Equipment life prediction method and device, electronic equipment and storage medium

Abstract

The invention provides an equipment life prediction method and device, electronic equipment and a storage medium, and relates to the technical field of neural networks. Comprising the steps that a feature sequence of to-be-predicted equipment is collected based on an electrical life test result, and the feature sequence comprises feature parameter sets under continuous time steps; decomposing and reconstructing the feature sequence according to the feature sequence of the to-be-predicted equipment and a plurality of preset scale factors to obtain at least one feature subsequence of the to-be-predicted equipment under each preset scale factor; determining the permutation entropy of the to-be-predicted equipment under each preset scale factor according to the sequence permutation mode of each feature subsequence of the to-be-predicted equipment under each preset scale factor, and determining a multi-scale feature matrix corresponding to the to-be-predicted equipment according to the permutation entropy under each preset scale factor; and according to the multi-scale feature matrix corresponding to the to-be-predicted device, using a pre-trained life prediction model to predict life information of the to-be-predicted device. The method improves the accuracy of the life prediction result.

Description

Technical Field

[0001] This application relates to the field of neural network technology, and more specifically, to a device lifetime prediction method, apparatus, electronic device, and storage medium. Background Technology

[0002] DC contactors are one of the main control devices in electrical drives, primarily used for frequently connecting and disconnecting DC circuits. Their performance stability and reliability directly affect the reliability of electrical equipment using the contactors and the safe operation of industrial production. By predicting the electrical life of DC contactors, potential equipment failure risks can be identified in a timely manner, allowing for appropriate maintenance measures to be taken to prevent equipment failures during operation and improve equipment reliability.

[0003] Currently, the electrical life of DC contactors is often predicted by the method of contact wear. This involves using specialized monitoring equipment to monitor the amount of contact wear in real time, analyze the amount of contact wear, and assess the lifespan of the DC contactor.

[0004] However, the above methods require specialized monitoring equipment and rely solely on contact wear for prediction, resulting in a limited range of predictive indicators and poor accuracy of the prediction results. Summary of the Invention

[0005] The purpose of this application is to address the shortcomings of the prior art by providing a method, apparatus, electronic device, and storage medium for predicting device electrical life, so as to improve the accuracy and efficiency of the prediction results.

[0006] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0007] In a first aspect, embodiments of this application provide a method for predicting equipment lifespan, including:

[0008] Based on the electrical life test results, a characteristic sequence of the device to be predicted is collected. The characteristic sequence includes a set of characteristic parameters under continuous time steps. Each set of characteristic parameters includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted.

[0009] Based on the feature sequence of the device to be predicted and multiple preset scale factors, the feature sequence is decomposed and reconstructed to obtain at least one feature subsequence of the device to be predicted under each preset scale factor.

[0010] Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, the arrangement entropy of the device to be predicted under each preset scale factor is determined, and based on the arrangement entropy under each preset scale factor, the multi-scale feature matrix corresponding to the device to be predicted is determined.

[0011] Based on the multi-scale feature matrix corresponding to the device to be predicted, a pre-trained lifetime prediction model is used to predict the lifetime information of the device to be predicted.

[0012] Optionally, the step of collecting the feature sequence of the device to be predicted based on the electrical lifetime test results includes:

[0013] Based on the electrical lifetime test performed on the device to be predicted, the current signal and voltage signal generated by the device to be predicted during the electrical lifetime test are collected;

[0014] Waveform analysis is performed on the current and voltage signals to obtain the characteristic sequence of the device to be predicted.

[0015] Optionally, the step of decomposing and reconstructing the feature sequence based on the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature sub-sequence of the device to be predicted under each preset scale factor includes:

[0016] Based on the target scale factor, determine each target time step to be extracted from the feature sequence;

[0017] Based on the sequence sorting pattern between each target time step and the feature parameter group under each target time step, the feature subsequences of the device to be predicted under the target scale factor are generated, wherein the sorting method of each target time step in each feature subsequence is different.

[0018] Optionally, determining the permutation entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor includes:

[0019] Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under the target scale factor, determine the probability of occurrence of each sequence arrangement pattern under the target scale factor;

[0020] Based on the occurrence probability of each sequence arrangement pattern under the target scale factor, determine the probability distribution of each sequence arrangement pattern under the target scale factor;

[0021] The permutation entropy under the target scale factor is determined based on the probability distribution of each sequence permutation pattern under the target scale factor.

[0022] Optionally, determining the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor includes:

[0023] The feature subsequences under each preset scale factor are integrated based on the permutation entropy under each preset scale factor to obtain the multi-scale feature matrix corresponding to the device to be predicted.

[0024] Optionally, the step of predicting the lifetime information of the device to be predicted using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted includes:

[0025] The multi-scale feature matrix is ​​input into the lifetime prediction model, and the lifetime prediction model determines the calculation weights corresponding to each time step in the multi-scale feature matrix based on the multi-scale feature matrix.

[0026] Based on the feature parameter groups of each time step in the multi-scale feature matrix and the calculation weights corresponding to each time step, a weighted calculation is performed to predict the lifetime information of the device to be predicted; the lifetime information is used to indicate the remaining number of times the device to be predicted can be used.

[0027] Optionally, the training process of the lifespan prediction model includes:

[0028] Collect sample feature sequences from multiple sample devices. The sample feature sequences contain a set of feature parameters of the sample devices at consecutive sample time steps, where each sample time step is labeled with a lifetime tag.

[0029] Based on the sample feature sequences of each sample device and multiple sample scale factors, the sample feature sequences of each sample device are decomposed and reconstructed to obtain at least one sample feature subsequence of each sample device under each sample scale factor.

[0030] Based on the sequence arrangement pattern of each sample feature subsequence under each sample scale factor, determine the sample multi-scale feature matrix corresponding to each sample device;

[0031] The lifetime prediction model is trained using the multi-scale feature matrix corresponding to each sample device.

[0032] Optionally, before decomposing and reconstructing the sample feature sequences of each sample device based on the sample feature sequences of each sample device and multiple sample scale factors to obtain at least one sample feature subsequence of each sample device under each sample scale factor, the method further includes:

[0033] Data normalization is performed on the sample feature sequences of each sample device, and the sample feature parameters in each sample feature sequence are scaled to a preset data range to obtain the processed sample feature sequences.

[0034] Optionally, the step of training the lifetime prediction model using the sample multi-scale feature matrix corresponding to each sample device includes:

[0035] The sample multi-scale feature matrix corresponding to each sample device is input into the initial lifetime prediction model. The initial lifetime prediction model iteratively corrects the model weights of the initial lifetime prediction model based on the sample multi-scale feature matrix to obtain the lifetime prediction model. The model weights are used to indicate the calculated weights corresponding to each time step in the feature sequence of the input model.

[0036] Optionally, the step of iteratively correcting the model weights of the initial lifetime prediction model based on the multi-scale feature matrix of each sample includes:

[0037] Based on the multi-scale feature matrix of each sample, determine the positional relationship between the time steps of each sample in the multi-scale feature matrix of each sample;

[0038] Based on the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample and the feature parameter set of each time step of each sample in the multi-scale feature matrix of each sample, the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step is predicted.

