Prediction method for insulation residual life of low-voltage electromagnetic coil driven by mechanism and data fusion

Abstract

The invention discloses a mechanism and data fusion driven low-voltage electromagnetic coil insulation residual life prediction method, and belongs to the technical field of power equipment state monitoring and predictive maintenance. The method comprises the following steps: firstly, constructing a creep finite element simulation model of the turn-to-turn insulating coating of the electromagnetic coil, and determining a creep quantity as an insulating health index with a clear physical meaning; then establishing a mapping relationship between electrical parameters such as high-frequency impedance and the like and creep variables, and realizing accurate trend prediction of the electrical parameters and the corresponding creep variables by adopting a lightweight deep learning model (Windows Mixer); and finally, fusing a failure threshold value determined based on a creep mechanism with a data-driven prediction result so as to improve generalization and accuracy of a prediction model. According to the invention, an intelligent prediction tool with high precision and low resource consumption can be provided for operation and maintenance of the electromagnetic coil.

Description

Technical Field

[0001] The present invention belongs to the technical field of power equipment health management, and particularly relates to a method for predicting the remaining insulation life of low-voltage electromagnetic coils by integrating electromagnetic mechanics mechanism modeling and lightweight deep learning algorithms, which is particularly applicable to the condition monitoring and preventive maintenance of low-voltage electromagnetic devices such as relays and contactors. Background Art

[0002] Low-voltage electromagnetic coils are the core components for electromagnetic conversion in power equipment, and the degradation of their insulation performance is one of the key factors affecting the safe and reliable operation of equipment. Traditional insulation degradation assessment methods include two categories: mechanism modeling and data-driven methods. Among them, the finite element simulation method based on creep mechanism relies on the physical structure of the equipment and mechanical laws for simulation, and has high generalization. However, when dealing with complex systems, problems such as simplification and inaccurate assumptions often occur in the simulation model. For example, when considering the nonlinear characteristics of materials or the actual operating environment of the equipment, significant deviations may occur between the simulation results and the actual working conditions. In addition, pure simulation models usually have difficulty in fully capturing the actual operating state of the equipment, and there are certain prediction limitations.

[0003] On the other hand, data-driven deep learning models have strong feature mining capabilities by performing high-precision fitting on historical data. However, in complex nonlinear systems, deep learning models are prone to falling into local optimal solutions, and usually rely on a large amount of training data, have strict requirements on data quality and distribution, and are prone to overfitting problems. In addition, the existing methods for determining the failure threshold of insulation systems are mainly based on the statistical correlation of accelerated aging test data, which essentially belongs to a black-box empirical method, lacks a clear physical mechanism explanation, and has poor working condition generalization.

[0004] Therefore, there is an urgent need in the prior art for a fusion prediction method that can not only maintain the generalization ability of the mechanism model, but also make full use of the prediction ability of the data-driven model and provide physical interpretability. Summary of the Invention

[0005] The purpose of the present invention is to address the deficiencies of the existing low-voltage electromagnetic coil insulation life prediction methods, such as weak physical interpretability, high complexity of the prediction model, and difficulty in clearly defining the failure threshold, and propose a mechanism and data fusion-driven method for predicting the remaining insulation life of low-voltage electromagnetic coils. This method realizes the prediction of the remaining insulation life with high generalization, high precision, and clear physical interpretation by integrating mechanism modeling and deep learning technology, effectively solves the contradiction between the generalization ability and prediction accuracy in the prior art, and improves the reliability and intelligent level of the operation and maintenance of electromagnetic coils.

[0006] The technical solution adopted by the present invention to achieve the above purpose is:

[0007] A method for predicting the remaining insulation life of low-voltage electromagnetic coils driven by mechanism and data fusion, comprising the following steps:

[0008] 1) Establish a three-dimensional geometric model according to the actual structural dimensions of the low-voltage electromagnetic coil to be measured;

[0009] 2) Based on the three-dimensional geometric model of the low-voltage electromagnetic coil, and based on the three-stage theory of creep, adopt the finite element analysis method to establish a creep simulation model of the inter-turn insulation coating of the electromagnetic coil;

[0010] 3) Apply high-frequency voltage signals with different frequencies to the electromagnetic coil to be measured, collect the high-frequency impedance data of the coil at different aging stages, and construct a high-frequency electrical parameter data set;

[0011] 4) Construct a mapping relationship between high-frequency electrical parameters and the creep amount of the insulation coating;

[0012] 5) Utilize the mapping relationship to construct and train a data-driven prediction model based on the WindowMixer model;

[0013] 6) Use the creep simulation model and the data-driven prediction model to perform a fusion prediction on the remaining insulation life.

