Equipment operation efficiency evaluation method, device, storage medium and computer equipment

Through the feature extraction and dimensionality reduction model combined with quantum optimization algorithm and dual loss function, combined with topological dynamic adjustment and variational mode, the problem of inefficiency of traditional equipment operating efficiency evaluation methods is solved, and more accurate and timely evaluation results are achieved.

CN118585926BActive Publication Date: 2024-12-03GUANGZHOU POWER SUPPLY BUREAU GUANGDONG POWER GRID CO LTD
View PDF 3 Cites 0 Cited by

Patent Information

Application Number
CN202411074381.8
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-08-07
Publication Date
2024-12-03
Estimated Expiration
2044-08-07

AI Technical Summary

Technical Problem

Traditional equipment operating efficiency evaluation methods are inefficient, and the accuracy and timeliness of evaluation results are difficult to meet the needs of modern industrial production.

Method used

The target feature extraction model expanded based on quantum optimization algorithm and the target feature dimensionality reduction model trained by dual loss function are adopted, combined with the topological dynamic adjustment mechanism and variational mode, in-depth data analysis and intelligent prediction are carried out to evaluate the operating performance of the equipment.

Benefits of technology

It improves the accuracy and timeliness of equipment operation efficiency evaluation, enhances the generalization ability of unseen samples, and provides more comprehensive and accurate evaluation results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118585926B_ABST
    Figure CN118585926B_ABST
Patent Text Reader

Abstract

The present application provides a method, device, storage medium and computer equipment for evaluating the equipment operation efficiency. The present application first uses two models to extract and reduce the dimension to obtain the target feature vector of the operation data. Among them, the target feature extraction model is obtained based on the sample operation data obtained by expanding the quantum optimization algorithm and training with the topological dynamic adjustment mechanism. The quantum optimization algorithm can improve the diversity of the generated samples, while the topological dynamic adjustment mechanism can adapt to the differences between different samples and delete redundant connections in the model to improve the learning efficiency and accuracy of the model; the target feature dimensionality reduction model is trained using a dual loss function, which can evaluate the performance of the model from multiple levels to improve the accuracy of the evaluation results. Finally, after mining the potential information of the target feature vector through variational mode decomposition, the target classifier is used to evaluate the performance of the decomposed target feature vector, which can improve the accuracy and timeliness of the evaluation results.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, storage medium and computer device for evaluating equipment operation efficiency. Background Art

[0002] In the modern industrial environment, equipment performance and operating efficiency are key factors in achieving production goals and economic benefits. The optimization of equipment performance not only affects product quality and production volume, but is also directly related to energy consumption and maintenance costs. As the core element of equipment performance optimization, the evaluation of equipment operating performance is crucial to conduct efficient and accurate evaluation of equipment operating performance.

[0003] However, traditional performance evaluation methods often rely on manual periodic inspections and simple sensor data collection, lacking in-depth data analysis and intelligent prediction capabilities, and are unable to cope with the complex and ever-changing modern industrial production scenarios. In short, with the increase in equipment types, the surge in data volume, and the increasing complexity of failure modes, traditional equipment operation performance evaluation methods are not only inefficient, but also the accuracy and timeliness of evaluation results are difficult to meet the needs of modern industrial production. Summary of the invention

[0004] The purpose of this application is to solve at least one of the above-mentioned technical defects, especially the technical defect that the equipment operation efficiency evaluation method in the prior art is not only inefficient, but also the accuracy and timeliness of the evaluation results cannot meet the needs of modern industrial production.

[0005] The present application provides a method for evaluating the operation efficiency of an equipment, the method comprising:

[0006] Obtain the operating data of the equipment to be evaluated;

[0007] Determine a target feature extraction model and a target feature dimensionality reduction model; wherein the target feature extraction model is obtained by training a preset initial feature extraction model using a topological dynamic adjustment mechanism based on sample operation data expanded by a quantum optimization algorithm; and the target feature dimensionality reduction model is obtained by training a preset initial feature dimensionality reduction model using a dual loss function;

[0008] Using the target feature extraction model to extract features from the operating data to obtain target data features, and inputting the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model;

[0009] The target feature vector is decomposed through a variational model, and the decomposed target feature vector is input into a predetermined target classifier to obtain an operating efficiency evaluation result of the device to be evaluated output by the target classifier.