[0039] The model weights of the initial lifetime prediction model are iteratively corrected based on the predicted lifetime of each sample device under the characteristic parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device.

[0040] Optionally, the step of iteratively correcting the model weights of the initial lifetime prediction model based on the predicted lifetime of each sample device under the feature parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device includes:

[0041] The current loss value of the initial lifetime prediction model is calculated based on the predicted lifetime of each sample device under the characteristic parameter group indicated at each sample time step and the lifetime label corresponding to each sample time step of each sample device.

[0042] If the current loss value is greater than a preset loss threshold, the current model weights of the initial lifetime prediction model are iteratively corrected until the current loss value is less than or equal to the preset loss threshold. Then, the correction is stopped, and the model weights at the point where the correction stops are used as the model weights of the initial lifetime prediction model to obtain the lifetime prediction model.

[0043] Secondly, embodiments of this application also provide an equipment life prediction device, including: a data acquisition module, a processing module, a determination module, and a prediction module;

[0044] The acquisition module is used to acquire the feature sequence of the device to be predicted based on the electrical life test results. The feature sequence includes a set of feature parameters under continuous time steps. Each set of feature parameters includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted.

[0045] The processing module is used to decompose and reconstruct the feature sequence based on the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the device to be predicted under each preset scale factor.

[0046] The determining module is used to determine the permutation entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, and to determine the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor.

[0047] The prediction module is used to predict the lifetime information of the device to be predicted by using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted.

[0048] Optionally, the acquisition module is specifically used to acquire the current signal and voltage signal generated by the device under prediction during the electrical lifetime test, based on the electrical lifetime test performed on the device under prediction;

[0049] Waveform analysis is performed on the current and voltage signals to obtain the characteristic sequence of the device to be predicted.

[0050] Optionally, the processing module is specifically used to determine each target time step to be extracted from the feature sequence based on the target scale factor;

[0051] Based on the sequence sorting pattern between each target time step and the feature parameter group under each target time step, the feature subsequences of the device to be predicted under the target scale factor are generated, wherein the sorting method of each target time step in each feature subsequence is different.

[0052] Optionally, the determining module is specifically used to determine the probability of occurrence of each sequence arrangement pattern under the target scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under the target scale factor;

[0053] Based on the occurrence probability of each sequence arrangement pattern under the target scale factor, determine the probability distribution of each sequence arrangement pattern under the target scale factor;

[0054] The permutation entropy under the target scale factor is determined based on the probability distribution of each sequence permutation pattern under the target scale factor.

[0055] Optionally, the determining module is specifically used to integrate the feature subsequences under each preset scale factor according to the permutation entropy under each preset scale factor to obtain the multi-scale feature matrix corresponding to the device to be predicted.

[0056] Optionally, the prediction module is specifically used to input the multi-scale feature matrix into the lifetime prediction model, and the lifetime prediction model determines the calculation weights corresponding to each time step in the multi-scale feature matrix based on the multi-scale feature matrix.

[0057] Based on the feature parameter groups of each time step in the multi-scale feature matrix and the calculation weights corresponding to each time step, a weighted calculation is performed to predict the lifetime information of the device to be predicted; the lifetime information is used to indicate the remaining number of times the device to be predicted can be used.

[0058] Optionally, it may also include: a training module;

[0059] The training module is used to collect sample feature sequences from multiple sample devices. The sample feature sequences contain a set of feature parameters of the sample devices at consecutive sample time steps, wherein each sample time step is labeled with a lifetime tag.

[0060] Based on the sample feature sequences of each sample device and multiple sample scale factors, the sample feature sequences of each sample device are decomposed and reconstructed to obtain at least one sample feature subsequence of each sample device under each sample scale factor.

[0061] Based on the sequence arrangement pattern of each sample feature subsequence under each sample scale factor, determine the sample multi-scale feature matrix corresponding to each sample device;

[0062] The lifetime prediction model is trained using the multi-scale feature matrix corresponding to each sample device.

[0063] Optionally, the training module is specifically used to perform data normalization processing on the sample feature sequences of each sample device, scaling the sample feature parameters in each sample feature sequence to a preset data range, so as to obtain the processed sample feature sequences.

[0064] Optionally, the training module is specifically used to input the multi-scale feature matrix of each sample device into the initial lifetime prediction model, and the initial lifetime prediction model iteratively corrects the model weights of the initial lifetime prediction model based on the multi-scale feature matrix of each sample to obtain the lifetime prediction model; the model weights are used to indicate the calculated weights corresponding to each time step in the feature sequence of the input model.

[0065] Optionally, the training module is specifically used to determine the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample based on the multi-scale feature matrix of each sample.

[0066] Based on the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample and the feature parameter set of each time step of each sample in the multi-scale feature matrix of each sample, the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step is predicted.

[0067] The model weights of the initial lifetime prediction model are iteratively corrected based on the predicted lifetime of each sample device under the characteristic parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device.

[0068] Optionally, the training module is specifically used to calculate the current loss value of the initial lifetime prediction model based on the predicted lifetime of each sample device under the feature parameter group indicated by each sample time step and the lifetime label corresponding to each sample time step of each sample device.

[0069] If the current loss value is greater than a preset loss threshold, the current model weights of the initial lifetime prediction model are iteratively corrected until the current loss value is less than or equal to the preset loss threshold. Then, the correction is stopped, and the model weights at the point where the correction stops are used as the model weights of the initial lifetime prediction model to obtain the lifetime prediction model.

[0070] Thirdly, embodiments of this application provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to implement the device lifetime prediction method provided in the first aspect.

[0071] Fourthly, embodiments of this application provide a computer-readable storage medium storing a computer program that, when executed by a processor, performs the device lifetime prediction method as provided in the first aspect.

[0072] The beneficial effects of this application are:

[0073] This application provides a method, apparatus, electronic device, and storage medium for predicting equipment lifetime, comprising: acquiring a feature sequence of the device to be predicted based on electrical lifetime test results, the feature sequence including a set of feature parameters at continuous time steps; decomposing and reconstructing the feature sequence according to the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature sub-sequence of the device to be predicted under each preset scale factor; determining the permutation entropy of the device to be predicted under each preset scale factor according to the sequence arrangement pattern of each feature sub-sequence of the device to be predicted under each preset scale factor, and determining the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor; and predicting the lifetime information of the device to be predicted using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted. This method captures the features of the feature sequence at different scale factors based on an improved multi-scale permutation entropy algorithm, calculates the permutation entropy of the features at each scale factor, and integrates the feature sequence based on the permutation entropy to obtain the multi-scale fused features corresponding to the feature sequence, thereby improving the accuracy of the extracted features. Therefore, based on the extracted features, equipment lifetime prediction can be performed, improving the accuracy of the prediction results.