[0014] The step 4) includes the following steps:

[0015] 4.1) Based on the collected high-frequency impedance data of the coil, identify and extract the resonant frequency when the reactance is zero, and use the inherent physical relationship between the resonant frequency and the self-inductance and capacitance of the coil to calculate the corresponding coil capacitance value;

[0016] 4.2) Convert the coil capacitance value into the inter-turn capacitance representing the inter-turn insulation state through an empirical formula;

[0017] 4.3) Use an optimization algorithm to inversely deduce the change in the outer diameter of the wire including the insulation coating according to the inter-turn capacitance, and then establish a quantitative mapping relationship model between high-frequency electrical parameters and the creep amount of the coil insulation coating.

[0018] The step 5) includes the following steps:

[0019] 5.1) Decompose the creep amount data of the coil insulation coating in the form of a normalized time series into a trend component and a seasonal component through the average pooling method with a fixed window; wherein, the trend component represents the long-term and coarse-grained information in the sequence, and the seasonal component represents the short-term and fine-grained periodic information;

[0020] 5.2) Predict the trend component through a fully connected layer;

[0021] 5.3) Predict the seasonal component;

[0022] 5.4) Add the predicted trend component and the seasonal component to obtain the completed predicted value.

[0023] Step 5.3) includes the following steps:

[0024] 5.3.1) Create a window with a fixed length centered at each time point in the time series, project each window into a higher-dimensional space using a linear transformation, and generate corresponding window tokens.

[0025] 5.3.2) Pass the window tokens through the in-window mixer and the inter-window mixer in sequence to capture the local relationships between the data inside the window and the long-term global dependencies between different windows, and flatten the window relationships into a one-dimensional vector.

[0026] 5.3.3) Use a fully connected layer to process the one-dimensional vector to generate the prediction result of the seasonal component.

[0027] Both the in-window mixer and the inter-window mixer adopt a full MLP architecture, including a Dropout layer, a fully connected layer, and a non-linear activation function, and output them to a two-dimensional layer for normalization processing.

[0028] Step 6) includes the following steps:

[0029] 6.1) Use the creep simulation model to determine the failure threshold, and insulation failure occurs when the creep amount of the insulation coating reaches the failure threshold.

[0030] 6.2) Adopt an adaptive weight fusion strategy with threshold constraint, use the failure threshold predicted by the creep simulation model as a constraint condition, and dynamically adjust the creep amount prediction result of the data-driven prediction model to make it closer to the true value.

[0031] 6.3) Use the data-driven prediction model with adjusted weights to predict the creep amount trend of the electromagnetic coil insulation coating, and compare it with the failure threshold. When the creep amount reaches the threshold, the corresponding time at this time is the remaining insulation life of the coil.

[0032] The specific method of using the creep simulation model to determine the failure threshold is:

[0033] Take the inflection point where the steady-state creep changes to accelerating creep in the three stages of creep as the failure threshold.

[0034] The adaptive weight fusion strategy with threshold constraint is specifically:

[0035] output = a × α + b × (1 - α)

[0036] where output represents the result output of the fusion prediction method, α represents the weight, a represents the result of the creep simulation model, and b represents the result of the data-driven prediction model.

[0037] The present invention has the following beneficial effects and advantages:

[0038] 1. The present invention proposes a novel insulation remaining life prediction method that combines the dual advantages of mechanism and data-driven approaches, integrating the generalization of physical mechanisms and the high prediction accuracy of deep learning models.

[0039] 2. The present invention defines the insulation layer failure threshold based on the creep mechanism, enhancing the physical interpretability and generalization of the model and overcoming the limitation of the traditional empirical threshold that is difficult to be universal.

[0040] 3. The present invention adopts the WindowMixer lightweight prediction model, significantly reducing the computational resources required for prediction and meeting the real-time requirements of industrial sites.

