An intelligent prediction method and system for the aging state of an aluminum alloy flexible cable

Abstract

The application discloses an intelligent prediction method and system for the aging state of an aluminum alloy flexible cable, and relates to the technical field of cables.The method comprises the following steps: obtaining multi-source operation data and prior material performance data of the current life cycle of a target cable; establishing a mechanical characteristic sequence, a load characteristic sequence and a condition vector of the target cable; according to the mechanical characteristic sequence, the load characteristic sequence and the condition vector, combining a pre-trained double-channel encoding-regression prediction model, performing segmented aging intensity prediction on the target cable, obtaining an aging intensity sequence, and then predicting the aging state information of the target cable.The technical problem of limited prediction accuracy and generalization ability caused by the complexity of offline detection operation, the single online monitoring parameter and the insufficient consideration of nonlinear time-varying effect in the prior art is solved.

Description

Technical Field

[0001] This application relates to the field of cable technology, specifically to an intelligent prediction method and system for the aging state of aluminum alloy flexible cables. Background Technology

[0002] Aluminum alloy flexible cables are widely used in many fields such as building power distribution, new energy, rail transportation, and industrial automation due to their advantages of light weight, high strength, corrosion resistance, good flexibility, and cost-effectiveness. However, during long-term operation, aluminum alloy flexible cables are inevitably affected by the combined effects of environmental factors, continuous electrical load, and mechanical stress, leading to the degradation of the mechanical properties of the conductor material, increased dielectric loss of the insulation layer, and damage to the structural integrity, i.e., aging. Cable aging is a gradual and complex physicochemical process. If the aging state cannot be accurately and timely assessed and predicted, it may lead to insulation breakdown, short circuit faults, and even serious safety accidents such as fires and large-scale power outages, causing huge losses to social production and life.

[0003] In existing technologies, methods for assessing and predicting the aging state of cables mainly include traditional offline testing methods and online monitoring methods based on single parameters. Traditional offline testing methods, such as taking cable samples for tensile testing, dielectric loss testing, and thermogravimetric analysis, can obtain relatively accurate aging performance data, but require power outages for sampling, are complex to operate, and are costly. Furthermore, they cannot achieve real-time monitoring of the status of cables in operation, making it difficult to reflect the overall aging distribution and dynamic trends of the cable. Online monitoring methods based on single parameters, while achieving online monitoring, suffer from limitations because cable aging is the result of multiple coupled factors. A single parameter often only reflects one aspect of the aging process, making it difficult to comprehensively and accurately characterize the overall aging degree and remaining lifespan of the cable, and prone to misjudgment or omission.

[0004] In addition, some prediction models rely heavily on empirical formulas or simple statistical models, which do not adequately consider the nonlinearity, time-varying nature, and multi-physics coupling effects in the cable aging process. This results in limited prediction accuracy and generalization ability, making it difficult to meet the actual needs of intelligent prediction of the aging state of aluminum alloy flexible cables under complex working conditions. Summary of the Invention

[0005] This application provides an intelligent prediction method and system for the aging state of aluminum alloy flexible cables, which solves the technical problems of existing technologies for predicting the aging state of aluminum alloy flexible cables, such as complex offline detection operations, single online monitoring parameters, and insufficient consideration of nonlinear time-varying effects, resulting in limited prediction accuracy and generalization ability.

[0006] The technical solution to the above-mentioned technical problems in this application is as follows: In a first aspect, this application provides an intelligent prediction method for the aging state of aluminum alloy flexible cables, the method comprising: Acquire multi-source operational data of the target cable during its current lifecycle, and simultaneously acquire prior material performance data of the target cable; By combining the multi-source operational data with the prior material performance data, a mechanical characteristic sequence, a load characteristic sequence, and a condition vector for the target cable are established. Based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, the aging intensity of the target cable is predicted in segments to obtain the aging intensity sequence. Based on the aging intensity sequence, the aging status information of the target cable is predicted.

[0007] Secondly, this application provides an intelligent prediction system for the aging state of aluminum alloy flexible cables, comprising: The multi-source data acquisition module is used to acquire multi-source operational data of the target cable during its current life cycle, and simultaneously acquire prior material performance data of the target cable. The feature sequence establishment module is used to combine the multi-source operating data and the prior material performance data to establish the mechanical feature sequence, load feature sequence and condition vector of the target cable; The prediction model training module is used to predict the segmented aging intensity of the target cable based on the mechanical feature sequence, the load feature sequence and the condition vector, combined with a pre-trained dual-channel encoding-regression prediction model, and to obtain the aging intensity sequence. The aging information prediction module is used to predict the aging status information of the target cable based on the aging intensity sequence.

[0008] This application provides one or more technical solutions, which have at least the following technical effects or advantages: This application provides an intelligent prediction method and system for the aging state of aluminum alloy flexible cables. First, it acquires multi-source operational data and prior material performance data of the target cable's current lifecycle, laying the data foundation for establishing feature sequences and aging prediction. Second, combining the multi-source operational data and prior material performance data, it constructs mechanical feature sequences, load feature sequences, and conditional vectors that reflect the cable's operational state and material properties. Third, it employs a pre-trained dual-channel encoding-regression prediction model to predict the segmented aging intensity of the target cable. This model effectively handles the differences in mechanical and load characteristics over time and enhances its adaptability to cables with different material properties through conditional modulation. Finally, based on the obtained aging intensity sequences, it predicts the aging state information of the target cable, including predicting aging life and determining the aging level. Through multi-source data fusion, dual-channel feature encoding, and multi-dimensional aging information output, it characterizes the aging process of the aluminum alloy flexible cable.