[0074] In addition, training a lifetime prediction model based on the Informer model framework can significantly improve the model's computational efficiency and its ability to capture long-term series dependencies, thereby improving the accuracy of the prediction results. Attached Figure Description

[0075] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0076] Figure 1 A flowchart illustrating a method for predicting equipment lifespan provided in an embodiment of this application;

[0077] Figure 2 A flowchart illustrating another device lifetime prediction method provided in this application embodiment;

[0078] Figure 3 A flowchart illustrating another equipment life prediction method provided in this application embodiment;

[0079] Figure 4 A flowchart illustrating another device lifetime prediction method provided in this application embodiment;

[0080] Figure 5 A flowchart illustrating another device lifetime prediction method provided in this application embodiment;

[0081] Figure 6 A flowchart illustrating a training method for a lifetime prediction model provided in an embodiment of this application;

[0082] Figure 7 A flowchart illustrating another training method for a lifetime prediction model provided in an embodiment of this application;

[0083] Figure 8 A flowchart illustrating another method for training a lifetime prediction model provided in an embodiment of this application;

[0084] Figure 9 This is a schematic diagram of a device for predicting device lifespan according to an embodiment of this application;

[0085] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0086] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0087] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0088] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0089] With the continuous development and improvement of power systems, the performance requirements for DC contactors are becoming increasingly stringent. Electrical life prediction, as a crucial method for evaluating DC contactor performance, is being applied more and more widely in power systems. Simultaneously, with the continuous development of technologies such as artificial intelligence and big data, electrical life prediction methods are constantly being innovated and improved, providing a more reliable guarantee for the stable operation of power systems.

[0090] This application provides a method for predicting equipment lifespan. It can capture the features of a feature sequence at different scale factors based on an improved multi-scale permutation entropy algorithm, obtain the multi-scale fusion features corresponding to the feature sequence, and improve the accuracy of the extracted features. Furthermore, based on the trained Informer lifespan prediction model, it can predict the remaining electrical life of a DC contactor, which can significantly improve the model's computational efficiency and its ability to capture long-term sequence dependencies, thereby improving the accuracy of the prediction results.

[0091] Figure 1 This is a flowchart illustrating a device lifetime prediction method provided in an embodiment of this application; the executing entity of this method can be a computing device such as a terminal or processor. Figure 1 As shown, the method may include:

[0092] S101. Based on the electrical life test results, collect the characteristic sequence of the device to be predicted. The characteristic sequence includes a set of characteristic parameters under continuous time steps.

[0093] Each characteristic parameter group includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted.

[0094] Of course, in practical applications, the characteristic parameters included in each characteristic parameter group are not limited to the three items listed above. For example, they may also include: the contact point temperature of the device to be predicted and the maximum voltage of the device to be predicted.

[0095] It is worth noting that when the feature parameter set includes five parameters—arc time of the device to be predicted, contact resistance of the device to be predicted, arc energy of the device to be predicted, contact point temperature of the device to be predicted, and maximum voltage of the device to be predicted—the model can obtain the optimal prediction result.

[0096] The arcing time is determined by the time interval between the arc's appearance and extinction. The arcing energy is determined by the accumulated voltage and current during the arc. The contact resistance is determined by measuring the ratio of voltage to current.

[0097] The device in this embodiment can refer to a DC contactor, and this solution can be used to predict the remaining electrical life of a DC contactor.

[0098] The electrical life test of the device to be predicted can be conducted to simulate the actual workload and frequency. The current and voltage information generated by the device during the electrical life test can be collected by the sensor, and the characteristic sequence of the device can be obtained based on the analysis of the current and voltage information.

[0099] Optionally, the feature sequence may include a set of feature parameters at continuous time steps, which can be obtained as time progresses during the electrical life test.

[0100] Each time step's feature parameter group may include multiple feature parameters. In this embodiment, only the feature parameters that have a significant impact on the lifespan of the device to be predicted are used as parameters in the feature parameter group, thereby reducing the amount of feature data processing and improving processing efficiency. S102. Based on the feature sequence of the device to be predicted and multiple preset scale factors, the feature sequence is decomposed and reconstructed to obtain at least one feature subsequence of the device to be predicted under each preset scale factor.

[0101] The scaling factor is used to indicate the filtering length of each time step in the feature sequence. Different feature subsequences are obtained by decomposing the feature sequence based on different scaling factors.

[0102] Optionally, multiple preset scale factors can be set to decompose and reconstruct the feature sequence under each preset scale factor, thereby obtaining at least one feature subsequence of the device to be predicted under each preset scale factor.

[0103] The reason why more than one feature subsequence is obtained under each preset scale factor is that after decomposing the feature sequence, each preset scale factor can have multiple reconstruction methods.

[0104] S103. Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, determine the arrangement entropy of the device to be predicted under each preset scale factor, and determine the multi-scale feature matrix corresponding to the device to be predicted based on the arrangement entropy under each preset scale factor.

[0105] Optionally, multiple feature subsequences can be generated under a preset scale factor, and each feature subsequence corresponds to a certain sequence arrangement pattern. Based on the sequence arrangement pattern of each feature subsequence, the arrangement entropy of the device to be predicted under a preset scale factor can be calculated.

[0106] Permutation entropy is an average entropy function used to measure the complexity of a time series and detect abrupt changes in dynamics. It reflects the complexity of the time series by transforming it into a series of permutations and calculating the entropy values ​​of these permutations. Here, the time series corresponds to the feature sequence mentioned above, and a feature subsequence can be called a permutation.

[0107] In some embodiments, feature integration is performed based on the permutation entropy of the device to be predicted under each preset scale factor, thereby obtaining the multi-scale feature matrix corresponding to the device to be predicted.

[0108] By using the above multi-scale processing method, the complexity of the feature sequence of the device to be predicted under different scale factors can be fully explored, thereby integrating the multi-scale feature matrix corresponding to the device to be predicted and improving the accuracy of the extracted features.

[0109] S104. Based on the multi-scale feature matrix corresponding to the device to be predicted, a pre-trained lifetime prediction model is used to predict the lifetime information of the device to be predicted.

[0110] In some embodiments, the multi-scale feature matrix corresponding to the device to be predicted can be used as the feature data of the device to be predicted and input into the pre-trained lifetime prediction model. The lifetime prediction model can then predict the lifetime information of the device to be predicted based on the multi-scale feature matrix.

[0111] Among them, lifetime information is used to indicate the remaining electrical lifetime of the device to be predicted.

[0112] In summary, the equipment life prediction method provided in this embodiment includes: collecting feature sequences of the equipment to be predicted based on electrical life test results, the feature sequences including feature parameter sets at continuous time steps; decomposing and reconstructing the feature sequences according to the feature sequences of the equipment to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the equipment to be predicted under each preset scale factor; determining the permutation entropy of the equipment to be predicted under each preset scale factor according to the sequence arrangement pattern of each feature subsequence of the equipment to be predicted under each preset scale factor, and determining the multi-scale feature matrix corresponding to the equipment to be predicted based on the permutation entropy under each preset scale factor; and predicting the life information of the equipment to be predicted using a pre-trained life prediction model based on the multi-scale feature matrix corresponding to the equipment to be predicted. This method captures the features of the feature sequences at different scale factors based on an improved multi-scale permutation entropy algorithm, calculates the permutation entropy of the features at each scale factor, and integrates the feature sequences based on the permutation entropy to obtain the multi-scale fusion features corresponding to the feature sequences, thereby improving the accuracy of the extracted features. Therefore, equipment life prediction based on the extracted features can improve the accuracy of the prediction results.