[0041] 4. Through the integration of mechanism and data-driven approaches, the present invention effectively improves the accuracy and reliability of the coil insulation life prediction, providing a solid theoretical and technical guarantee for the operation and maintenance of power equipment. Brief Description of the Drawings

[0042] Figure 1 is a schematic diagram of the overall process of a method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by the fusion of mechanism and data according to the present invention;

[0043] Figure 2 is a schematic diagram of the geometric modeling and finite element mesh division of an electromagnetic coil;

[0044] Figure 3 is a schematic diagram of the stress distribution simulation of the inter-turn insulation coating of an electromagnetic coil during the creep process;

[0045] Figure 4 is a schematic diagram of the windowmixer model structure used in the present invention;

[0046] Figure 5 is a schematic diagram of the method structure based on the fusion of a mechanism model and a data-driven model;

[0047] Figure 6 is a schematic diagram of the implementation effect of the fusion model for predicting the remaining life of the insulation layer of an electromagnetic coil. Detailed Embodiment

[0048] The following further describes the present invention in detail with reference to the drawings and embodiments.

[0049] As Figure 1 shown, a method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by the fusion of mechanism and data includes the following steps:

[0050] (1) Establish a geometric model of the electromagnetic coil to be measured:

[0051] According to the actual structural dimensions of the low-voltage electromagnetic coil to be measured, including parameters such as the coil bobbin diameter, bobbin height, wire outer diameter, wire inner diameter, and wire insulation coating thickness, an accurate three-dimensional geometric model is established as the basis for subsequent creep simulation analysis.

[0052] (2) Establish a creep simulation model for the inter-turn insulation coating of the electromagnetic coil:

[0053] Based on the three-dimensional geometric model established in step (1), and based on the three-stage creep theory (primary creep, steady-state creep, and accelerating creep), a creep simulation model for the inter-turn insulation coating of the electromagnetic coil is established using the finite element analysis method;

[0054] Specifically, the simulation model uses an improved time-hardening creep model. In the finite element analysis, the actual temperature, material properties, and initial stress distribution during the operation of the coil are input to obtain the creep simulation data of the insulation coating thickness changing with time, and its creep characteristics are analyzed to clarify the creep amount as a physical health index for insulation performance degradation.

[0055] (3) Injection of high-frequency electrical signals and acquisition of response signals:

[0056] A high-frequency signal generator is used to apply high-frequency voltage signals with different frequencies to the electromagnetic coil to be measured. The coil generates a response current signal under the excitation of this signal, and an impedance analysis device is used to collect the high-frequency response data of the coil at different aging stages, so as to obtain a high-frequency electrical parameter data set containing the characteristics of the coil impedance, reactance, and capacitance.

[0057] (4) Establish the mapping relationship between high-frequency electrical parameters and the creep amount of the insulation coating:

[0058] Based on the collected high-frequency impedance data of the coil, the resonance frequency when the reactance is zero is identified and extracted, and the corresponding coil capacitance value is calculated using the inherent physical relationship between this resonance frequency and the self-inductance and capacitance of the coil. Subsequently, through an empirical formula, the calculated coil capacitance value is further converted into an inter-turn capacitance that can characterize the inter-turn insulation state. Using an optimization algorithm, the change in the outer diameter of the wire including the insulation coating is deduced from the inter-turn capacitance, so as to finally realize the conversion from measurable high-frequency electrical parameters to the creep amount of the coil insulation coating, and establish a clear and effective quantitative mapping relationship model.

[0059] (5) Data-driven creep amount prediction based on the WindowMixer model:

[0060] This step aims to utilize the mapping relationship between high-frequency electrical parameters and the creep variable of the insulation coating established in step (4), convert the historical high-frequency electrical parameter data into corresponding creep variable data as input, train a data-driven prediction model based on the WindowMixer architecture, and predict the creep variable trend in the future period. The specific process of the data-driven prediction model is described as follows:

[0061] Specifically, it includes the following sub-steps:

[0062] ① Time series decomposition:

[0063] The normalized time series is decomposed into a trend component and a seasonal component through the average pooling method with a fixed window; the trend component represents the long-term and coarse-grained information in the sequence, capturing the overall change trend, while the seasonal component represents the short-term and fine-grained periodic information;