[0009] Through the above technical solutions, this application effectively improves the accuracy and generalization ability of predicting cable aging status, overcomes the limitations of traditional offline testing and single-parameter online monitoring, and provides strong technical support for the safe operation and maintenance of aluminum alloy flexible cables. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating an intelligent prediction method for the aging state of an aluminum alloy flexible cable provided in an embodiment of this application. Figure 2 This is a schematic diagram of the structure of an intelligent prediction system for the aging state of an aluminum alloy flexible cable provided in an embodiment of this application.

[0012] The components represented by each number in the attached diagram are explained below: Multi-source data acquisition module 11, feature sequence establishment module 12, prediction model training module 13, and aging information prediction module 14. Detailed Implementation

[0013] This application provides an intelligent prediction method and system for the aging state of aluminum alloy flexible cables, which addresses the technical problems of existing technologies for predicting the aging state of aluminum alloy flexible cables, such as complex offline detection operations, single online monitoring parameters, and insufficient consideration of nonlinear time-varying effects, resulting in limited prediction accuracy and generalization ability.

[0014] Example 1, as Figure 1 As shown in the figure, this application provides an intelligent prediction method for the aging state of aluminum alloy flexible cables, including: S10: Acquire multi-source operational data of the target cable during its current lifecycle, and simultaneously acquire prior material performance data of the target cable; In this embodiment of the application, multi-source operating data of the current life cycle of the target cable and the prior material performance data of the target cable are first obtained. The multi-source operating data includes environmental state data collected during operation, and the prior material performance data includes the intrinsic material parameters of the target cable at the time of manufacture.

[0015] This includes acquiring multi-source operational data for the current lifecycle of the target cable, prior to which the following is included: By using correlation analysis, we traverse each candidate feature in the multi-source operational data to determine the degree of correlation with historical aging data. Features with a correlation degree higher than a preset correlation threshold are extracted and stored as environmental status data lists, load monitoring data lists, and mechanical monitoring data lists, respectively. Iterate through the correlation between each candidate parameter in the prior material performance data and the historical aging data, extract parameters with a correlation higher than the preset correlation threshold, and store them as a prior material performance data list.

[0016] In this embodiment, firstly, features strongly correlated with aging status are selected through association analysis, and redundant and irrelevant candidate features are eliminated. This effectively reduces the input dimension of the subsequent model, reduces the amount of computation, and avoids interference from irrelevant features with the prediction results. The association analysis method is gray association analysis, which quantifies the degree of correlation by calculating the gray correlation degree between each candidate feature and the historical aging amount. It can handle association analysis scenarios with small samples and multiple factors, and the screening results are highly reliable, which is suitable for the feature screening requirements of multi-source data in this application.

[0017] Secondly, features with correlation higher than a preset correlation threshold are extracted and categorized into three monitoring data lists and a priori material performance data list according to data type. This facilitates the targeted construction of different feature sequences, providing a data foundation for dual-channel feature coding. In this step, after feature selection through correlation analysis, when actually collecting target cable data, only the feature data within the corresponding list needs to be collected, eliminating the need to collect all candidate data. This effectively reduces the workload and cost of on-site data collection and improves the overall efficiency of the prediction process.

[0018] Specifically, step S10 in the method includes: Collect environmental status data, which includes at least ambient temperature data, ambient wind field data, and ambient light intensity data; Collect load monitoring data, which includes at least operating current data and power factor data; Collect mechanical monitoring data, which includes at least triaxial vibration acceleration data collected by vibration sensors installed on the target cable body or support structure; Based on the specifications of the target cable, the prior material performance data is extracted, wherein the prior material performance data includes at least conductor mechanical performance data and insulation sheath thermo-oxidative aging data.

[0019] In this embodiment, firstly, ambient temperature and light intensity affect the thermal-oxidative aging rate of insulation materials, and ambient wind field changes the heat dissipation efficiency and vibration characteristics of the cable surface. Incorporating these into the environmental condition data can fully cover the impact of environmental factors on cable aging.

[0020] Secondly, the operating current determines the Joule heating of the cable conductor, and combined with the power factor, it can accurately reflect the actual load level of the cable. During long-term operation, the cable is affected by ambient wind, vibration of surrounding equipment, and fluctuations in load current, which will generate continuous mechanical vibration. Long-term vibration will cause relative displacement and interface wear between the insulation layer and the conductor, further accelerating aging. Collecting triaxial vibration acceleration data can completely capture the mechanical vibration state of the cable.

[0021] Finally, conductor mechanical property data and insulation sheath thermo-oxidative aging data are extracted from the target cable's specifications. This allows the intrinsic aging characteristics of the materials to be incorporated into the prediction process, avoiding prediction deviations caused by material differences between different cable models. The conductor mechanical property data includes tensile strength, elongation, conductivity, and hardness; the insulation sheath thermo-oxidative aging data includes carbonyl index, oxidation induction time, tensile strength retention rate, elongation at break retention rate, volume resistivity, dielectric loss factor, hardness change, yellowing index, crosslinking density, gel content, thermal weight loss rate, and water absorption rate.

[0022] S20: Combining the multi-source operating data with the prior material performance data, establish the mechanical characteristic sequence, load characteristic sequence, and condition vector of the target cable; In this embodiment, based on the screened mechanical monitoring data, mechanical feature sequences corresponding to the time steps are obtained by organizing the data in the time sampling order. Each time step retains the statistical characteristics of triaxial vibration acceleration, including mean, peak value, and root mean square, which can reflect the mechanical stress state of the cable at different times. Similarly, the operating current and power factor in the load monitoring data are organized according to the same time segmentation rules to obtain the load feature sequence, which reflects the change law of cable load level over time. After normalizing the screened prior material performance data, a fixed-dimensional condition vector is spliced ​​to input the intrinsic material attribute information of the target cable into the model, so as to realize the prediction and adaptation of cables with different material parameters.