[0113] Figure 2 This is a flowchart illustrating another equipment lifetime prediction method provided in an embodiment of this application; optionally, in step S101, based on the electrical lifetime test results, collecting the feature sequence of the equipment to be predicted may include:

[0114] S201. Based on the electrical life test performed on the device to be predicted, collect the current signal and voltage signal generated by the device to be predicted during the electrical life test.

[0115] The electrical life test of DC contactors aims to evaluate their electrical life and performance stability by simulating actual working conditions and subjecting the contactors to periodic switching operations.

[0116] The DC contactor electrical life test bench is the key equipment for conducting this experiment. It typically consists of a high-voltage power supply, a photoelectric sampling system, a data acquisition system, and a control system. These systems work together to achieve precise control and monitoring of the DC contactor.

[0117] The general procedure of the experiment is as follows:

[0118] Connect the device: Connect the test device to the DC contactor under test, ensuring the connection is correct and secure.

[0119] Parameter settings: Set the voltage value of the high-voltage power supply, test cycle, and other parameters according to the experimental requirements.

[0120] Start the test: Turn on the power and control system of the test equipment to start the test program. During the test, the photoelectric sampling system will collect the switching status of the contactor in real time, and the data acquisition system will process and analyze this data.

[0121] Monitoring and Testing: Real-time monitoring of test results is necessary during the testing process to ensure the accuracy and reliability of the test data. Simultaneously, it is also crucial to observe the contactor's operating status to promptly identify and address any issues.

[0122] Data recording: After the test is completed, the test data is recorded and analyzed to evaluate the electrical life and performance stability of the DC contactor.

[0123] Taking the AC-3 experiment as an example, the experiment can be conducted through the experimental platform to obtain multiple data tables. Each data table can record the experimental data of the device to be predicted at one time step. The experimental data can include the voltage and current signals generated by the device to be predicted at the time step.

[0124] S202. Perform waveform analysis on the current and voltage signals to obtain the characteristic sequence of the device to be predicted.

[0125] By performing waveform analysis on the voltage and current signals generated at each time step, a set of characteristic parameters can be extracted from the experimental data at each time step based on waveform analysis software.

[0126] The time steps are continuous. Based on the experimental data at each time step, the feature sequence of the device to be predicted can be obtained. The feature sequence includes each time step arranged in chronological order and the feature parameter group at each time step.

[0127] Figure 3 This is a flowchart illustrating another equipment lifetime prediction method provided in this application embodiment; optionally, in step S102, the feature sequence is decomposed and reconstructed according to the feature sequence of the equipment to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the equipment to be predicted under each preset scale factor, which may include:

[0128] S301. Based on the target scale factor, determine each target time step to be extracted from the feature sequence.

[0129] This embodiment uses the generation method of feature subsequences under the target scale factor as an example for illustration. The target time steps to be extracted can be determined from the feature sequence according to the screening length indicated by the target scale factor.

[0130] Assuming the feature sequence contains 8 consecutive time steps: {1,2,3,4,5,6,7,8}, and the target scale factor is 2, then time steps 1, 3, 5, and 7 can be determined as the target time steps.

[0131] S302. Based on the sequence sorting pattern between each target time step and the feature parameter group under each target time step, generate each feature subsequence of the device to be predicted under the target scale factor.

[0132] The sorting method for each target time step in each feature subsequence is different.

[0133] There can be multiple sorting patterns among the target time steps identified above, and at least one feature subsequence can be generated under different sorting patterns. Thus, based on the sorting patterns of each sequence among the target time steps, various sorting combinations can be performed on each target time step, and a feature subsequence is generated under each sorting combination, thereby obtaining the feature subsequences of the device to be predicted under the target scale factor.

[0134] The sequence sorting mode can indicate the sorting method between time steps, such as sequential, reverse, or interleaved sorting, which are all different sorting methods.

[0135] In the cross-arrangement method, a variety of feature subsequences can be generated.

[0136] In another feasible approach, when reconstructing the feature sequence based on the target scale factor, instead of extracting time steps from the feature sequence, the feature subsequence corresponding to the target scale factor is calculated for each time step based on the target scale factor.

[0137] For example, assuming the feature sequence contains 8 consecutive time steps: {1,2,3,4,5,6,7,8}, and the target scale factor is 3, the feature parameter sets of time step 1, time step 2, and time step 3 can be averaged to obtain the new feature parameter set of time step 1, and the feature parameter sets of time step 4, time step 5, and time step 6 can be averaged to obtain the new feature parameter set of time step 2; or the feature parameter sets of time step 2, time step 3, and time step 4 can be averaged to obtain the new feature parameter set of time step 3. That is, the scale factor determines how many consecutive time steps need to be calculated.

[0138] Therefore, in this way, the feature sequence of the device to be predicted can be coarsened at multiple scales by each preset scale factor, so as to obtain at least one feature subsequence of the device to be predicted under each preset scale factor.

[0139] The multi-scale feature matrix corresponding to the device to be predicted is obtained by processing in the above way. Based on the multi-scale feature matrix, the dynamic features of the device to be predicted can be effectively mined, thereby improving the accuracy of subsequent predictions.

[0140] Figure 4 This is a flowchart illustrating another device lifetime prediction method provided in this application embodiment; optionally, in step S103, determining the arrangement entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor may include:

[0141] S401. Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under the target scale factor, determine the probability of occurrence of each sequence arrangement pattern under the target scale factor.

[0142] For the target scale factor, the extracted target time steps can be sorted in multiple sorting modes, and the number of occurrences of each sorting mode can be calculated to determine the probability of occurrence of each sorting mode.

[0143] For each sequence arrangement pattern, we need to calculate its probability of occurrence. This can be obtained by dividing the number of occurrences of each sequence arrangement pattern by the total number of occurrences.

[0144] S402. Determine the probability distribution of each sequence arrangement pattern under the target scale factor based on the probability of occurrence of each sequence arrangement pattern under the target scale factor.

[0145] Based on the probability of occurrence of each sequence arrangement pattern under the target scale factor, the probability distribution of each sequence arrangement pattern under the target scale factor can be calculated.

[0146] By compiling the probabilities of all sequence permutations into a list or table, we can obtain the probability distribution of each sequence permutation under the target scale factor.

[0147] S403. Determine the permutation entropy under the target scale factor based on the probability distribution of each sequence permutation pattern under the target scale factor.

[0148] Assume that the probability distribution of each sequence arrangement pattern under the target scaling factor is P1, P2, P3...Pk;

[0149] For each sequence permutation pattern, calculate its log probability log2(Pi), multiply the probability of each sequence permutation pattern by its log probability, and sum over all sequence permutation patterns to obtain the permutation entropy under the target scale factor.

[0150] Optionally, in step S103, determining the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor may include: integrating each feature subsequence under each preset scale factor based on the permutation entropy under each preset scale factor to obtain the multi-scale feature matrix corresponding to the device to be predicted.

[0151] In one implementation, the multi-scale feature matrix corresponding to the device to be predicted can be obtained by simply averaging the feature subsequences under each preset scale factor based on the permutation entropy under each preset scale factor.

[0152] In another feasible approach, a weighted average can be calculated for each feature subsequence under each preset scale factor based on the importance or weight of each preset scale. This method can better reflect the contribution of each scale, but the weight values ​​need to be determined in advance.