[0064] ② Trend prediction:

[0065] The trend component is directly predicted through a fully connected layer (FC layer) to maintain the overall structural characteristics of the trend information;

[0066] ③ Window embedding of the seasonal component:

[0067] The seasonal component is further characterized using a window embedding module;

[0068] Specifically, a window with a fixed length (such as a length of 17 time points, window radius w = 8) is created centered on each time point, and each window is projected into a higher-dimensional space using a linear transformation to generate corresponding window tokens, realizing the local pattern representation of the seasonal component;

[0069] ④ Intra-Window-Mixer and Inter-Window-Mixer:

[0070] The window tokens first pass through the Intra-Window-Mixer to capture the local relationships between the data within the window;

[0071] Then, they pass through the Inter-Window-Mixer to capture the long-term global dependencies between different windows;

[0072] Both mixers adopt a fully MLP (Multi-Layer Perceptron) architecture, including a Dropout layer, a Fully Connected Layer, and a non-linear Activation Function, and are combined with two-dimensional Layer Normalization (LayerNorm2D) to improve the stability of the model;

[0073] ⑤ Marker flattening and seasonal component prediction:

[0074] The window markers processed by the in-window and inter-window mixers are flattened into a one-dimensional vector and fed into another fully connected layer to generate the prediction results of the seasonal components;

[0075] ⑥ Trend and seasonal component fusion:

[0076] Add the trend component and the seasonal component predicted in the previous steps to achieve the fusion of the trend and seasonal patterns and generate the complete predicted value;

[0077] ⑦ Final prediction output:

[0078] The final creep variable prediction results obtained through the above steps can be directly used for the next step of the fusion of the mechanism model and the data-driven model, and then achieve the accurate prediction of the remaining insulation life of the low-voltage electromagnetic coil.

[0079] Compared with traditional deep learning models, the WindowMixer model has a smaller parameter scale and lower computational complexity, thus significantly improving the prediction speed and accuracy, and is more suitable for actual industrial application scenarios.

[0080] (6) Fusion prediction of the mechanism model and the data-driven model:

[0081] Use the creep simulation model (step 2) to determine the failure threshold, which is clearly defined as insulation failure when the creep variable of the insulation coating reaches the determined failure threshold;

[0082] outp = mechanism model result (a) × α + data-driven model result (b) × (1 - α) #(1)

[0083] Adopt an adaptive weight fusion strategy with threshold constraints, as shown in equation (1), use the failure threshold predicted by the mechanism model as a constraint condition, dynamically adjust the creep variable prediction results of the data-driven model in step (5), form a fusion model with the advantages of both mechanism and data-driven, realize the correction and optimization of the prediction results, and ensure the generalization and physical interpretability of the model.

[0084] (7) Prediction of the remaining insulation life and operation and maintenance decision-making based on the fusion model:

[0085] Predict the creep variable trend of the electromagnetic coil insulation coating through a mechanism and data fusion prediction model, compare it with the failure threshold, and determine the remaining insulation life of the coil;

[0086] Based on the prediction results, formulate a targeted maintenance strategy or replacement plan, and take maintenance measures in advance before the creep variable of the insulation coating reaches the failure threshold, so as to reduce the risk of sudden failures and achieve low-cost and high-efficiency intelligent operation and maintenance of equipment.

[0087] Furthermore, to improve the prediction accuracy and generalization, the improved time-hardening creep model in step (2) is preferably a model that considers the nonlinear creep characteristics of the material and temperature sensitivity, and the effectiveness of the model is verified through experimental simulations and actual accelerated aging experiments.

[0088] Furthermore, in step (5), the optimizer used for training the WindowMixer model is Adam, the initial learning rate is 0.001, and a dynamic learning rate adjustment strategy is adopted to quickly achieve model convergence and reduce the risk of overfitting.

[0089] Furthermore, the fusion process in step (6) preferably adopts a method of dynamically adjusting the adaptive weight based on the prediction error. When the prediction error of the data-driven model increases, the weight of the mechanism model is increased to effectively avoid the overfitting phenomenon of the data-driven model.