[0023] Specifically, step S20 in the method includes: The multi-source operational data is preprocessed, and based on the preprocessing results, environmental temperature features, environmental wind field features, and environmental illumination features are extracted from the preprocessed environmental state data. Time-frequency analysis was performed on the preprocessed mechanical monitoring data to determine the time-frequency vibration characteristics; The mechanical feature sequence is obtained by splicing together the ambient temperature feature, the ambient wind field feature, the ambient light feature, and the time-frequency vibration feature; By splicing together the preprocessed load monitoring data, the ambient temperature characteristics, the ambient wind field characteristics, and the ambient light characteristics, the load characteristic sequence is obtained. Based on the prior material performance data, the conditional vector, including the conductor mechanical property sub-vector and the thermo-oxidative aging sub-vector, is extracted.

[0024] In this embodiment, the multi-source operational data is first preprocessed, specifically including outlier removal, missing value interpolation and normalization, to eliminate numerical differences between features of different dimensions. Outlier removal uses the Grubbs test to identify and remove abnormal data that exceeds the confidence interval. Missing values ​​are filled using cubic spline interpolation based on the feature data of adjacent time steps to ensure the temporal continuity of the feature sequence.

[0025] Then, time-frequency analysis was performed on the mechanical monitoring data. Energy characteristics of different frequency bands were extracted through wavelet decomposition, and combined with time-domain statistical features to obtain complete time-frequency vibration characteristics. This reflects the mechanical stress information contained in the cable vibration signal and captures the changes in vibration characteristics caused by aging. Specifically, the original vibration signal was decomposed into approximation coefficients and detail coefficients at different scales through wavelet decomposition. The energy proportion of each coefficient was calculated and concatenated with time-domain statistical features to obtain time-frequency vibration characteristics. These characteristics were then combined with environmental features extracted from environmental data to obtain a complete mechanical feature sequence, fully covering the coupled influence of mechanical stress and environmental factors on cable aging.

[0026] For the load characteristic sequence, the preprocessed raw load monitoring data, such as operating current and power factor, are concatenated with the corresponding ambient temperature, wind field, and illumination characteristics. Each time step yields a complete set of load-related features, which are then arranged in chronological order to obtain the load characteristic sequence. This sequence reflects the impact of the cable's load state coupled with environmental factors on the aging process, more closely mirroring the aging process under the combined effects of multiple factors in actual operation. For the condition vector, the normalized conductor mechanical performance parameters are organized into conductor mechanical performance sub-vectors, and the normalized insulation sheath thermo-oxidative aging parameters are organized into thermo-oxidative aging sub-vectors. The two sub-vectors are then directly concatenated to obtain a fixed-dimensional condition vector, which fully carries the intrinsic material property information of the target cable.

[0027] S30: Based on the mechanical feature sequence, the load feature sequence and the condition vector, and combined with the pre-trained dual-channel coding-regression prediction model, the aging intensity of the target cable is predicted in segments to obtain the aging intensity sequence. In this embodiment, the mechanical feature sequence and the load feature sequence are respectively input into two encoding branches of the pre-trained dual-channel encoding-regression prediction model. The mechanical feature encoding branch uses a one-dimensional convolutional neural network combined with a gated recurrent unit to extract deep features in the time dimension, while the load feature encoding branch uses a Transformer encoder to extract long-range temporal dependent features. The deep features extracted by the two branches are fused through an attention fusion module, and then the fused features are modulated through a conditional vector. The modulated features are input into a fully connected regression layer, and the aging intensity of the corresponding segment is output at each time step. The prediction results of all segments are spliced ​​together to obtain the complete aging intensity sequence.

[0028] Specifically, based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, the target cable is segmented to predict its aging strength, obtaining an aging strength sequence. Prior to this, the process includes: Based on the preset first load window length and first overlap ratio, the load feature sequence is segmented by a sliding window to obtain a load feature segment set; Based on the preset second mechanical window length and second overlap ratio, the mechanical feature sequence is segmented by a sliding window to obtain a mechanical feature segment set; The mechanical feature segment set and the load feature segment set are time-series aligned and a mapping relationship is established, wherein each load feature segment corresponds to N mechanical feature segments; Wherein, the length of the first load window is N times the length of the second mechanical window, and N is a positive integer greater than 2.

[0029] In this embodiment, firstly, the load feature sequence is segmented by a sliding window according to a preset first load window length and a first overlap ratio to obtain load segments for different time intervals. Then, the mechanical feature sequence is segmented by a sliding window according to a preset second mechanical window length and a second overlap ratio to obtain mechanical feature segments for different time intervals. The first load window length is N times the second mechanical window length because the impact of cable load changes on aging has a cumulative effect and a longer time scale, while the impact of mechanical vibration characteristics on aging is more high-frequency, requiring finer time granularity. Therefore, using a segmentation strategy with different lengths adapts to the time scale characteristics of the two types of features, enabling more accurate capture of the aging effects of different features.

[0030] After segmentation, a temporal mapping relationship is established between the mechanical feature segments and the corresponding load feature segments within the same time interval. This completes the alignment processing of the input data. The temporal mapping relationship facilitates the model's extraction of feature information at different scales while ensuring the correspondence between the two types of features in the time dimension, avoiding temporal misalignment that could interfere with feature fusion. During model training, historical data of aluminum alloy flexible cables that have undergone aging testing are used as the training set. The aging intensity of each segment obtained from actual testing is used as the label. The mean squared error loss function is used to optimize the model parameters, ultimately resulting in a converged pre-trained dual-channel encoding-regression prediction model. After training, the input features are processed according to the same segment alignment rules during the prediction phase. The input to the model yields the aging intensity prediction results for each segment, which are then concatenated to form a complete aging intensity sequence.