[0153] The above method can effectively capture the dynamic features of feature sequences at different scale factors, thereby improving the comprehensiveness and accuracy of feature extraction. The extracted multi-scale feature matrix can be directly used as input data into the lifetime prediction model for lifetime prediction.

[0154] Figure 5 This is a flowchart illustrating another equipment lifetime prediction method provided in an embodiment of this application; optionally, in step S104, predicting the lifetime information of the equipment to be predicted using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the equipment to be predicted may include:

[0155] S501. Input the multi-scale feature matrix into the lifetime prediction model. The lifetime prediction model determines the calculation weights corresponding to each time step in the multi-scale feature matrix based on the multi-scale feature matrix.

[0156] The lifetime prediction model in this embodiment can adopt the Informer model framework. The Informer model consists of an encoder, a decoder, and a fully connected layer. The encoder converts the input sequence into a series of coded representations, and the decoder generates the output sequence based on the coded representations.

[0157] The encoder accepts the input sequence and maps it to a high-dimensional vector, then feeds it to the decoder to generate the output sequence. Before the input sequence enters the encoder, time information is embedded into the sequence through sine and cosine functions to preserve the position and time information in the sequence data.

[0158] Optionally, in this embodiment, the extracted multi-scale feature matrix is ​​input into the lifetime prediction model. The lifetime prediction model calculates the attention weight matrix corresponding to the multi-scale feature matrix based on the multi-scale feature matrix, thereby determining the calculation weight corresponding to each time step in the multi-scale feature matrix based on the attention weight matrix.

[0159] The computational weights corresponding to each time step determine the computational weights of the feature parameter groups at each time step when participating in the calculation of the prediction results.

[0160] S502. Based on the feature parameter groups of each time step in the multi-scale feature matrix and the calculation weights corresponding to each time step, perform weighted calculations to predict the lifespan information of the device to be predicted.

[0161] Lifetime information is used to indicate the remaining number of uses of the device to be predicted.

[0162] Optionally, based on the feature parameter groups of each time step in the multi-scale feature matrix and the corresponding calculation weights of each time step, the feature parameter groups of each time step can be weighted and calculated, and the lifetime information of the device to be predicted can be obtained based on the calculation results.

[0163] In some embodiments, lifetime information can be characterized as a percentage, which indicates the percentage of remaining usage times of the device to be predicted relative to the total available usage times. Based on the percentage and the total available usage times at full lifetime, the remaining electrical lifetime, i.e., the remaining usage times, can be calculated.

[0164] In other embodiments, lifetime information can be represented by data that directly indicates the remaining number of uses of the device to be predicted, without the need for percentage-based conversion calculations.

[0165] The specific lifetime information is represented in the same way as the lifetime label representation in the sample data during model training.

[0166] Figure 6 A flowchart illustrating a training method for a lifespan prediction model provided in this application embodiment; optionally, the training process of the lifespan prediction model may include:

[0167] S601. Collect sample feature sequences from multiple sample devices. The sample feature sequences contain a set of feature parameters of the sample devices at consecutive sample time steps, wherein each sample time step is marked with a lifetime label.

[0168] Optionally, sample feature sequences of multiple sample devices can be collected, with each sample device being a sample DC contactor. By conducting electrical lifetime experiments on each sample device, sample feature sequences of each sample device can be collected. Similar to the feature sequences of the device to be predicted, the sample feature sequences of the sample devices also contain a set of characteristic parameters of the sample device at consecutive sample time steps. In addition, each sample time step is marked with a lifetime label.

[0169] It is worth noting that during the electrical lifetime test of the sample equipment, the remaining number of uses of the sample equipment decreases continuously over time. Therefore, the remaining number of uses indicated by the lifetime labels of each sample time step arranged in sequence in the sample feature sequence is continuously decreasing.

[0170] S602. Based on the sample feature sequences of each sample device and multiple sample scale factors, decompose and reconstruct the sample feature sequences of each sample device to obtain at least one sample feature subsequence of each sample device under each sample scale factor.

[0171] Similar to the above processing, the sample feature sequences of each sample device can be decomposed and reconstructed according to the sample scale factor to obtain at least one sample feature subsequence of the sample device under each sample scale factor.

[0172] The specific decomposition and reconstruction process will not be detailed here.

[0173] S603. Based on the sequence arrangement pattern of each sample feature subsequence of each sample device under each sample scale factor, determine the sample multi-scale feature matrix corresponding to each sample device.

[0174] Similarly, based on the sequence arrangement pattern of each sample feature subsequence of each sample device under each sample scale factor, the arrangement entropy of each sample device under each sample scale factor can be determined. Then, based on the arrangement entropy of each sample device under each sample scale factor, the sample feature subsequences under each sample scale factor are integrated to obtain the sample multi-scale feature matrix corresponding to each sample device.

[0175] S604. Using the multi-scale feature matrix of each sample device, a lifetime prediction model is trained.

[0176] Optionally, the multi-scale feature matrix of each sample device can be used as training sample data and input into the Informer model to train a lifetime prediction model.

[0177] It is worth noting that in some embodiments, the sample feature sequences of each sample device can be directly input into the model as input data. The model processes the sample feature sequences based on the multi-scale permutation entropy algorithm to obtain the corresponding multi-scale feature matrix. The model then trains the network parameters based on the multi-scale feature matrix.

[0178] Optionally, in step S602, before decomposing and reconstructing the sample feature sequences of each sample device according to the sample feature sequences of each sample device and multiple sample scale factors to obtain at least one sample feature subsequence of each sample device under each sample scale factor, the method further includes: performing data normalization processing on the sample feature sequences of each sample device, scaling each sample feature parameter in each sample feature sequence to a preset data range, and obtaining the processed sample feature sequences.

[0179] In some embodiments, the sample feature sequences may be preprocessed before processing to accelerate model training and enhance model robustness.

[0180] Optionally, a data normalization method can be used to scale the values ​​of each feature parameter in the feature parameter group at each time step in the sample feature sequence to a new data range, typically between -1 and 1.

[0181] The implementation steps of the data normalization method are roughly as follows:

[0182] 1) Calculate the statistic:

[0183] Median (Q2): The middle value of the data, that is, the value that is in the middle position after the data is sorted.

[0184] First quartile (Q1): Dividing the data into one-quarters, Q1 is the smallest one-quarter value.

[0185] Third quartile (Q3): Dividing the data into quartiles, Q3 is the largest quartile value.

[0186] 2) Perform data scaling:

[0187] Use the following formula to normalize the original data to a new data range:

[0188] Xnorm = (X - Q2) / (Q3 - Q1)

[0189] Where X refers to the original data, and Xnorm refers to the normalized data.

[0190] In electrical life tests, extreme values ​​and outliers are often generated due to noise. Data normalization is robust to outliers, meaning that even when extreme values ​​or outliers exist in the data, it can still effectively scale the data without being affected by the outliers.

[0191] In some embodiments, data standardization can be used instead of data normalization to effectively handle anomalous outlier data.

