Secondary equipment reliability evaluation method

Abstract

The application provides a secondary equipment reliability evaluation method, and belongs to the technical field of secondary equipment evaluation. The method comprises the following steps: constructing a secondary equipment reliability evaluation integrated model, the integrated model comprising multiple parallel individual discriminators, the individual discriminators being trained by using operation data of normal secondary equipment, and the optimization target of the individual discriminators being to minimize reconstruction error of input and output data; optimizing the secondary equipment reliability evaluation integrated model, eliminating the individual discriminators with low accuracy, and obtaining an optimal integrated model; and inputting current operation data of the secondary equipment into the optimal integrated model for reliability evaluation. The application generates multiple parallel basic models by using a Bagging integrated learning method, comprehensively determines the results of the multiple individual discriminators to determine the secondary equipment fault, and can improve the stability of the model as a whole. Moreover, the application optimizes the integrated model by using an integrated model optimization method, and can improve the accuracy of the secondary equipment reliability evaluation.

Description

Technical Field

[0001] This invention belongs to the field of secondary equipment evaluation technology, specifically relating to a secondary equipment reliability evaluation method based on reconstruction error calculation and ensemble learning. Background Technology

[0002] Secondary equipment plays a crucial role in power systems, functioning as voltage transformers, power distributors, and voltage controllers. It is a vital component of the power system, and its safe and reliable operation is essential for the system's safety and stability. Therefore, precise fault diagnosis and reliability analysis of secondary equipment are necessary to facilitate accurate and scientific power emergency response and support. Traditionally, maintenance personnel conduct regular inspections of secondary equipment to promptly identify potential faults. However, with the expansion of power grids, the number of secondary equipment has increased exponentially, significantly increasing the workload for maintenance personnel and also raising the risks to power grid operation.

[0003] With the widespread application of intelligent secondary equipment in smart substations, power systems are able to acquire and store detailed operating data of intelligent secondary equipment. How to diagnose whether there are faults by analyzing the historical and current operating data of secondary equipment is an urgent technical problem to be solved. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a secondary equipment reliability assessment method based on reconstruction error calculation and ensemble learning, in order to achieve high-precision assessment of secondary equipment reliability, in order to address the shortcomings of the prior art.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A method for assessing the reliability of secondary equipment, comprising:

[0007] An integrated model for reliability assessment of secondary equipment is constructed. The integrated model includes multiple parallel individual discriminators. The individual discriminators are trained using the operating data of normal secondary equipment. The optimization objective of the individual discriminators training is to minimize the reconstruction error of the input and output data.

[0008] Optimize the integrated model for secondary equipment reliability assessment, remove the individual discriminators with low accuracy, and obtain the optimal integrated model;

[0009] The current operating data of the secondary equipment is input into the optimal integrated model for reliability assessment.

[0010] Furthermore, the integrated model combines the predictions of all the individual discriminators to obtain the evaluation result.

[0011] Furthermore, the evaluation result of the ensemble model is expressed as follows: Where n is the number of the individual discriminators, y k This is the predicted value of the k-th individual discriminator.

[0012] Furthermore, the optimization objective of the individual discriminator training is: Where X in For input secondary equipment operation data, Reconstruct the data at the output end, where L is X. in and The reconstruction error, m is X in and Dimension, x in,, For X in The element value at time t in the i-th dimension, x out,i,t for The element value at time t in the i-th dimension.

[0013] Furthermore, the judgment rule of the individual discriminator is as follows: Where X i in This indicates the input from the secondary device to be evaluated. σ represents the output of the secondary device to be judged, and σ represents the threshold for determining whether a potential fault has occurred.

[0014] Furthermore, the operating data of the secondary equipment are normalized using the Z-score method before being input.

[0015] Furthermore, the Z-score method is expressed as: Where μ is the mean of all sample data, σ ′ Let X be the standard deviation of all sample data, and X be the element value of an individual.

[0016] Furthermore, the input vectors of the individual discriminator all contain time series data.