[0010] Optionally, the process of expanding the sample operation data includes:

[0011] Collect the real operation data generated by multiple devices to be evaluated during operation;

[0012] Determine a target data expansion model; the target data expansion model is obtained by training a preset initial data expansion model using a quantum optimization algorithm and a quality evaluation function;

[0013] Inputting the real operation data into the target data expansion model to obtain the expanded operation data output by the target data expansion model;

[0014] The performance type of each real operating data is determined, and each real operating data and the corresponding extended operating data are type-labeled based on each performance type to form sample operating data.

[0015] Optionally, determining the target data expansion model includes:

[0016] Initializing network parameters of a generator and a discriminator in a generative adversarial network to form an initial data expansion model, and performing quantum state optimization on the initial data expansion model to generate quantum parameters;

[0017] Using the generator to generate predicted operation data of the real operation data collected in advance based on the quantum parameters, and using the discriminator to discriminate the authenticity of the real operation data and the predicted operation data to obtain a discrimination result;

[0018] A quality assessment function is used to perform quality assessment on the predicted operation data to obtain an assessment result, and based on the assessment result, the network parameters of the initial data expansion model are optimized using the predicted operation data and the discrimination result until the optimization round reaches a preset optimization round threshold, thereby generating a target data expansion model.

[0019] Optionally, the process of determining the target feature extraction model includes:

[0020] Initialize the network parameters and topological structure of the pre-trained neural network to form an initial feature extraction model;

[0021] Inputting the pre-acquired sample operation data into the initial feature extraction model to obtain the predicted data features output by the initial feature extraction model;

[0022] With the goal of making the predicted data features approach the real data features of the sample data features, optimizing the network parameters or topological structure of the initial feature extraction model;

[0023] When the optimization round of the initial feature extraction model reaches a preset optimization round threshold, the optimized initial feature extraction model is used as the target feature extraction model.

[0024] Optionally, the optimizing the network parameters or topological structure of the initial feature extraction model includes:

[0025] Determining a loss function and a topology adjustment index of the initial feature extraction model based on the network parameters and the topology structure, and determining an optimization direction of the initial feature extraction model based on the loss function and the topology adjustment index;

[0026] When the optimization direction is parameter optimization, a loss value of the initial feature extraction model is calculated based on the predicted feature vector, and a network parameter of the initial feature extraction model is optimized according to the loss value;

[0027] When the optimization direction is topology optimization, the topology reconstruction strategy of the initial feature extraction model is determined, and the topology structure of the initial feature extraction model is optimized according to the topology reconstruction strategy. In addition, the performance of the optimized initial feature extraction model is evaluated using a topology reconstruction feedback mechanism, and the topology reconstruction strategy is adjusted according to the evaluation results.

[0028] Optionally, the initial feature dimensionality reduction model includes an autoencoder, a microadjustable layer and a decoder;

[0029] Determining the target feature dimensionality reduction model includes:

[0030] Initializing the network parameters of the initial feature dimensionality reduction model, and using the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features;

[0031] Performing dimension adjustment on the initial low-dimensional features through the micro-adjustable layer to obtain final low-dimensional features, and inputting the final low-dimensional features into the decoder to obtain predicted high-dimensional features output by the decoder;

[0032] A dual loss function is used to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features, and the network parameters of the initial feature dimensionality reduction model are optimized based on the dimensionality loss value until the optimization round reaches a preset optimization round threshold, thereby generating a target feature dimensionality reduction model.

[0033] Optionally, the process of determining the target classifier includes:

[0034] Initializing network parameters of an extreme learning machine and determining a minimized loss function of the extreme learning machine;

[0035] Inputting the pre-acquired sample training data into the extreme learning machine to obtain a predicted value output by the extreme learning machine;

[0036] Based on the predicted value, the network parameters of the extreme learning machine are optimized using the minimization loss function until the optimization round reaches a preset optimization round threshold, thereby generating a target classifier.

[0037] The present application also provides a device for evaluating equipment operation efficiency, comprising:

[0038] A data acquisition module, used to acquire the operating data of the equipment to be evaluated;

[0039] A model determination module is used to determine a target feature extraction model and a target feature dimensionality reduction model; wherein the target feature extraction model is obtained by training a preset initial feature extraction model based on sample operation data expanded by a quantum optimization algorithm using a topological dynamic adjustment mechanism; and the target feature dimensionality reduction model is obtained by training a preset initial feature dimensionality reduction model using a dual loss function;

[0040] A feature extraction module is used to extract features from the operating data using the target feature extraction model to obtain target data features, and input the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model;

[0041] The performance evaluation module is used to decompose the target feature vector through a variational model, and input the decomposed target feature vector into a predetermined target classifier to obtain the operating performance evaluation result of the device to be evaluated output by the target classifier.