[0192] Optionally, in step S604, the lifetime prediction model is trained using the sample multi-scale feature matrix corresponding to each sample device. This may include: inputting the sample multi-scale feature matrix corresponding to each sample device into the initial lifetime prediction model, and iteratively correcting the model weights of the initial lifetime prediction model based on the sample multi-scale feature matrix to obtain the lifetime prediction model; the model weights are used to indicate the calculated weights corresponding to each time step in the feature sequence of the input model.

[0193] In some embodiments, after inputting the sample multi-scale feature matrix corresponding to each sample device into the initial lifetime prediction model, the initial lifetime prediction model can perform model prediction and parameter training processes based on the sample multi-scale feature matrix to continuously iterate and correct the model weights of the initial lifetime prediction model. The model weights are used to indicate the calculation weights of each time step in the input data input into the model.

[0194] The initial lifetime prediction model refers to the lifetime prediction model before the network parameters are trained. By training the model weights of the initial lifetime prediction model, the initial lifetime prediction model with the trained model weights can be used as the desired lifetime prediction model.

[0195] Figure 7 A flowchart illustrating another training method for a lifetime prediction model provided in this application embodiment; optionally, the step of iteratively correcting the model weights of the initial lifetime prediction model based on the multi-scale feature matrix of each sample may include:

[0196] S701. Based on the multi-scale feature matrix of each sample, determine the positional relationship between the time steps of each sample in the multi-scale feature matrix of each sample.

[0197] In some embodiments, in order to preserve the positional relationship between time steps of each sample, time information can first be embedded into the multi-scale feature matrix of each sample using sine and cosine functions.

[0198] Sine and Cosine Functions: The basic idea of ​​time embedding is to use sine and cosine functions to generate fixed periodic signals that change with increasing time steps. Specifically, for each time step t, the following time embedding vector can be calculated:

[0199]

[0200] In the formula: t is the index of the time step, i is the dimension index of the time embedding, and d model It is the number of hidden units in the model (usually corresponding to the dimension of the encoder output).

[0201] Function: The temporal embedding vector is added to the input representation, so that the encoding of each time step depends not only on its numerical value, but also on information about its position and time. This helps the model better understand the context and order of different time points in the sequence.

[0202] Therefore, based on the multi-scale feature matrix of each sample, the temporal embedding vector of each sample time step can be extracted to determine the positional relationship between each sample time step.

[0203] S702. Based on the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample and the feature parameter set of each time step of each sample in the multi-scale feature matrix of each sample, predict the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step.

[0204] Next, the encoder uses self-attention to learn the dependencies and importance between time steps of each sample. Self-attention allows the model to consider information from all other time steps in the sequence when processing each time step, thus better capturing the long-term dependencies of the time series.

[0205] It's worth noting that the self-attention mechanism used in this embodiment can be the ProbSparse self-attention mechanism. The ProbSparse mechanism introduces sparsity, meaning that significant attention connections exist only between a few positions. This significantly reduces computational cost, especially important when processing long sequences. The ProbSparse mechanism doesn't randomly select attention connections; instead, it determines whether attention weights are zero based on a probability distribution (such as a normal distribution). This probabilistic selection can maintain the model's expressive power to some extent while reducing unnecessary computational overhead. By reducing unnecessary connections and computations, the ProbSparse mechanism focuses more on important dependencies, thus more effectively capturing associations between long sequences.

[0206] Optionally, based on the dependencies between each sample time step and the characteristic parameters in each sample time step, the predicted lifetime of each sample device under the characteristic parameter set indicated by each sample time step can be predicted.

[0207] S703. Based on the predicted lifetime of each sample device under the characteristic parameter group indicated by each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device, iteratively correct the model weights of the initial lifetime prediction model.

[0208] Each sample time step of each sample device is also marked with a lifetime label. Therefore, based on the predicted lifetime of each sample device under the characteristic parameter group indicated by each sample time step and the lifetime label corresponding to each sample device at each sample time step, the model weights of the initial lifetime prediction model can be iteratively corrected.

[0209] Figure 8 This is a flowchart illustrating another training method for a lifetime prediction model provided in this application embodiment; optionally, in step S703, the model weights of the initial lifetime prediction model are iteratively corrected based on the predicted lifetime of each sample device under the feature parameter group indicated by each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device, which may include:

[0210] S801. Calculate the current loss value of the initial lifetime prediction model based on the predicted lifetime of each sample device under the characteristic parameter group indicated at each sample time step and the lifetime label corresponding to each sample time step of each sample device.

[0211] In some embodiments, during each round of iterative training, the loss value of the initial lifetime prediction model in the current round can be calculated based on the predicted lifetime of each sample device under the feature parameter group indicated by each sample time step and the lifetime label corresponding to each sample time step of each sample device calculated in the current round.

[0212] S802. If the current loss value is greater than the preset loss threshold, the current model weights of the initial lifetime prediction model are iteratively corrected until the current loss value is less than or equal to the preset loss threshold. Then, the correction is stopped, and the model weights at the time of stopping the correction are used as the model weights of the initial lifetime prediction model to obtain the lifetime prediction model.

[0213] If the current loss value is greater than the preset loss threshold, the model is considered to have a large loss. In this case, the current model weights of the initial lifetime prediction model can be corrected. Based on the corrected initial lifetime prediction model, the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step in the new round is recalculated. The loss value of the initial lifetime prediction model in the new round is calculated according to the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step and the lifetime label corresponding to each sample device at each sample time step. Training stops when the loss value meets the preset loss threshold. The current model weights of the initial lifetime prediction model are used as the target model weights, and the initial lifetime prediction model with the target model weights is used as the trained lifetime prediction model.

[0214] In other implementations, the stopping condition for model training can be determined by setting the number of iterations. Training can be stopped when the preset number of iterations is reached.

[0215] In summary, the equipment life prediction method provided in this embodiment includes: collecting feature sequences of the equipment to be predicted based on electrical life test results, the feature sequences including feature parameter sets at continuous time steps; decomposing and reconstructing the feature sequences according to the feature sequences of the equipment to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the equipment to be predicted under each preset scale factor; determining the permutation entropy of the equipment to be predicted under each preset scale factor according to the sequence arrangement pattern of each feature subsequence of the equipment to be predicted under each preset scale factor, and determining the multi-scale feature matrix corresponding to the equipment to be predicted based on the permutation entropy under each preset scale factor; and predicting the life information of the equipment to be predicted using a pre-trained life prediction model based on the multi-scale feature matrix corresponding to the equipment to be predicted. This method captures the features of the feature sequences at different scale factors based on an improved multi-scale permutation entropy algorithm, calculates the permutation entropy of the features at each scale factor, and integrates the feature sequences based on the permutation entropy to obtain the multi-scale fusion features corresponding to the feature sequences, thereby improving the accuracy of the extracted features. Therefore, equipment life prediction based on the extracted features can improve the accuracy of the prediction results.

[0216] In addition, training a lifetime prediction model based on the Informer model framework can significantly improve the model's computational efficiency and its ability to capture long-term series dependencies, thereby improving the accuracy of the prediction results.

[0217] The following describes the apparatus, equipment, and storage medium used to implement the device lifetime prediction method provided in this application. The specific implementation process and technical effects are described above and will not be repeated below.