[0017] Secondary systems are a crucial line of defense for ensuring the safe and stable operation of the power grid. The National Energy Administration, in its "Several Provisions on the Safety Management of Power Secondary Systems," stipulates that power companies and related power users should establish a dual prevention system for secondary system safety and strengthen the management of secondary system safety risks and the investigation and rectification of potential hazards. However, with the large-scale development of new energy sources and their high proportion of grid connection, the types of power secondary systems are constantly increasing, and their safe and stable operation has been affected to some extent. Accidents involving injuries or fatalities caused by backfeeding from secondary equipment have already occurred. Therefore, quantifying the failure probability and consequences of secondary equipment is urgently needed.

[0018] Currently, the mainstream methods for addressing the condition assessment of secondary equipment fall into three categories: secondary system evaluation guidelines for power system enterprises, condition assessment models evolved from primary equipment, and data-driven secondary system evaluation methods.

[0019] The secondary equipment evaluation guidelines for power system enterprises select status assessment information from aspects such as the completeness of pre-commissioning data, historical fault records, maintenance records, and real-time monitoring records. The degree of impact of these information on the equipment status is determined by an expert system. Although this method is simple and practical, its accuracy is insufficient. Furthermore, the expert system's judgment relies on its recorded experience and knowledge, thus requiring a lot of complex and profound professional knowledge and long-term accumulated experience, which is beyond the scope of general technical personnel and therefore difficult to operate. In addition, the expert system updates slowly, making it difficult to adapt to a wide variety of secondary equipment, and its judgment results cannot be accurately quantified.

[0020] State evaluation models evolved from primary systems can, to some extent, avoid the influence of subjective human factors on evaluation results. However, the indicators they utilize, such as mean time between failures (MTBF) and component availability, cannot fully reflect the equipment's state. For example, patent document CN102723775A proposes a reliability assessment method for secondary systems in intelligent substations. This method summarizes secondary equipment in the secondary system into different components constituting various subsystems and uses a reliability model to obtain the availability of each component. However, when calculating availability, it only uses parameters such as the component's mean time between failures (MTBF), mean time to repair (MTBT), and the average time between two failures, without considering the operating parameters of the secondary equipment. Therefore, the obtained availability cannot fully reflect the true state of the equipment.

[0021] Data-driven secondary system evaluation methods mainly model the large amount of historical data stored in the system to achieve secondary system fault early warning and anomaly monitoring. However, different devices have significant differences in historical data and evaluation indicators, so the existing data-driven methods have poor universality.

[0022] With the rise of intelligent secondary equipment, a large amount of information reflecting the status of secondary equipment has emerged. To make full use of this information to address the problem of rapid diagnosis of secondary circuit faults, it is necessary to establish a data-driven, generalized, and quantitative evaluation method for secondary equipment.

[0023] To address the issue of secondary equipment condition assessment based on historical data, patent document CN110580578A proposes a multi-layered evaluation method for the operational quality of secondary systems in intelligent substations. This method uses rough set theory to establish an incomplete evaluation system for secondary equipment in intelligent substations, establishes membership function values ​​for different degradation levels of secondary equipment, and uses fuzzy comprehensive evaluation to assess the condition of secondary equipment in intelligent substations. However, this method does not fully utilize historical operating data of secondary equipment, sets relatively coarse evaluation levels, and lacks sufficient precision in reliability quantification.

[0024] However, the operational data of secondary equipment is diverse and the correlation between data is weak. In this case, neural network technology has great application potential in the reliability assessment of secondary equipment in smart substations. All neurons in a neural network (generally referring to a fully connected neural network) are interconnected between layers and have strong nonlinear fitting capabilities, capable of fitting almost any function. Therefore, it is suitable for extracting, analyzing, and quantifying the acquired data from secondary equipment.

[0025] However, the training results of neural networks are prone to instability. To address this issue, this invention utilizes the Bagging ensemble learning method to generate multiple parallel base models (individual discriminators). The judgment results of these multiple base models are then combined to determine secondary equipment faults, thus improving the overall stability of the model. Furthermore, this invention optimizes the ensemble model using an ensemble model optimization method, eliminating base models with low accuracy to improve the accuracy of secondary equipment reliability assessment.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0027] This invention proposes a secondary equipment reliability assessment method based on reconstruction error calculation and ensemble learning. Leveraging the strong generalization ability and reduced error probability of ensemble models, a secondary equipment fault early warning ensemble model is constructed. The ensemble model uses the Bagging ensemble learning method to generate multiple parallel individual discriminators and employs voting merging to make predictions, thereby improving the overall stability and generalization ability of the model and reducing the error probability. Furthermore, to ensure the stability of the individual discriminators, this invention also employs ensemble learning to construct a thermal runaway alarm framework.