[0042] The present application also provides a storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the device operation efficiency evaluation method as described in any of the above embodiments.

[0043] The present application also provides a computer device, comprising: one or more processors, and a memory;

[0044] The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the device operation efficiency evaluation method as described in any one of the above embodiments are performed.

[0045] It can be seen from the above technical solutions that the embodiments of the present application have the following advantages:

[0046] The equipment operation efficiency evaluation method, device, storage medium and computer equipment provided in the present application can determine the target feature extraction model and the target feature dimensionality reduction model after obtaining the operation data of the equipment to be evaluated. Among them, the target feature extraction model is based on the sample operation data obtained by the quantum optimization algorithm, and is trained by the topological dynamic adjustment mechanism. The sample operation data obtained by the quantum optimization algorithm is diverse, and the topological dynamic adjustment mechanism can better adapt to the differences between different samples during the model training process, and can identify and delete redundant connections in the model. Therefore, the feature extraction efficiency of the model trained in this way is high, and the generalization ability on unseen samples can be improved, thereby improving the accuracy of the evaluation results; and the target feature dimensionality reduction model is trained by the dual loss function. Compared with the traditional single loss function, the dual loss function may evaluate the performance of the model from multiple angles or levels, thereby providing a more comprehensive and accurate evaluation result. Therefore, using these two models, it is possible to efficiently generate target feature vectors with high quality in the operation data. Finally, the present application can decompose the target feature vector through a variational model to mine the potential information of the target feature vector. In this way, when the target classifier is used to evaluate the equipment operating performance of the decomposed target feature vector, the accuracy and timeliness of the evaluation results can be greatly improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0048] Figure 1 A schematic diagram of a flow chart of a method for evaluating equipment operation efficiency provided in an embodiment of the present application;

[0049] Figure 2 A flowchart of a target data expansion model determination process provided in an embodiment of the present application;

[0050] Figure 3 A schematic diagram of a process for determining a target feature extraction model provided in an embodiment of the present application;

[0051] Figure 4 A schematic diagram of a process for determining a target feature dimensionality reduction model provided in an embodiment of the present application;

[0052] Figure 5 A schematic diagram of the structure of a device operation efficiency evaluation device provided in an embodiment of the present application;

[0053] Figure 6 A schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] Traditional performance evaluation methods often rely on manual periodic inspections and simple sensor data collection, lacking in-depth data analysis and intelligent prediction capabilities, and are unable to cope with the complex and ever-changing modern industrial production scenarios. In short, with the increase in equipment types, the surge in data volume, and the increasing complexity of failure modes, traditional equipment operation performance evaluation methods are not only inefficient, but also the accuracy and timeliness of evaluation results are difficult to meet the needs of modern industrial production.

[0056] Based on this, this application proposes the following technical solutions, see below for details:

[0057] In one embodiment, Figure 1 As shown, Figure 1 A schematic diagram of a flow chart of a method for evaluating the performance of equipment operation provided in an embodiment of the present application; the present application provides a method for evaluating the performance of equipment operation, see below for details:

[0058] S110: Obtaining operating data of the device to be evaluated.

[0059] In this step, when the user needs to perform performance evaluation on a device, the computer device may first obtain the operating data of the device to be evaluated, and then perform performance evaluation on the device to be evaluated based on the operating data.

[0060] Specifically, the operating data here can include not only the basic parameter data of the equipment to be evaluated, but also the real-time data generated by the equipment to be evaluated during operation. The basic parameter data can include basic information such as the model, serial number, performance parameters and production date of the equipment to be evaluated, while the real-time data can be data monitored in real time by sensors installed on computer equipment, such as temperature, humidity, pressure, vibration, current, voltage and other data, to reflect the operating environment and working status of the equipment to be evaluated.

[0061] Furthermore, when the computer device obtains the operating data of the device to be evaluated, it can read and store the operating data in chronological order to improve the efficiency of subsequent data processing and evaluation. Moreover, after reading the operating data, the computer device can also perform preprocessing operations on the operating data, including but not limited to denoising, data cleaning, and data normalization.

[0062] To elaborate, denoising refers to removing noise and interference from operating data to improve data quality and availability; data cleaning refers to removing outliers, supplementing missing values, and smoothing data to improve data accuracy and completeness; and data normalization refers to normalizing operating data to ensure that the weights of data in different dimensions are the same to avoid data bias and errors.