[0031] Specifically, step S30 in the method includes: Each mechanical feature segment is input into the mechanical feature encoding channel of the dual-channel encoding-regression prediction model, the load feature segment corresponding to the mechanical feature segment is input into the load feature encoding channel of the dual-channel encoding-regression prediction model, and the condition vector is simultaneously injected into the mechanical feature encoding channel and the load feature encoding channel. In the mechanical feature encoding channel, after linear projection of the mechanical feature segments, feature condition modulation is performed based on the condition vector, and then processed by multi-head self-attention encoding and feedforward network to obtain the mechanical strength encoding vector. In the load feature encoding channel, after linear projection of the load feature segments, feature condition modulation is performed based on the condition vector, and the load intensity encoding vector is obtained after multi-head self-attention encoding and feedforward network processing. Attention feature fusion is performed on the mechanical strength encoding vector and the load strength encoding vector to obtain a fused encoding vector; The fused encoding vector is input into the aging intensity prediction head and the aging amount prediction head respectively to obtain the aging intensity value and aging amount value corresponding to each mechanical feature segment. Arrange the aging intensity values ​​corresponding to all the mechanical feature segments in chronological order to obtain the aging intensity sequence.

[0032] In this embodiment, firstly, the mechanical feature segments and the corresponding load feature segments are input into the corresponding encoding channels. Then, the two types of segment features are linearly projected and mapped to a feature space of the same dimension. Then, the projection features of the two channels are conditionally modulated by conditional vectors to integrate the intrinsic material properties of the target cable into the encoding process, so that the encoding process can adapt to the cable feature processing requirements of different material parameters.

[0033] Then, the dependencies within the features are captured by multi-head self-attention encoding, and deep nonlinear features are further extracted by combining a feedforward network, resulting in mechanical strength encoding vector and load strength encoding vector. The mechanical feature encoding channel focuses on capturing the coupling effect between high-frequency vibration features and environmental factors, while the load feature encoding channel focuses on capturing the influence of long-cycle load accumulation effect. The independent encoding of the two channels can avoid mutual interference between features of different scales and ensure the targeting of feature extraction.

[0034] Furthermore, after obtaining the two types of encoding vectors, feature fusion is achieved through a cross-attention module. The mechanical strength encoding vector is used as the query, and the load strength encoding vector is used as the key and value to calculate the attention weight. The feature information of the two different scales is fused to obtain a fused encoding vector containing the coupling effect of multiple factors.

[0035] Finally, the fused encoding vector is fed into two prediction heads. The aging intensity prediction head outputs the aging intensity value of the current segment, and the aging amount prediction head outputs the cumulative aging amount corresponding to the current segment. The aging intensity values ​​obtained by arranging all mechanical feature segments in chronological order are spliced ​​together to form a complete aging intensity sequence that reflects the aging rate changes at different stages of the cable's entire life cycle, providing a basis for subsequent prediction of remaining service life.

[0036] Furthermore, the construction of the dual-channel encoding-regression prediction model includes: A dual-channel encoder layer is established, wherein the dual-channel encoder layer includes a mechanical feature encoding channel and a load feature encoding channel arranged in parallel, and both the mechanical feature encoding channel and the load feature encoding channel include a linear projection module, a conditional modulation module, a multi-head self-attention module and a feedforward network module connected in sequence; Based on the prior aging database, coded sample data is established, and the mechanical feature encoding channel is trained with the sample mechanical feature segmentation and sample condition vector in the coded sample data as input and the joint loss function including reconstruction loss and physical consistency loss as the objective. Using the sample payload feature segments in the encoded sample data and the conditional vector as input, and with the goal of minimizing the joint loss function, the payload feature encoding channel is trained. The encoded sample data is input into the trained mechanical feature encoding channel and load feature encoding channel to obtain dual-channel encoded output and concatenate them into sample fusion encoded features. Combined with the true value aging state, regression prediction samples are constructed. Using the sample fusion encoding features as input and the true aging state as supervision, a regression prediction layer based on machine learning regression is constructed and trained, wherein the regression prediction layer includes an aging intensity prediction head and an aging amount prediction head. Connect the input of the trained regression prediction layer to the output of the dual-channel encoder layer to obtain the dual-channel encoder-regression prediction model.

[0037] In this embodiment, firstly, the basic structure of a dual-channel encoder layer is constructed. Two encoding channels are set in parallel, corresponding to mechanical and load features at different time scales. Within each channel, a linear projection module unifies the feature dimension, and then a conditional modulation module incorporates the intrinsic properties of the material. Subsequently, a multi-head self-attention module captures temporal dependencies, and finally, a feedforward network module extracts deep nonlinear features, completing the basic encoder structure. The conditional modulation module, based on a feature linear modulation mechanism, uses conditional gating vectors and conditional bias vectors generated from conditional vectors to perform element-wise scaling and translation of the projected features.

[0038] Then, the historical cable operation data labeled in the prior aging database is used to construct the coding samples. The two coding channels are pre-trained respectively. During the training process, not only is the reconstruction loss used to ensure the accuracy of feature extraction, but also the physical consistency loss is introduced to constrain the coding features to conform to the physical laws of the aging process, so as to avoid the model learning erroneous feature associations that violate the aging mechanism and ensure the rationality of the pre-trained coding.

[0039] After pre-training two encoding channels, all encoded samples are passed through the trained dual-channel encoder to obtain the encoded output. The sample fusion encoding features are obtained by cross-attention fusion and combined with the aging state of the actual detection ground truth corresponding to the sample to construct the training samples for regression prediction.

[0040] Finally, using the fused coding features as input and the ground truth aging state as the supervision label, a regression prediction layer containing two prediction heads is trained. The aging intensity prediction head learns to predict the aging rate per unit time period, and the aging amount prediction head learns to predict the cumulative aging degree of the corresponding time period. After training, the regression prediction layer is connected and integrated with the dual-channel encoder to obtain a complete pre-trained dual-channel encoder-regression prediction model.