[0218] Figure 9This is a schematic diagram of a device for predicting the lifespan of an object according to an embodiment of this application. The function implemented by the device corresponds to the steps performed by the method described above. The device can be understood as the server or the processor of the server, or as a component that implements the function of this application under the control of the server, independent of the server or the processor. Optionally, the device may include: a data acquisition module 110, a processing module 120, a determination module 130, and a prediction module 140.

[0219] The acquisition module 110 is used to acquire the characteristic sequence of the device to be predicted based on the electrical life test results. The characteristic sequence includes a set of characteristic parameters under continuous time steps. Each set of characteristic parameters includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted.

[0220] The processing module 120 is used to decompose and reconstruct the feature sequence based on the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the device to be predicted under each preset scale factor.

[0221] The determination module 130 is used to determine the permutation entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, and to determine the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor.

[0222] The prediction module 140 is used to predict the lifetime information of the device to be predicted by using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted.

[0223] Optionally, the acquisition module 110 is specifically used to acquire the current signal and voltage signal generated by the device under test during the electrical life test based on the electrical life test performed on the device under test;

[0224] Waveform analysis of current and voltage signals yields the characteristic sequence of the device to be predicted.

[0225] Optionally, the processing module 120 is specifically used to determine each target time step to be extracted from the feature sequence based on the target scale factor;

[0226] Based on the sequence sorting patterns between each target time step and the feature parameter set under each target time step, feature subsequences of the device to be predicted under the target scale factor are generated. Among them, the sorting method of each target time step in each feature subsequence is different.

[0227] Optionally, the determining module 130 is specifically used to determine the probability of occurrence of each sequence arrangement pattern under the target scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under the target scale factor.

[0228] Based on the probability of occurrence of each sequence arrangement pattern under the target scale factor, determine the probability distribution of each sequence arrangement pattern under the target scale factor;

[0229] The permutation entropy under the target scale factor is determined based on the probability distribution of each sequence permutation pattern under the target scale factor.

[0230] Optionally, the determining module 130 is specifically used to integrate the feature subsequences under each preset scale factor according to the permutation entropy under each preset scale factor to obtain the multi-scale feature matrix corresponding to the device to be predicted.

[0231] Optionally, the prediction module 140 is specifically used to input the multi-scale feature matrix into the lifetime prediction model, and the lifetime prediction model determines the calculation weights corresponding to each time step in the multi-scale feature matrix based on the multi-scale feature matrix.

[0232] Based on the feature parameter groups of each time step in the multi-scale feature matrix and the corresponding calculation weights of each time step, a weighted calculation is performed to predict the lifetime information of the device to be predicted; the lifetime information is used to indicate the remaining number of times the device to be predicted can be used.

[0233] Optionally, it may also include: a training module;

[0234] The training module is used to collect sample feature sequences from multiple sample devices. The sample feature sequences contain feature parameter sets of the sample devices at consecutive sample time steps, where each sample time step is labeled with a lifetime tag.

[0235] Based on the sample feature sequences of each sample device and multiple sample scale factors, the sample feature sequences of each sample device are decomposed and reconstructed to obtain at least one sample feature subsequence of each sample device under each sample scale factor.

[0236] Based on the sequence arrangement pattern of each sample feature subsequence under each sample scale factor, determine the sample multi-scale feature matrix corresponding to each sample device;

[0237] The lifetime prediction model is trained using the multi-scale feature matrix corresponding to each sample device.

[0238] Optionally, the training module is specifically used to perform data normalization processing on the sample feature sequences of each sample device, scaling the sample feature parameters in each sample feature sequence to a preset data range, so as to obtain the processed sample feature sequences.

[0239] Optionally, the training module is specifically used to input the multi-scale feature matrix of each sample device into the initial lifetime prediction model. The initial lifetime prediction model iteratively corrects the model weights of the initial lifetime prediction model based on the multi-scale feature matrix of each sample to obtain the lifetime prediction model. The model weights are used to indicate the calculated weights corresponding to each time step in the feature sequence of the input model.

[0240] Optionally, the training module is specifically used to determine the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample based on the multi-scale feature matrix of each sample.

[0241] Based on the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample and the feature parameter set of each time step of each sample in the multi-scale feature matrix of each sample, the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step is predicted.

[0242] Based on the predicted lifetime of each sample device under the characteristic parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device, the model weights of the initial lifetime prediction model are iteratively corrected.

[0243] Optionally, the training module is specifically used to calculate the current loss value of the initial lifetime prediction model based on the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step and the lifetime label corresponding to each sample time step of each sample device.

[0244] If the current loss value is greater than the preset loss threshold, the current model weights of the initial lifetime prediction model are iteratively corrected until the current loss value is less than or equal to the preset loss threshold. Then, the correction is stopped, and the model weights at the point where the correction stops are used as the model weights of the initial lifetime prediction model to obtain the lifetime prediction model.

[0245] The above-described device is used to execute the method provided in the foregoing embodiments, and its implementation principle and technical effect are similar, so they will not be described again here.

[0246] These modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more digital signal processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). Alternatively, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a system-on-a-chip (SOC).

[0247] The modules described above can be connected or communicate with each other via wired or wireless connections. Wired connections can include metal cables, optical fibers, hybrid cables, or any combination thereof. Wireless connections can include connections via LAN, WAN, Bluetooth, ZigBee, or NFC, or any combination thereof. Two or more modules can be combined into a single module, and any module can be divided into two or more units. Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems and devices described above can be referred to the corresponding processes in the method embodiments, and will not be repeated here.

[0248] Figure 10 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. The device may be a computing device with data processing capabilities.

[0249] The device includes: a processor 801 and a storage medium 802.

[0250] Storage medium 802 is used to store programs, and processor 801 calls the programs stored in storage medium 802 to execute the above method embodiments. The specific implementation and technical effects are similar, and will not be described in detail here.

[0251] The storage medium 802 stores program code, which, when executed by the processor 801, causes the processor 801 to perform various steps in the device lifetime prediction method according to various exemplary embodiments of this application as described in the "Exemplary Methods" section above.

[0252] The processor 801 can be a general-purpose processor, such as a central processing unit (CPU), digital signal processor (DSP), application-specific integrated circuit (ASIC), field-programmable gate array (FPGA), or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component, capable of implementing or executing the methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor.

[0253] Storage medium 802, as a non-volatile computer-readable storage medium, can be used to store non-volatile software programs, non-volatile computer-executable programs, and modules. The storage medium can include at least one type of storage medium, such as flash memory, hard disk, multimedia card, card-type storage medium, random access memory (RAM), static random access memory (SRAM), programmable read-only memory (PROM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), magnetic storage medium, magnetic disk, optical disk, etc. The storage medium is any other medium capable of carrying or storing desired program code in the form of instructions or data structures that can be accessed by a computer, but is not limited thereto. In the embodiments of this application, storage medium 802 can also be a circuit or any other device capable of implementing storage functions for storing program instructions and / or data.

[0254] Optionally, this application also provides a program product, such as a computer-readable storage medium, including a program that, when executed by a processor, performs the above-described method embodiments.