[0063] S120: Determine a target feature extraction model and a target feature dimensionality reduction model.

[0064] In this step, after obtaining the operating data of the device to be evaluated through step S110, the computer device can also determine the target feature extraction model and the target feature dimensionality reduction model, so as to perform feature extraction and feature dimensionality reduction on the operating data through these two models, thereby improving the efficiency and accuracy of subsequent performance evaluation.

[0065] Among them, the target feature extraction model refers to a model that extracts features from the input operating data and obtains the target data features. This model is based on the sample operating data expanded by the quantum optimization algorithm, and uses a topological dynamic adjustment mechanism to train the preset initial feature extraction model; and the target feature dimensionality reduction model refers to a model that performs feature dimensionality reduction on the input target data features and obtains the target feature vector. This model is obtained by training the preset initial feature dimensionality reduction model using a dual loss function.

[0066] It is understandable that a quantum optimization algorithm refers to an optimization algorithm based on a quantum computer. Its core idea is to use the parallel computing power of a quantum computer and the superposition and interference of quantum states. It can process multiple computing tasks at the same time, accelerate the computing speed, and quickly find the optimal solution or a solution close to the optimal solution in the search space. Therefore, this application can expand a small amount of running data into a large amount of sample running data through a quantum optimization algorithm to generate more diverse and higher quality sample running data under limited resources, and use these data as input data for model training, which can improve the generalization ability and accuracy of the model.

[0067] After the sample running data is input into the initial feature extraction model, the model can also be trained in combination with the topology dynamic adjustment mechanism. The topology dynamic adjustment mechanism here refers to dynamically adjusting the model's topology structure or network parameters according to certain rules or conditions, so that the initial feature extraction model can dynamically adjust the number of network layers, number of neurons, connection methods, etc. to adapt to the feature distribution learning of different training samples, thereby improving the accuracy and efficiency of the model.

[0068] In addition, when training the target feature dimensionality reduction model, the present application can combine the dual loss function to train the preset initial feature dimensionality reduction model. The dual loss function here refers to a method of solving the original problem by constructing a dual problem and using the properties of the dual variable and the dual problem in the optimization problem. When training the initial feature dimensionality reduction model, the present application can introduce a dual loss function to provide a clear goal for the model training process. By minimizing the dual loss function, the initial feature dimensionality reduction model can optimize multiple goals at the same time, thereby better retaining the original information of the data during the feature dimensionality reduction process, thereby enhancing the generalization ability and accuracy of the model.

[0069] S130: extracting features from the operating data using the target feature extraction model to obtain target data features, and inputting the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model.

[0070] In this step, after obtaining the target feature extraction model and the target feature dimensionality reduction model through step S120, the computer device can use the target feature extraction model to extract features from the operating data to obtain target data features, and then the target data features can be input into the target feature dimensionality reduction model to obtain the target feature vector output by the target feature dimensionality reduction model.

[0071] Specifically, after the computer device inputs the operation data into the target feature extraction model, the target feature extraction model can first identify the features of the operation data, extract the identified features and then output them to obtain the target data features. Then, the computer device can input the target data features output by the target feature extraction model into the target feature dimensionality reduction model, so that the target feature dimensionality reduction model can map the target data features from high-dimensional features to low-dimensional space, form a low-dimensional target feature vector and then output it as the basic data for the performance evaluation of the device to be evaluated.

[0072] S140: Decomposing the target feature vector through a variational model, and inputting the decomposed target feature vector into a predetermined target classifier to obtain an operating efficiency evaluation result of the device to be evaluated output by the target classifier.

[0073] In this step, after obtaining the target feature vector of the device to be evaluated through step S130, the computer device can decompose the target feature vector through a variational model to mine the potential information of the target feature vector, and then the decomposed target feature vector can be input into a predetermined target classifier to obtain the operating performance evaluation result of the device to be evaluated output by the target classifier.

[0074] Among them, variational mode refers to a technology that decomposes complex information into multiple modal components. These modal components can represent different components or specific characteristic trends in the information, so as to effectively reveal the hidden structure and potential characteristics in the information. Therefore, the present application can use variational mode to decompose the target feature vector to mine the potential information of the target feature vector, thereby improving the accuracy of subsequent performance evaluation.