[0090] Embodiment

[0091] Taking the accelerated aging test at 260°C as an example, the technical solution of the present invention will be described in detail below:

[0092] Step 1: Geometric modeling and finite element simulation

[0093] Taking a certain type of low-voltage electromagnetic coil as the object to be measured, the geometric structure parameters of the coil are as follows:

[0094]

[0095] Establish the coil geometric model with the above dimensions and structures, as Figure 2 shown. Using ANSYS finite element analysis software, the material parameters are set as follows:

[0096]

[0097] Using the improved time-hardening creep model for simulation, setting the test temperature to 260°C and the aging duration to 80,000 seconds. The simulation obtains the curve of the change trend of the coil insulation coating thickness over time, as Figure 3 shown.

[0098] Step 2: Electrical parameter measurement and establishment of the mapping relationship between creep variables

[0099] Under the condition of constant temperature at 260 °C, an accelerated aging test is carried out on the coil to be measured. Using a high-frequency signal generator and an impedance analyzer, electrical parameters such as the impedance and capacitance of the coil in the high-frequency band are measured in real time. The test sampling frequency is once per minute, and continuous measurement is carried out until the end of the test.

[0100] Based on the creep variable data and high-frequency electrical parameters obtained from the simulation, a mapping relationship model between high-frequency electrical parameters and creep variables is established by using the multiple nonlinear regression analysis method.

[0101] Step 3: Data-driven model prediction (WindowMixer model)

[0102] The WindowMixer model is trained based on historical electrical data. The structure of the WindowMixer model is as Figure 4 shown, including instance normalization, time window embedding, Intra-Window Mixer and Inter-Window Mixer modules, to efficiently capture local dynamics and global dependencies in time-series data, predict the change trend of electrical parameters at future moments, and convert them into corresponding predicted values of the creep variable of the insulating layer through the mapping relationship model.

[0103] Model training parameter settings: the window size is 128, the prediction length is 10 sampling points, the Adam optimizer is selected as the optimizer, and the initial learning rate is set to 0.001.

[0104] Step 4: Mechanism and data-driven fusion prediction

[0105] Based on the creep mechanism model, the failure threshold is determined to be a creep variable of 1.96e-5 mm. The threshold-constrained adaptive weight fusion method is used to fuse and correct the threshold with the prediction results of the data-driven model to obtain the final prediction result of the creep variable, and the corresponding remaining insulation life is deduced.

[0106] Step 5: Verification of prediction results and performance analysis

[0107] The high-frequency impedance change data obtained from the actual accelerated aging test is used for verification. The prediction results are as Figure 5As shown in Table 1, it can be seen that the proposed method based on the fusion of mechanism and data-driven has improved the prediction accuracy by 20.18% compared with the deep learning-based model (WindowMixer). The reasons for the improvement in accuracy are analyzed as follows: Although the deep learning-based prediction model has strong following ability and can better continue the trend in historical data, however, due to the fact that the true creep curve of the coil turn-to-turn insulation layer shows a three-stage change characteristic, especially in the last stage, the slope of the creep curve will mutate, and it is difficult for the deep learning model to capture this important change from the early aging data. The proposed method provides the inflection point information of the second and third stages of creep degradation to the deep learning model by integrating the mechanism model of creep degradation. Therefore, the prediction accuracy is effectively improved.

[0108] Table 1 Comparison of the life prediction accuracy of the proposed prediction model under different prediction length conditions

[0109]

[0110] According to the results of the fusion prediction model, predict the remaining life of the electromagnetic coil to be tested under the constant temperature condition of 260 °C. Make a reasonable operation and maintenance decision based on the prediction results, and implement maintenance and replacement in advance before the insulation layer reaches the creep threshold to effectively avoid the risk of sudden failure.

[0111] In summary, the present invention significantly improves the performance of predicting the remaining life of the insulation of low-voltage electromagnetic coils through clear mechanism analysis and innovative model fusion strategies, and has strong theoretical and practical application values.