[0255] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0256] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0257] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0258] The integrated units implemented as software functional units described above can be stored in a computer-readable storage medium. These software functional units, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for predicting equipment lifespan, characterized in that, include: Based on the electrical life test results, a characteristic sequence of the device to be predicted is collected. The characteristic sequence includes a set of characteristic parameters under continuous time steps. Each set of characteristic parameters includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted. Based on the feature sequence of the device to be predicted and multiple preset scale factors, the feature sequence is decomposed and reconstructed to obtain at least one feature subsequence of the device to be predicted under each preset scale factor. Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, the arrangement entropy of the device to be predicted under each preset scale factor is determined, and based on the arrangement entropy under each preset scale factor, the multi-scale feature matrix corresponding to the device to be predicted is determined. Based on the multi-scale feature matrix corresponding to the device to be predicted, a pre-trained lifetime prediction model is used to predict the lifetime information of the device to be predicted.

2. The method according to claim 1, characterized in that, The feature sequence of the device to be predicted, based on the electrical lifetime test results, includes: Based on the electrical lifetime test performed on the device to be predicted, the current signal and voltage signal generated by the device to be predicted during the electrical lifetime test are collected; Waveform analysis is performed on the current and voltage signals to obtain the characteristic sequence of the device to be predicted.

3. The method according to claim 1, characterized in that, The step of decomposing and reconstructing the feature sequence based on the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature sub-sequence of the device to be predicted under each preset scale factor includes: Based on the target scale factor, determine each target time step to be extracted from the feature sequence; Based on the sequence sorting pattern between each target time step and the feature parameter group under each target time step, the feature subsequences of the device to be predicted under the target scale factor are generated, wherein the sorting method of each target time step in each feature subsequence is different.

4. The method according to claim 3, characterized in that, The step of determining the permutation entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor includes: Based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under the target scale factor, determine the probability of occurrence of each sequence arrangement pattern under the target scale factor; Based on the occurrence probability of each sequence arrangement pattern under the target scale factor, determine the probability distribution of each sequence arrangement pattern under the target scale factor; The permutation entropy under the target scale factor is determined based on the probability distribution of each sequence permutation pattern under the target scale factor.

5. The method according to claim 1, characterized in that, The step of determining the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor includes: The feature subsequences under each preset scale factor are integrated based on the permutation entropy under each preset scale factor to obtain the multi-scale feature matrix corresponding to the device to be predicted.

6. The method according to claim 1, characterized in that, The step of predicting the lifetime information of the device to be predicted using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted includes: The multi-scale feature matrix is ​​input into the lifetime prediction model, and the lifetime prediction model determines the calculation weights corresponding to each time step in the multi-scale feature matrix based on the multi-scale feature matrix. Based on the feature parameter groups of each time step in the multi-scale feature matrix and the calculation weights corresponding to each time step, a weighted calculation is performed to predict the lifetime information of the device to be predicted; the lifetime information is used to indicate the remaining number of times the device to be predicted can be used.

7. The method according to any one of claims 1-6, characterized in that, The training process of the lifespan prediction model includes: Collect sample feature sequences from multiple sample devices. The sample feature sequences contain a set of feature parameters of the sample devices at consecutive sample time steps, where each sample time step is labeled with a lifetime tag. Based on the sample feature sequences of each sample device and multiple sample scale factors, the sample feature sequences of each sample device are decomposed and reconstructed to obtain at least one sample feature subsequence of each sample device under each sample scale factor. Based on the sequence arrangement pattern of each sample feature subsequence under each sample scale factor, determine the sample multi-scale feature matrix corresponding to each sample device; The lifetime prediction model is trained using the multi-scale feature matrix corresponding to each sample device.

8. The method according to claim 7, characterized in that, Before the step of decomposing and reconstructing the sample feature sequences of each sample device based on the sample feature sequences of each sample device and multiple sample scale factors to obtain at least one sample feature subsequence of each sample device under each sample scale factor, the method further includes: Data normalization is performed on the sample feature sequences of each sample device, and the sample feature parameters in each sample feature sequence are scaled to a preset data range to obtain the processed sample feature sequences.

9. The method according to claim 7, characterized in that, The lifespan prediction model is trained using the multi-scale feature matrix corresponding to each sample device, including: The sample multi-scale feature matrix corresponding to each sample device is input into the initial lifetime prediction model. The initial lifetime prediction model iteratively corrects the model weights of the initial lifetime prediction model based on the sample multi-scale feature matrix to obtain the lifetime prediction model. The model weights are used to indicate the calculated weights corresponding to each time step in the feature sequence of the input model.

10. The method according to claim 9, characterized in that, The step of iteratively correcting the model weights of the initial lifetime prediction model based on the multi-scale feature matrix of each sample includes: Based on the multi-scale feature matrix of each sample, determine the positional relationship between the time steps of each sample in the multi-scale feature matrix of each sample; Based on the positional relationship between time steps of each sample in the multi-scale feature matrix of each sample and the feature parameter set of each time step of each sample in the multi-scale feature matrix of each sample, the predicted lifetime of each sample device under the feature parameter set indicated by each sample time step is predicted. The model weights of the initial lifetime prediction model are iteratively corrected based on the predicted lifetime of each sample device under the characteristic parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the sample multi-scale feature matrix of each sample device.

11. The method according to claim 10, characterized in that, The step of iteratively refining the model weights of the initial lifetime prediction model based on the predicted lifetime of each sample device under the feature parameter set indicated at each sample time step and the lifetime label corresponding to each sample time step in the multi-scale feature matrix of each sample device includes: The current loss value of the initial lifetime prediction model is calculated based on the predicted lifetime of each sample device under the characteristic parameter group indicated at each sample time step and the lifetime label corresponding to each sample time step of each sample device. If the current loss value is greater than a preset loss threshold, the current model weights of the initial lifetime prediction model are iteratively corrected until the current loss value is less than or equal to the preset loss threshold. Then, the correction is stopped, and the model weights at the point where the correction stops are used as the model weights of the initial lifetime prediction model to obtain the lifetime prediction model.

12. A device for predicting equipment lifespan, characterized in that, include: The module includes a data acquisition module, a processing module, a determination module, and a prediction module. The acquisition module is used to acquire the feature sequence of the device to be predicted based on the electrical life test results. The feature sequence includes a set of feature parameters under continuous time steps. Each set of feature parameters includes at least: the arcing time of the device to be predicted, the contact resistance of the device to be predicted, and the arcing energy of the device to be predicted. The processing module is used to decompose and reconstruct the feature sequence based on the feature sequence of the device to be predicted and multiple preset scale factors to obtain at least one feature subsequence of the device to be predicted under each preset scale factor. The determining module is used to determine the permutation entropy of the device to be predicted under each preset scale factor based on the sequence arrangement pattern of each feature subsequence of the device to be predicted under each preset scale factor, and to determine the multi-scale feature matrix corresponding to the device to be predicted based on the permutation entropy under each preset scale factor. The prediction module is used to predict the lifetime information of the device to be predicted by using a pre-trained lifetime prediction model based on the multi-scale feature matrix corresponding to the device to be predicted.

13. An electronic device, characterized in that, include: The device includes a processor, a storage medium, and a bus, wherein the storage medium stores program instructions executable by the processor, and when the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the program instructions to implement the device lifetime prediction method as described in any one of claims 1 to 11.

14. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that is executed by a processor to implement the device lifetime prediction method as described in any one of claims 1 to 11.

Images