[0041] For example, a dual-channel encoding-regression prediction model is trained with an initial learning rate of 0.001. The AdamW optimizer is used for parameter updates, and the batch size is set to 32. The pre-training phase consists of 50 rounds, with model weights saved every 10 rounds. The weights with the lowest loss on the validation set are selected as the pre-training results. In the regression prediction layer training phase, the learning rate is adjusted to 0.0001, and the AdamW optimizer is also used. The model is trained for 20 rounds, with weights saved every 5 rounds. The weights with the lowest prediction error on the validation set are selected as the final model parameters to ensure the model's generalization ability and prediction accuracy on the target task.

[0042] Furthermore, establishing a dual-channel encoder layer also includes: An orthogonal test factor table is set up based on the aging influencing factors of the target cable. An orthogonal test scheme is designed based on the orthogonal test factor table, and an accelerated aging test is performed to obtain the comprehensive aging rate response value of each test group. A range analysis is performed on the orthogonal test results to calculate the range value of each factor. The range value is then normalized and used as the initial weight value of the linear projection module in the mechanical feature encoding channel and the load feature encoding channel. Based on physical mechanism analysis, a set of constraint sets on the magnitude relationship between input features and an interaction matrix of promoting and antagonizing relationships are established. During model training, after each gradient update, the weights of the linear projection module are verified based on the set of size relationship constraints and the interaction matrix of facilitating and antagonistic relationships.

[0043] In this embodiment of the application, firstly, an orthogonal test is conducted to investigate the degree of influence of different influencing factors on the aging of aluminum alloy flexible cables. Specifically, an orthogonal test factor table is set up, and key influencing factors such as mechanical vibration stress, operating load, ambient temperature, and humidity are listed as orthogonal test factors. Multiple different levels are set for each factor, and multiple sets of accelerated aging tests are carried out according to the orthogonal test scheme to measure the comprehensive aging rate response value corresponding to each set of tests.

[0044] Then, range analysis was performed on the experimental results to calculate the range value corresponding to each factor. After normalization, the weight coefficients reflecting the influence of each factor were obtained. These weight coefficients were directly used as the initial weights of the linear projection modules in the two encoding channels to replace random initialization, so that the initial state of the model conforms to the physical law of aging effect, reducing the convergence time of model training and improving the final prediction accuracy.

[0045] Specifically, range normalization is achieved by subtracting the minimum range value from the range value of each factor, and then dividing by the difference between the maximum and minimum range values ​​of all factors. The result is mapped to the [0,1] interval to obtain the normalized weight coefficient of each influencing factor. This ensures that the relative size of the weight coefficient is consistent with the range value ranking of the influence of each factor on aging, and can accurately reflect the strength of different factors on cable aging.

[0046] Subsequently, based on the physical mechanism of cable aging, the size relationship constraints between different input features and the promoting or antagonistic relationships between different factors were compiled. Corresponding size relationship constraint sets and promoting / antagonistic relationship interaction matrices were established. The size relationship constraint set defines the weight order between feature pairs, and the promoting / antagonistic relationship interaction matrix uses positive one to represent promoting relationships, negative one to represent antagonistic relationships, and zero to represent no interaction. After the gradient update is completed in each round of model training, the weights of the linear projection module are verified and corrected according to the above constraints. Weight update results that do not conform to physical laws are eliminated to ensure that the feature weights learned by the model conform to the actual aging mechanism and avoid erroneous predictions that violate physical common sense.

[0047] S40: Based on the aging intensity sequence, predict the aging status information of the target cable.

[0048] In this embodiment, the aging rate change of the target cable in each period of the entire operating cycle can be intuitively obtained from the aging intensity sequence. Combined with the current operating time and the cumulative aging amount, the current overall aging status level can be calculated. At the same time, by extrapolating the aging intensity change trend, the remaining service life of the target cable when it reaches the aging failure threshold can be predicted, thus completing the final intelligent prediction of the aging status.

[0049] Specifically, step S40 in the method includes: Curve fitting is performed on the aging intensity sequence to determine a smooth aging intensity curve; Based on the aging intensity curve, the aging values ​​corresponding to all time windows are cumulatively summed to obtain the cumulative fitted aging amount. Based on the cumulative fitted aging amount, combined with the failure threshold aging amount, the predicted aging life of the target cable is calculated. The cumulative fitted aging amount is matched with a preset aging level threshold range to determine the aging level of the target cable. The predicted aging life and the aging level are the aging state information.

[0050] In this embodiment, firstly, the discrete aging intensity values ​​obtained from each segment are curve fitted to remove the local fluctuations caused by segment prediction, resulting in a smooth and continuous aging intensity curve that reflects the aging rate change with the running time. This curve can more clearly reflect the trend of aging rate change with the running process. For example, the fitting calculation is completed by combining the least squares method with cubic spline interpolation to ensure the continuity and smoothness of the curve at each segment node and avoid large oscillations or sudden changes in the fitting results.

[0051] Then, based on the fitted aging intensity curve, the aging intensity of each time period is integrated to obtain the cumulative fitted aging amount under any running time, which intuitively reflects the overall aging degree of the cable. Specifically, the aging amount values ​​of each time window segment are directly summed to obtain the cumulative fitted aging amount that is consistent with the integral result of the aging intensity curve. The calculation process is simple and the accuracy meets the engineering prediction requirements.

[0052] Then, the cumulative fitted aging amount reaching the preset failure threshold aging amount is used as the failure judgment condition. The total predicted operating time of the cable from the time it is put into operation to the time it reaches the failure threshold is extrapolated and calculated. Subtracting the operating time of the cable, the remaining predicted aging life of the target cable can be obtained. The current cumulative fitted aging amount is compared and matched with the preset multi-level aging level threshold range to determine the current aging level of the target cable, providing a clear reference for operation and maintenance decisions.

