maintenance strategies based on the instantaneous risk index and the deterioration resilience value.

[0123] The execution process of the system part of this application embodiment is the same as that of the method part of the embodiment described above, and will not be repeated here.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A data-driven intelligent early warning method for the deterioration risk of organic heat transfer fluids, characterized in that, The method includes the following steps: S1. Acquire multimodal monitoring data of the organic heat transfer fluid, wherein the multimodal monitoring data includes operating parameters, medium quality parameters, and acoustic emission signals; S2. Preprocess and extract features from the multimodal monitoring data to obtain the entropy yield and acoustic emission activity index; S3. Construct a feature vector based on the entropy yield, the acoustic emission activity index and the medium quality parameters, input it into a pre-trained mechanism classification neural network model, and obtain the probability of the organic heat carrier being in each preset degradation mechanism. S4. Based on the preset degradation mechanism with the highest determination probability, select the corresponding target branch neural network from the pre-trained physical constraint branch neural network, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value. S5. Based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid, predict the time required for each indicator to reach the safety threshold, take the minimum value among the times required for each indicator to reach the safety threshold as the predicted failure time, and calculate the deterioration toughness value in combination with the planned maintenance time. S6. Based on the instantaneous risk index prediction value and the deterioration resilience value, determine the comprehensive risk level and generate a corresponding resilience maintenance strategy.

2. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 1, characterized in that, In S2, the formula for calculating the entropy production rate is as follows: ； in, This represents the entropy yield of the organic heat transfer fluid. This represents the total thermal power of the input system. This represents the effective thermal power output by the system. This indicates the average temperature of the furnace. Indicates ambient temperature.

3. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 1, characterized in that, In step S2, the formula for calculating the acoustic emission activity index is as follows: ； in, This indicates the acoustic emission activity index of the organic heat transfer fluid. This represents the root mean square value of the measured acoustic emission signal. This represents the critical root-mean-square value of the acoustic emission signal. This represents the root mean square value of the background noise of the system under normal conditions.

4. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 1, characterized in that, In S3, the preset degradation mechanism includes at least the oxidation degradation mechanism and the thermal cracking-coking degradation mechanism.

5. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 4, characterized in that, In step S4, based on the preset degradation mechanism with the highest determination probability, a corresponding target branch neural network is selected from the pre-trained physical constraint branch neural network, including: When the probability of determining the oxidative degradation mechanism is at its maximum, the target branch neural network is determined to be the first physical constraint branch neural network; the first physical constraint loss function used by the first physical constraint branch neural network during training is constructed based on the calculation formula of the first degradation immunity index; When the probability of determining the thermal cracking-coking degradation mechanism is at its maximum, the target branch neural network is determined to be the second physical constraint branch neural network; the second physical constraint loss function used by the second physical constraint branch neural network during training is constructed based on the calculation formula of the second degradation immunity index.

6. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 5, characterized in that, The formula for calculating the first physical constraint loss function is as follows: ； ； ； in, Represents the first physical constraint loss function. This indicates task loss due to oxidative degradation mechanism. This indicates the physical constraint loss in the oxidative degradation mechanism. This represents the hyperparameter of the oxidative degradation mechanism, where i represents the i-th sample and N represents the total number of samples. This represents the predicted value of the immune index. This indicates the true value of the immune index representing the oxidative degradation mechanism. This represents the predicted value of the instantaneous risk index. This indicates the true value of the instantaneous risk index for the oxidative degradation mechanism; The formula for calculating the true value of the immune index of the oxidative degradation mechanism is as follows: ； in, , , These represent the acid value weighting coefficient, entropy yield weighting coefficient, and nitrogen sealing weighting coefficient, respectively, where A represents the current acid value. Indicates the acid value safety threshold. This represents the entropy yield of the organic heat transfer fluid. The critical threshold representing the entropy productivity. This indicates the effectiveness of the nitrogen blanketing system.

7. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 5, characterized in that, The formula for calculating the second physical constraint loss function is as follows: ； ； ； in, This represents the second physical constraint loss function. This indicates the task loss due to the thermal cracking-coking degradation mechanism. This indicates the physical constraint loss in the thermal cracking-coking degradation mechanism. This represents the hyperparameter of the thermal cracking-coking degradation mechanism, where i represents the i-th sample and N represents the total number of samples. This represents the predicted value of the immune index. This indicates the true value of the immune index for the thermal decomposition-coking degradation mechanism. This represents the predicted value of the instantaneous risk index. This represents the true value of the instantaneous risk index for the thermal cracking-coking degradation mechanism; The formula for calculating the true value of the immune index of the thermal decomposition-coking degradation mechanism is as follows: ； in, , , , These represent the viscosity weighting coefficient, residual carbon weighting coefficient, acoustic emission activity index weighting coefficient, and entropy yield weighting coefficient, respectively. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value. Indicates acoustic emission activity index, This represents the entropy yield of the organic heat transfer fluid. The critical threshold representing the entropy production rate.

8. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to any one of claims 6 or 7, characterized in that, The formula for calculating the true value of the instantaneous risk index is as follows: ； ； in, R represents the true value of the instantaneous risk index for degradation mechanism of type X, and R represents the base risk index. , , These represent the acid value weighting coefficient, viscosity weighting coefficient, and residual carbon weighting coefficient, respectively, with A representing the current acid value. Indicates the acid value safety threshold. Indicates the current kinematic viscosity. Indicates the safe threshold for kinematic viscosity. This indicates the current residual carbon value. This indicates the safe threshold for residual carbon value. This represents the true value of the immune index for the degradation mechanism of type X. This represents the weighting coefficient of acoustic emission activities in the risk. This indicates the acoustic emission activity index.

9. The data-driven intelligent early warning method for the degradation risk of organic heat transfer fluids according to claim 1, characterized in that, In step S5, the formula for calculating the deterioration toughness value is as follows: ； in, Indicates the deterioration toughness value. Indicates the predicted failure time. Indicates the planned maintenance time.

10. A data-driven intelligent early warning system for the deterioration risk of organic heat transfer fluids, used to execute the data-driven intelligent early warning method for the deterioration risk of organic heat transfer fluids as described in any one of claims 1-9, characterized in that, The system includes the following modules: The data acquisition module is used to acquire multimodal monitoring data of the organic heat transfer fluid, including operating parameters, medium quality parameters, and acoustic emission signals. The feature extraction module is used to preprocess and extract features from the multimodal monitoring data to obtain the entropy yield and acoustic emission activity index. The degradation mechanism classification module is used to construct a feature vector based on the entropy yield, the acoustic emission activity index and the medium quality parameters, input it into a pre-trained mechanism classification neural network model, and obtain the determination probability of the organic heat carrier being in each preset degradation mechanism. The instantaneous risk index calculation module is used to select the corresponding target branch neural network from the pre-trained physical constraint branch neural network according to the preset degradation mechanism with the highest judgment probability, input the feature vector into the target branch neural network, and obtain the instantaneous risk index prediction value. The degradation toughness value calculation module is used to predict the time required for each indicator to reach the safety threshold based on the historical medium quality parameters and historical entropy yield of the organic heat transfer fluid. The minimum value among the times required for each indicator to reach the safety threshold is used as the predicted failure time, and the degradation toughness value is calculated in combination with the planned maintenance time. The early warning module is used to determine the comprehensive risk level and generate a corresponding resilience maintenance strategy based on the instantaneous risk index prediction value and the deterioration resilience value.

Images