[0028] To improve the accuracy of ensemble model evaluation, this invention further optimizes the ensemble model using an ensemble model optimization method. The optimal ensemble model contains only individual discriminators with high accuracy, eliminating those with low accuracy. Specifically, during optimization, error-split theory is used to decompose the error of the ensemble model onto the features of the individual discriminators, determining the individual discriminators that make up the best ensemble model, thus identifying the optimal ensemble model.

[0029] The core idea of ​​unsupervised learning is to classify data and uncover potential patterns in unlabeled datasets. Reconstruction error is a crucial classification criterion in unsupervised learning methods; it is calculated by subtracting the original data from the reconstructed data. In this invention, the data from potentially faulty secondary equipment differs from the normal equipment data used during training, resulting in a larger reconstruction error. Conversely, the normal secondary equipment data closely resembles the training samples, leading to a smaller reconstruction error. Therefore, the reconstruction error of the measured secondary equipment data can serve as a criterion for determining the presence of potential faults in the secondary equipment. Based on this, an ensemble model is formed using multiple reconstruction error models to further reduce missed and false positives for potentially faulty equipment. Finally, an optimization method using the ensemble model is employed to improve the accuracy of secondary equipment anomaly warnings. The effectiveness of this method was verified using data from a smart substation protection device and its associated energy storage battery, confirming the proposed method's effectiveness.

[0030] The integrated model of this invention utilizes the Z-score method to normalize the input vectors of different features during training and evaluation, transforming the data to have a mean of 0 and a standard deviation of 1. This eliminates the influence of the dimensions of each feature vector when calculating reconstruction errors, thereby improving the stability of the model. The Z-score method has the advantages of being computationally simple and easy to use. When the operating data of secondary equipment is disorganized and it is impossible to determine the maximum and minimum values, the Z-score method can easily achieve data normalization.

[0031] This invention quantifies the failure probability and consequences of secondary equipment, making the design and operation of secondary systems more reliable. During the construction and operation of smart grids, when a secondary device in a smart substation malfunctions, the invention can quickly identify the problem and promptly replace it, thereby reducing the time required for substation operation and maintenance, shortening power outage time, and significantly improving the reliability, safety, and economy of power supply. Attached Figure Description

[0032] The present invention will now be described in further detail with reference to the accompanying drawings.

[0033] Figure 1 : Flowchart of the evaluation method of this invention;

[0034] Figure 2 : A schematic diagram of the integrated model for secondary equipment reliability assessment of this invention;

[0035] Figure 3 : Flowchart for calculating the reconfiguration error of the secondary equipment in this invention. Detailed Implementation

[0036] To better understand the present invention, the content of the invention is further clearly illustrated below with reference to embodiments and accompanying drawings. However, the scope of protection of the present invention is not limited to the embodiments described below. Numerous specific details are set forth in the following description to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the present invention can be practiced without one or more of these details.

[0037] See Figures 1-3 The purpose of this embodiment is to provide a method for evaluating the reliability of secondary equipment. For example... Figure 1 As shown, the evaluation method includes:

[0038] Step S1: Construct an integrated model for secondary equipment reliability assessment.

[0039] This step leverages the strong generalization ability of ensemble models, which reduces the probability of model errors, to construct an ensemble model for secondary equipment reliability assessment. The overall framework of the ensemble model is as follows: Figure 2 As shown.

[0040] The ensemble model utilizes the Bagging ensemble learning method to generate multiple parallel individual discriminators, and employs voting merging to make predictions. To ensure the stability of the individual discriminators, an ensemble learning method is used to construct a thermal runaway alarm framework. Different normal secondary equipment sample sets are used (…). Figure 2 Multiple individual discriminators were obtained by training on a sub-training set (in the dataset). Figure 2 The basic model in the model is used to synthesize the predictions of all individual discriminators to obtain the evaluation result, as shown in equation (1). The probability P of secondary equipment failure can be obtained by n individual discriminators as follows:

[0041]

[0042] In the formula y k This is the predicted value of the discriminator for the k-th individual.