[0053] Finally, the predicted aging life and current aging level are output as the aging status information of the target cable, completing the intelligent prediction process of the aging status of the entire aluminum alloy flexible cable.

[0054] In summary, compared with existing technologies, this application independently encodes mechanical and load features by inputting them into different encoding channels at different scales. At the same time, it introduces conditional modulation to incorporate the intrinsic properties of cable materials, which can specifically extract feature information of different influencing factors, avoid mutual interference between multi-scale features, and combine physical constraint initialization and training constraints to ensure that the model learning process conforms to the aging physical mechanism. This effectively improves the accuracy and reliability of aging state prediction for aluminum alloy flexible cables, solves the problems of traditional prediction methods being difficult to adapt to cables with different material parameters and insufficient feature extraction targeting, and provides a more accurate aging state reference for power cable operation and maintenance.

[0055] In summary, the embodiments of this application have at least the following technical effects: This application provides an intelligent prediction method for the aging state of aluminum alloy flexible cables. First, it acquires multi-source operational data and prior material performance data of the target cable's current lifecycle. Second, it combines the multi-source operational data and prior material performance data to construct a mechanical feature sequence, a load feature sequence, and a conditional vector reflecting the cable's operational state and material properties. Third, it uses a pre-trained dual-channel encoding-regression prediction model to predict the segmented aging intensity of the target cable. Finally, based on the obtained aging intensity sequence, it predicts the aging state information of the target cable, including predicting the aging life and determining the aging level.

[0056] Through the above technical solutions, this application effectively improves the accuracy and generalization ability of predicting cable aging status, overcomes the limitations of traditional offline testing and single-parameter online monitoring, and provides strong technical support for the safe operation and maintenance of aluminum alloy flexible cables.

[0057] Example 2, as Figure 2 As shown, based on the same inventive concept as the intelligent prediction method for the aging state of aluminum alloy flexible cables provided in Embodiment 1, this application also provides an intelligent prediction system for the aging state of aluminum alloy flexible cables, including: The multi-source data acquisition module 11 is used to acquire multi-source operating data of the target cable during its current life cycle, and to simultaneously acquire prior material performance data of the target cable. The feature sequence establishment module 12 is used to combine the multi-source operating data and the prior material performance data to establish the mechanical feature sequence, load feature sequence and condition vector of the target cable; The prediction model training module 13 is used to predict the segmented aging intensity of the target cable based on the mechanical feature sequence, the load feature sequence and the condition vector, combined with the pre-trained dual-channel encoding-regression prediction model, and to obtain the aging intensity sequence. The aging information prediction module 14 is used to predict the aging status information of the target cable based on the aging intensity sequence.

[0058] Furthermore, in one embodiment, acquiring multi-source operational data of the target cable's current lifecycle includes, prior to: By using correlation analysis, we traverse each candidate feature in the multi-source operational data to determine the degree of correlation with historical aging data. Features with a correlation degree higher than a preset correlation threshold are extracted and stored as environmental status data lists, load monitoring data lists, and mechanical monitoring data lists, respectively. Iterate through the correlation between each candidate parameter in the prior material performance data and the historical aging data, extract parameters with a correlation higher than the preset correlation threshold, and store them as a prior material performance data list.

[0059] In one embodiment, the multi-source data acquisition module 11 is specifically used for: Collect environmental status data, which includes at least ambient temperature data, ambient wind field data, and ambient light intensity data; Collect load monitoring data, which includes at least operating current data and power factor data; Collect mechanical monitoring data, which includes at least triaxial vibration acceleration data collected by vibration sensors installed on the target cable body or support structure; Based on the specifications of the target cable, the prior material performance data is extracted, wherein the prior material performance data includes at least conductor mechanical performance data and insulation sheath thermo-oxidative aging data.

[0060] In one embodiment, the feature sequence establishment module 12 is specifically used for: The multi-source operational data is preprocessed, and based on the preprocessing results, environmental temperature features, environmental wind field features, and environmental illumination features are extracted from the preprocessed environmental state data. Time-frequency analysis was performed on the preprocessed mechanical monitoring data to determine the time-frequency vibration characteristics; The mechanical feature sequence is obtained by splicing together the ambient temperature feature, the ambient wind field feature, the ambient light feature, and the time-frequency vibration feature; By splicing together the preprocessed load monitoring data, the ambient temperature characteristics, the ambient wind field characteristics, and the ambient light characteristics, the load characteristic sequence is obtained. Based on the prior material performance data, the conditional vector, including the conductor mechanical property sub-vector and the thermo-oxidative aging sub-vector, is extracted.

[0061] Further, in one embodiment, based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, segmented aging strength prediction of the target cable is performed to obtain an aging strength sequence. Prior to this, the process includes: Based on the preset first load window length and first overlap ratio, the load feature sequence is segmented by a sliding window to obtain a load feature segment set; Based on the preset second mechanical window length and second overlap ratio, the mechanical feature sequence is segmented by a sliding window to obtain a mechanical feature segment set; The mechanical feature segment set and the load feature segment set are time-series aligned and a mapping relationship is established, wherein each load feature segment corresponds to N mechanical feature segments; Wherein, the length of the first load window is N times the length of the second mechanical window, and N is a positive integer greater than 2.