[0112] The above embodiments are only used to illustrate the technical solutions of the present invention and do not constitute a limitation on the protection scope of the present invention. Any equivalent transformation or improvement made by those skilled in the art on the basis of the technical solutions of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for predicting the remaining insulation life of low-voltage electromagnetic coils driven by mechanism and data fusion, characterized in that It includes the following steps: 1) Establish a three-dimensional geometric model according to the actual structural dimensions of the low-voltage electromagnetic coil to be measured; 2) Based on the three-dimensional geometric model of the low-voltage electromagnetic coil, and based on the three-stage theory of creep, adopt the finite element analysis method to establish a creep simulation model of the inter-turn insulation coating of the electromagnetic coil; 3) Apply high-frequency voltage signals with different frequencies to the electromagnetic coil to be measured, collect the high-frequency impedance data of the coil at different aging stages, and construct a high-frequency electrical parameter data set; 4) Construct a mapping relationship between high-frequency electrical parameters and the creep amount of the insulation coating; 5) Use the mapping relationship to construct and train a data-driven prediction model based on the WindowMixer model; 6) Use the creep simulation model and the data-driven prediction model to perform a fusion prediction on the remaining insulation life.

2. A method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 1, characterized in that The step 4) includes the following steps: 4.1) Based on the collected high-frequency impedance data of the coil, identify and extract the resonance frequency when the reactance is zero, and use the inherent physical relationship between the resonance frequency and the self-inductance and capacitance of the coil to calculate the corresponding coil capacitance value; 4.2) Convert the coil capacitance value into the inter-turn capacitance characterizing the inter-turn insulation state through an empirical formula; 4.3) Use an optimization algorithm to inversely deduce the change in the outer diameter of the wire including the insulation coating according to the inter-turn capacitance, and then establish a quantitative mapping relationship model between high-frequency electrical parameters and the creep amount of the coil insulation coating.

3. A method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 1, characterized in that The step 5) includes the following steps: 5.1) Decompose the creep amount data of the coil insulation coating in the form of a normalized time series into a trend component and a seasonal component through the average pooling method with a fixed window; wherein, the trend component represents the long-term and coarse-grained information in the sequence, and the seasonal component represents the short-term and fine-grained periodic information; 5.2) Predict the trend component through a fully connected layer; 5.3) Predict the seasonal component; 5.4) Add the predicted trend component and the seasonal component to obtain the complete predicted value.

4. A method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 3, characterized in that The step 5.3) includes the following steps: 5.3.1) Create a window with a fixed length centered on each time point in the time series, and project each window into a higher-dimensional space using a linear transformation to generate corresponding window tokens; 5.3.2) Pass the window tokens through the in-window mixer and the between-window mixer in sequence to capture the local relationship between the data inside the window and the long-term global dependence between different windows in sequence, and flatten the window relationship into a one-dimensional vector; 5.3.3) Use a fully connected layer to process the one-dimensional vector to generate the prediction result of the seasonal component.

5. A method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 4, characterized in that, Both the in-window mixer and the between-window mixer adopt a full MLP architecture, including a Dropout layer, a fully connected layer, and a non-linear activation function, and output it to a two-dimensional layer for normalization processing.

6. A prediction method for the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 1, characterized in that, The step 6) includes the following steps: 6.1) Use the creep simulation model to determine the failure threshold, and insulation failure occurs when the creep amount of the insulation coating reaches the failure threshold; 6.2) Adopt an adaptive weight fusion strategy with threshold constraint, use the failure threshold predicted by the creep simulation model as a constraint condition, and dynamically adjust the creep amount prediction result of the data-driven prediction model to make it closer to the true value; 6.3) Use the data-driven prediction model after weight adjustment to predict the creep variable trend of the electromagnetic coil insulation coating, and compare it with the failure threshold. When the creep variable reaches the threshold, the corresponding time at this time is the remaining insulation life of the coil.

7. A prediction method for the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 6, characterized in that The specific method for determining the failure threshold using the creep simulation model is as follows: Take the inflection point where the steady-state creep in the three stages of creep changes to accelerated creep as the failure threshold.

8. A method for predicting the remaining insulation life of a low-voltage electromagnetic coil driven by mechanism and data fusion according to claim 6, characterized in that, The adaptive weight fusion strategy with threshold constraint is specifically as follows: output = a×α + b×(1 - α) Among them, output represents the result output of the fusion prediction method, α represents the weight, a represents the result of the creep simulation model, and b represents the result of the data-driven prediction model.

Images