[0043] The early warning accuracy of the ensemble model depends on the feature mining capabilities of the individual discriminators, and the stability of the individual discriminators determines whether the model can accurately identify potentially faulty secondary equipment across different test datasets. Furthermore, the parameters of the individual discriminators are significantly affected by the input sample set. Therefore, ensemble learning can be used to synthesize the judgment results of multiple individual discriminators to determine secondary equipment faults, thereby improving the overall stability of the model.

[0044] The core idea of ​​unsupervised learning is to classify data and uncover potential patterns in unlabeled datasets. Reconstruction error is a crucial classification criterion for unsupervised learning methods, and it can be obtained by calculating the difference between the original data and the reconstructed data. For example... Figure 3As shown, this invention utilizes unsupervised learning to assess the reliability of secondary equipment by calculating potential fault risks. First, an individual discriminator is trained using normal secondary equipment operating data. Then, the operating status of the secondary equipment is determined by measuring the reconstruction error of the data to be discriminated. During this process, the data from potentially faulty secondary equipment differs from the normal equipment data used during training, resulting in a larger reconstruction error. Conversely, the normal secondary equipment operating data is close to the training samples, leading to a smaller reconstruction error. Therefore, the measured reconstruction error of the secondary equipment data can serve as a criterion for determining the presence of potential faults in the secondary equipment.

[0045] The input vector during the training of the individual discriminator is denoted as X. in =[X1,X2,...,X n ], where X in The input consists of data features of the secondary device to be tested, with the specific input dimensions determined by the measurement feature parameters of the secondary device. Each vector contains a time series, such as X1 = [x 1,1 ,...x 1,t ,...x 1,T ], where x 1,t This represents a feature value at time t in the time series, where T is the total duration. All the above variables are normalized data. The output of the individual discriminator is the reconstructed data of the input features, denoted as... The dimensions of the input and output variables are equal. The optimization objective during model training is to minimize the reconstruction error of the input and output data, as shown in the following equation:

[0046]

[0047] In the formula: L is the input X in With output The reconstruction error; m is X ib and Dimension; x in,, For X in The element value at time t in the i-th dimension; x out,i,t for The element value at time t in the i-th dimension.

[0048] Once the individual discriminators mentioned above have been trained, they form an ensemble model.

[0049] Step S2: Optimize the integrated model for secondary equipment reliability assessment.

[0050] Step S1 utilizes a generalized reliability ensemble learning method. Using the concept of ensemble learning, a corresponding number of individual discriminators are trained using data from multiple normal secondary equipment of the same type. The reliability is then calculated using equation (1). Therefore, the accuracy of secondary equipment condition assessment depends on the accuracy of the ensemble model.

[0051] However, the above methods cannot determine the optimal training dataset to determine the optimal discriminator set; that is, there is no standard for selecting an ensemble model. To determine the individual discriminators that make up the optimal ensemble model, it is first necessary to clarify the relationship between the accuracy of the individual discriminators, the diversity of the individual discriminators, and the accuracy of the ensemble model. Error-divergence theory decomposes the error of the ensemble model into the features of the individual discriminators, providing a theoretical basis for selecting the optimal individual discriminators. Specifically, it can be described as follows: Assuming T individual discriminators 1,..., T To form an ensemble model, the output is obtained by weighted averaging when using this ensemble model:

[0052]

[0053] In the formula, w i For discriminator h i The weights, and determined by w i ≥0 and Constraints. Given a sample x, the divergence of the individual discriminators can be defined as:

[0054] A( i |x)=( i (x)-H(x)) 2 (4)

[0055] In the formula, H(x) represents the output of the ensemble model, and the divergence of the ensemble model can be defined as the weighted average of the divergences of the individual discriminators:

[0056]

[0057] Assuming the true nature of sample x is f(x), the weighted average error of the individual discriminator over the entire sample can be expressed as:

[0058]

[0059] Combining equations (5) and (6), the error of the integrated model can be expressed as:

[0060]

[0061] Since error represents accuracy and divergence represents diversity, it can be seen from equation (7) that the smaller the error, the higher the accuracy of the individual discriminator; the larger the divergence, the greater the diversity of the individual discriminators, and the smaller the error of the ensemble model. Therefore, in order to establish a generalized secondary equipment status assessment method, given M individual discriminators and m samples to be discriminated, the discrimination results are recorded in the following matrix:

[0062]

[0063] Let U = P T P, then the diagonal element U of Uii Individual discriminator h i The number of incorrect judgments, its off-diagonal element U ij This indicates that the individual discriminator h i with h j The number of simultaneous errors. Therefore, U contains both the error of individual discriminators and their divergence. The elements in U are normalized using the following formula.