[0062] In one embodiment, the prediction model training module 13 is specifically used for: Each mechanical feature segment is input into the mechanical feature encoding channel of the dual-channel encoding-regression prediction model, the load feature segment corresponding to the mechanical feature segment is input into the load feature encoding channel of the dual-channel encoding-regression prediction model, and the condition vector is simultaneously injected into the mechanical feature encoding channel and the load feature encoding channel. In the mechanical feature encoding channel, after linear projection of the mechanical feature segments, feature condition modulation is performed based on the condition vector, and then processed by multi-head self-attention encoding and feedforward network to obtain the mechanical strength encoding vector. In the load feature encoding channel, after linear projection of the load feature segments, feature condition modulation is performed based on the condition vector, and the load intensity encoding vector is obtained after multi-head self-attention encoding and feedforward network processing. Attention feature fusion is performed on the mechanical strength encoding vector and the load strength encoding vector to obtain a fused encoding vector; The fused encoding vector is input into the aging intensity prediction head and the aging amount prediction head respectively to obtain the aging intensity value and aging amount value corresponding to each mechanical feature segment. Arrange the aging intensity values ​​corresponding to all the mechanical feature segments in chronological order to obtain the aging intensity sequence.

[0063] Furthermore, the construction of the dual-channel encoding-regression prediction model includes: A dual-channel encoder layer is established, wherein the dual-channel encoder layer includes a mechanical feature encoding channel and a load feature encoding channel arranged in parallel, and both the mechanical feature encoding channel and the load feature encoding channel include a linear projection module, a conditional modulation module, a multi-head self-attention module and a feedforward network module connected in sequence; Based on the prior aging database, coded sample data is established, and the mechanical feature encoding channel is trained with the sample mechanical feature segmentation and sample condition vector in the coded sample data as input and the joint loss function including reconstruction loss and physical consistency loss as the objective. Using the sample payload feature segments in the encoded sample data and the conditional vector as input, and with the goal of minimizing the joint loss function, the payload feature encoding channel is trained. The encoded sample data is input into the trained mechanical feature encoding channel and load feature encoding channel to obtain dual-channel encoded output and concatenate them into sample fusion encoded features. Combined with the true value aging state, regression prediction samples are constructed. Using the sample fusion encoding features as input and the true aging state as supervision, a regression prediction layer based on machine learning regression is constructed and trained, wherein the regression prediction layer includes an aging intensity prediction head and an aging amount prediction head. Connect the input of the trained regression prediction layer to the output of the dual-channel encoder layer to obtain the dual-channel encoder-regression prediction model.

[0064] Furthermore, establishing a dual-channel encoder layer also includes: An orthogonal test factor table is set up based on the aging influencing factors of the target cable. An orthogonal test scheme is designed based on the orthogonal test factor table, and an accelerated aging test is performed to obtain the comprehensive aging rate response value of each test group. A range analysis is performed on the orthogonal test results to calculate the range value of each factor. The range value is then normalized and used as the initial weight value of the linear projection module in the mechanical feature encoding channel and the load feature encoding channel. Based on physical mechanism analysis, a set of constraint sets on the magnitude relationship between input features and an interaction matrix of promoting and antagonizing relationships are established. During model training, after each gradient update, the weights of the linear projection module are verified based on the set of size relationship constraints and the interaction matrix of facilitating and antagonistic relationships.

[0065] In one embodiment of the application, the aging information prediction module 14 is specifically used for: Curve fitting is performed on the aging intensity sequence to determine a smooth aging intensity curve; Based on the aging intensity curve, the aging values ​​corresponding to all time windows are cumulatively summed to obtain the cumulative fitted aging amount. Based on the cumulative fitted aging amount, combined with the failure threshold aging amount, the predicted aging life of the target cable is calculated. The cumulative fitted aging amount is matched with a preset aging level threshold range to determine the aging level of the target cable. The predicted aging life and the aging level are the aging state information.

Claims

1. A method for intelligently predicting the aging state of aluminum alloy flexible cables, characterized in that, include: Acquire multi-source operational data of the target cable during its current lifecycle, and simultaneously acquire prior material performance data of the target cable; By combining the multi-source operational data with the prior material performance data, a mechanical characteristic sequence, a load characteristic sequence, and a condition vector for the target cable are established. Based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, the aging intensity of the target cable is predicted in segments to obtain the aging intensity sequence. Based on the aging intensity sequence, the aging status information of the target cable is predicted.

2. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, Acquire multi-source operational data for the current lifecycle of the target cable, and simultaneously acquire prior material performance data for the target cable, including: Collect environmental status data, which includes at least ambient temperature data, ambient wind field data, and ambient light intensity data; Collect load monitoring data, which includes at least operating current data and power factor data; Collect mechanical monitoring data, which includes at least triaxial vibration acceleration data collected by vibration sensors installed on the target cable body or support structure; Based on the specifications of the target cable, the prior material performance data is extracted, wherein the prior material performance data includes at least conductor mechanical performance data and insulation sheath thermo-oxidative aging data.

3. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, Combining the multi-source operational data and the prior material performance data, a mechanical characteristic sequence, a load characteristic sequence, and a conditional vector for the target cable are established, including: The multi-source operational data is preprocessed, and based on the preprocessing results, environmental temperature features, environmental wind field features, and environmental illumination features are extracted from the preprocessed environmental state data. Time-frequency analysis was performed on the preprocessed mechanical monitoring data to determine the time-frequency vibration characteristics; The mechanical feature sequence is obtained by splicing together the ambient temperature feature, the ambient wind field feature, the ambient light feature, and the time-frequency vibration feature; By splicing together the preprocessed load monitoring data, the ambient temperature characteristics, the ambient wind field characteristics, and the ambient light characteristics, the load characteristic sequence is obtained. Based on the prior material performance data, the conditional vector, including the conductor mechanical property sub-vector and the thermo-oxidative aging sub-vector, is extracted.

4. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, Based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, the aging strength of the target cable is predicted in segments to obtain the aging strength sequence. Prior to this, the following steps are included: Based on the preset first load window length and first overlap ratio, the load feature sequence is segmented by a sliding window to obtain a load feature segment set; Based on the preset second mechanical window length and second overlap ratio, the mechanical feature sequence is segmented by a sliding window to obtain a mechanical feature segment set; The mechanical feature segment set and the load feature segment set are time-series aligned and a mapping relationship is established, wherein each load feature segment corresponds to N mechanical feature segments; Wherein, the length of the first load window is N times the length of the second mechanical window, and N is a positive integer greater than 2.

5. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 4, characterized in that, Based on the mechanical feature sequence, the load feature sequence, and the condition vector, and combined with a pre-trained dual-channel encoding-regression prediction model, the aging intensity of the target cable is predicted in segments to obtain an aging intensity sequence, including: Each mechanical feature segment is input into the mechanical feature encoding channel of the dual-channel encoding-regression prediction model, the load feature segment corresponding to the mechanical feature segment is input into the load feature encoding channel of the dual-channel encoding-regression prediction model, and the condition vector is simultaneously injected into the mechanical feature encoding channel and the load feature encoding channel. In the mechanical feature encoding channel, after linear projection of the mechanical feature segments, feature condition modulation is performed based on the condition vector, and then processed by multi-head self-attention encoding and feedforward network to obtain the mechanical strength encoding vector. In the load feature encoding channel, after linear projection of the load feature segments, feature condition modulation is performed based on the condition vector, and the load intensity encoding vector is obtained after multi-head self-attention encoding and feedforward network processing. Attention feature fusion is performed on the mechanical strength encoding vector and the load strength encoding vector to obtain a fused encoding vector; The fused encoding vector is input into the aging intensity prediction head and the aging amount prediction head respectively to obtain the aging intensity value and aging amount value corresponding to each mechanical feature segment. Arrange the aging intensity values ​​corresponding to all the mechanical feature segments in chronological order to obtain the aging intensity sequence.

6. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, The construction of the dual-channel encoding-regression prediction model includes: A dual-channel encoder layer is established, wherein the dual-channel encoder layer includes a mechanical feature encoding channel and a load feature encoding channel arranged in parallel, and both the mechanical feature encoding channel and the load feature encoding channel include a linear projection module, a conditional modulation module, a multi-head self-attention module and a feedforward network module connected in sequence; Based on the prior aging database, coded sample data is established, and the mechanical feature encoding channel is trained with the sample mechanical feature segmentation and sample condition vector in the coded sample data as input and the joint loss function including reconstruction loss and physical consistency loss as the objective. Using the sample payload feature segments in the encoded sample data and the conditional vector as input, and with the goal of minimizing the joint loss function, the payload feature encoding channel is trained. The encoded sample data is input into the trained mechanical feature encoding channel and load feature encoding channel to obtain dual-channel encoded output and concatenate them into sample fusion encoded features. Combined with the true value aging state, regression prediction samples are constructed. Using the sample fusion encoding features as input and the true aging state as supervision, a regression prediction layer based on machine learning regression is constructed and trained, wherein the regression prediction layer includes an aging intensity prediction head and an aging amount prediction head. Connect the input of the trained regression prediction layer to the output of the dual-channel encoder layer to obtain the dual-channel encoder-regression prediction model.

7. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 6, characterized in that, Establishing a dual-channel encoder layer also includes: An orthogonal test factor table is set up based on the aging influencing factors of the target cable. An orthogonal test scheme is designed based on the orthogonal test factor table, and an accelerated aging test is performed to obtain the comprehensive aging rate response value of each test group. A range analysis is performed on the orthogonal test results to calculate the range value of each factor. The range value is then normalized and used as the initial weight value of the linear projection module in the mechanical feature encoding channel and the load feature encoding channel. Based on physical mechanism analysis, a set of constraint sets on the magnitude relationship between input features and an interaction matrix of promoting and antagonizing relationships are established. During model training, after each gradient update, the weights of the linear projection module are verified based on the set of size relationship constraints and the interaction matrix of facilitating and antagonistic relationships.

8. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, Based on the aging intensity sequence, predicting the aging state information of the target cable also includes: Curve fitting is performed on the aging intensity sequence to determine a smooth aging intensity curve; Based on the aging intensity curve, the aging values ​​corresponding to all time windows are cumulatively summed to obtain the cumulative fitted aging amount. Based on the cumulative fitted aging amount, combined with the failure threshold aging amount, the predicted aging life of the target cable is calculated. The cumulative fitted aging amount is matched with a preset aging level threshold range to determine the aging level of the target cable. The predicted aging life and the aging level are the aging state information.

9. The intelligent prediction method for the aging state of aluminum alloy flexible cables as described in claim 1, characterized in that, Prior to obtaining multi-source operational data for the target cable's current lifecycle, including: By using correlation analysis, we traverse each candidate feature in the multi-source operational data to determine the degree of correlation with historical aging data. Features with a correlation degree higher than a preset correlation threshold are extracted and stored as environmental status data lists, load monitoring data lists, and mechanical monitoring data lists, respectively. Iterate through the correlation between each candidate parameter in the prior material performance data and the historical aging data, extract parameters with a correlation higher than the preset correlation threshold, and store them as a prior material performance data list.

10. An intelligent prediction system for the aging state of aluminum alloy flexible cables, characterized in that, A smart prediction method for the aging state of an aluminum alloy flexible cable according to any one of claims 1-9 includes: The multi-source data acquisition module is used to acquire multi-source operational data of the target cable during its current life cycle, and simultaneously acquire prior material performance data of the target cable. The feature sequence establishment module is used to combine the multi-source operating data and the prior material performance data to establish the mechanical feature sequence, load feature sequence and condition vector of the target cable; The prediction model training module is used to predict the segmented aging intensity of the target cable based on the mechanical feature sequence, the load feature sequence and the condition vector, combined with a pre-trained dual-channel encoding-regression prediction model, and to obtain the aging intensity sequence. The aging information prediction module is used to predict the aging status information of the target cable based on the aging intensity sequence.

Images