[0064]

[0065] In the formula, M is the number of samples to be judged. According to the definition of formula (7), if If every element in the set is minimized, then the ensemble model performs best. This is a good approximation of the ensemble error. Therefore, the ensemble model optimization problem can be transformed into the following quadratic integer programming problem:

[0066]

[0067] In the formula, the binary scalar x i This indicates whether the i-th individual discriminator is selected into the ensemble model, and k is the size of the optimized ensemble model.

[0068] After this step, the optimal ensemble model contains only the individual discriminators with high accuracy, eliminating those with low accuracy. Secondary equipment involves many types, and the size of the ensemble model for each type should be determined based on the actual situation. This step can be used to find the optimal ensemble model for a given ensemble model.

[0069] Step S3: Conduct a reliability assessment of the secondary equipment.

[0070] This step involves inputting the current operating data of the secondary equipment into various association rules and integration models based on real-time monitoring data to conduct a reliability assessment.

[0071] The reconstruction error of the secondary equipment under test is calculated by the individual discriminator, and the judgment rule is as follows:

[0072]

[0073] In the formula: X′ in This represents the input of a secondary device that needs to be evaluated; y represents the output of a secondary device to be judged; y represents the judgment result of the individual discriminator of the corresponding secondary device; σ represents the judgment threshold for whether a potential fault has occurred. By completing the model calculation of the reconstruction error of the sample data of the device to be judged, if the reconstruction error is above the set threshold, the secondary device is judged as having a potential fault, and the input value is 1, indicating that it has not passed the reliability assessment; if the reconstruction error is less than the set threshold, the secondary device is judged as normal, and the input value is 0, indicating that it has passed the reliability assessment. Finally, the integrated model's evaluation result of the reliability of the secondary device is obtained by combining formula (1).

[0074] The ensemble model of this invention uses the Z-score method to normalize the vectors input during training and evaluation, transforming the data to have a mean of 0 and a standard deviation of 1, thereby eliminating the influence of the dimensions of each feature vector when calculating reconstruction errors. The formula for the Z-score method is:

[0075]

[0076] In the formula, μ is the mean of all sample data, and σ ′ Let X be the standard deviation of all sample data, and X be the element value of an individual.

[0077] The following uses battery data and protection device data from a smart substation as an example to verify the effectiveness of the above evaluation method and the optimization effect of the integrated model. Specifically, the results obtained by executing S1+S3 (before optimization) and S1+S2+S3 (after optimization) under the same set of data are analyzed.

[0078] The power supply battery equipment in the secondary equipment of this intelligent substation involves 50 battery groups. Parameters for each group include voltage, current, state of charge, and temperature, with a time span of approximately six months and a sampling frequency of 10Hz. Due to the highly variable operating conditions of batteries in non-charging states, even between normal sample sets, the data varies significantly; therefore, only actual charging data is used. However, since the charging frequency and duration differ for each battery group, the shortest charging data length is used as the baseline. Two energy storage battery groups are identified as faulty secondary equipment. The secondary equipment includes 20 protection devices, with parameters for each protection device including temperature, operating voltage, transmitted light intensity, and received light intensity. One of these protection devices is faulty.

[0079] The following are common early warning indicators used in the industry:

[0080]

[0081] F1 = 2 × PRE × REC / (PRE + REC)

[0082] In the formula: TP / TN represents positive samples classified as positive / negative, and FP / FN represents negative samples classified as positive / negative. ACC is precision; PRE is accuracy; REC is recall; F1 is the harmonic mean of precision and recall, reflecting both the model's precision and recall. Higher values ​​for these metrics indicate better model classification performance. Furthermore, when determining the secondary device ranking metric, the probability of warnings for all secondary devices in the same test set is calculated and sorted from highest to lowest. A lower ranking value for the detected secondary device indicates better model performance.

[0083] After preprocessing (normalizing) the data from the battery device, ensemble models before and after optimization were trained, and fault prediction was performed on each. The results are as follows:

[0084] Table 1. Performance comparison between the ensemble model before and after optimization.

[0085]

[0086] Table 2. Early Warning Probability and Ranking of Secondary Equipment Power Supply Batteries

[0087]

[0088] Table 1 shows that the unoptimized ensemble model missed some faulty secondary devices, while the optimized ensemble model did not miss any. Furthermore, the optimized ensemble model improved classification accuracy, precision, and recall by 7.2%, 25%, and 50%, respectively, compared to the unoptimized model. Table 2 shows that, comparing the unoptimized and optimized models, the predicted probability ranking of actual faulty batteries after optimization was 1st and 2nd, respectively, while before optimization it was only 4th and 6th. Therefore, the ensemble optimization improved the model accuracy.

[0089] After preprocessing (normalizing) the data from the protection device, ensemble models before and after optimization were trained, and fault prediction was performed on each. The results are as follows:

[0090] Table 3. Performance comparison of the ensemble model before and after optimization.

[0091]

[0092]

[0093] Table 4. Early Warning Probability and Ranking of Intelligent Substation Protection Devices

[0094]

[0095] Table 3 shows that neither the pre-optimization nor the post-optimization integrated model missed any faulty secondary equipment. However, the post-optimization integrated model improved the classification accuracy, classification precision, and F1 score by 15%, 75%, and 0.6, respectively, compared to the pre-optimization model. Table 4 shows that the pre-warning probability ranking of the actual faulty battery was 1st after optimization, while it was 3rd before optimization. Therefore, the integrated optimization improved the model's accuracy in identifying protection devices.

[0096] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A method of secondary equipment reliability assessment, characterized by: The method comprises the following steps: constructing a secondary equipment reliability evaluation integrated model, the integrated model comprising a plurality of parallel individual discriminators, the individual discriminators being trained by using operation data of normal secondary equipment, and an optimization objective of the individual discriminators being to minimize reconstruction error of input and output data; optimizing the secondary equipment reliability evaluation integrated model, eliminating the individual discriminators with low accuracy, and obtaining an optimal integrated model; inputting current operation data of the secondary equipment into the optimal integrated model for reliability evaluation; the method for optimizing the secondary equipment reliability evaluation integrated model comprises the following steps: The error of the ensemble model is represented as where is the weighted average error of individual discriminators over the whole sample, is the divergence of the ensemble model; , , where is the weight of individual discriminators , is the output of individual discriminators for a sample x. The discrimination results of M individual discriminators and m samples to be discriminated are recorded in a matrix P, and the matrix element Pij is 0 when the individual discriminator j correctly discriminates the sample i, otherwise 1. value. Let and using normalizing the elements in the matrix U; An optimization problem defining the ensemble model is , , where binary scalar represents whether the i-th individual discriminator is selected into the ensemble model or not, and k is the size of the optimized ensemble model. optimizing the integrated model by using the optimization problem.

2. The method of claim 1, wherein: The integrated model obtains an evaluation result by comprehensively considering prediction values of all the individual discriminators.

3. The method of claim 2, wherein: The evaluation result of the integrated model is represented as where n is the number of individual discriminators, is the prediction value of the kth individual discriminator.

4. The method of claim 1, wherein: The optimization objective of the individual discriminator training is wherein is the input secondary equipment operation data, is the output end reconstruction data, L is and is the reconstruction error of and dimension, is the element value of the i-th dimension at time t in is the element value of the i-th dimension at time t in 5. The method of claim 4, wherein: The judgment rule of the individual discriminator is wherein represents the input of the secondary equipment to be judged, represents the output of the secondary equipment to be judged, and σ represents a judgment threshold value of whether a potential fault has occurred.

6. The method of claim 1, wherein: The operation data of the secondary equipment are normalized by using a Z-score method before being inputted.

7. The method of claim 6, wherein: The Z-score method is represented as where μ is the mean of all sample data, and σ is the standard deviation of all sample data, and X is the elemental value of the individual.

8. The method of claim 1, wherein: The input vectors of the individual discriminators all comprise time series.

Images