[0075] Specifically, when the computer device performs performance evaluation on the decomposed target feature vector, it can first load a pre-trained target classifier, which can be a machine learning model, such as an extreme learning machine, a support vector machine, a random forest, a neural network, etc., so that the decomposed target feature vector can be input into the target classifier, so that the target classifier can perform performance prediction on the input data and output the operating performance evaluation result of the device to be evaluated.

[0076] Furthermore, the present application can classify the performance level of the equipment according to the specific needs of the evaluation business, and then use different training data to train the target classifier according to the performance labels obtained by the classification, so that the target classifier can output the corresponding performance label during the performance evaluation process as the operating performance evaluation result of the equipment to be evaluated. For example, in a target classifier, the operating performance evaluation result of its evaluation output may include low performance, medium performance, high performance and high performance.

[0077] In the above embodiment, after obtaining the operation data of the device to be evaluated, the target feature extraction model and the target feature dimensionality reduction model can be determined. Among them, the target feature extraction model is based on the sample operation data obtained by the quantum optimization algorithm, and is trained by the topological dynamic adjustment mechanism. The sample operation data obtained by the quantum optimization algorithm is diverse, and the topological dynamic adjustment mechanism can better adapt to the differences between different samples during the model training process, and can identify and delete redundant connections in the model. Therefore, the feature extraction efficiency of the model trained in this way is high, and the generalization ability on unseen samples can be improved, thereby improving the accuracy of the evaluation results; and the target feature dimensionality reduction model is trained by the dual loss function. Compared with the traditional single loss function, the dual loss function may evaluate the performance of the model from multiple angles or levels, thereby providing a more comprehensive and accurate evaluation result. Therefore, using these two models, it is possible to efficiently generate a target feature vector with high quality in the operation data. Finally, the present application can decompose the target feature vector through the variational mode to mine the potential information of the target feature vector, so that when the target classifier is used to evaluate the equipment operation efficiency of the decomposed target feature vector, the accuracy and timeliness of the evaluation results can be greatly improved.

[0078] In one embodiment, the process of expanding the sample operation data in step S120 may include:

[0079] S1211: Collecting actual operation data generated by multiple devices to be evaluated during operation.

[0080] S1212: Determine a target data expansion model; the target data expansion model is obtained by training a preset initial data expansion model using a quantum optimization algorithm and a quality evaluation function.

[0081] S1213: Input the actual operation data into the target data expansion model to obtain the expanded operation data output by the target data expansion model.

[0082] S1214: Determine the performance type of each real operating data, and perform type annotation on each real operating data and the corresponding extended operating data based on each performance type to form sample operating data.

[0083] In this embodiment, in the process of expanding the sample operation data, the computer device can first collect the real operation data generated by multiple devices to be evaluated during operation, and determine the target data expansion model for data expansion, so that after the computer device inputs the real operation data into the target data expansion model, it can obtain the expanded operation data output by the target data expansion model. Finally, the computer device can determine the performance type of each real operation data, and based on each performance type, each real operation data and the corresponding expanded operation data are type-labeled to form sample operation data. The performance types here can include low performance, medium performance, high performance and high performance.

[0084] It is understandable that the training of the model depends on the quantity and quality of the training data, and the real operation data collected by the computer equipment is affected by the operating environment, resulting in insufficient effective data and unable to support the training of large parameter models. Therefore, this application can expand the sample of the collected real operation data to train the target feature extraction model. Among them, the real operation data refers to the operating data such as temperature, humidity, pressure, vibration, current, voltage, etc. collected from the running equipment, which can reflect the operation of the equipment during actual use.

[0085] Specifically, when collecting real operation data, the computer device can collect the operation data through sensors or other monitoring devices installed on the device. The real operation data collected here needs to cover devices of different types and different usage scenarios to ensure that the collected data is representative. Then the computer device can design and generate a target data expansion model based on the goals of maintaining the distribution characteristics of the data and increasing the diversity of the data, so that the real operation data can be input into the target data expansion model, and the expanded operation data corresponding to each real operation data can be expanded. The expanded operation data is basically consistent with the real operation data in distribution, so its corresponding performance type is consistent with the real operation data used for expansion. For each real operation data and expanded operation data, it is necessary to mark its corresponding performance type to form sample operation data to ensure the integrity and availability of the training data. At the same time, in order to ensure sample balance, the number of sample data of each performance type in the sample operation data should be balanced to ensure the accuracy and stability of the model during training and evaluation.

[0086] Furthermore, in the process of constructing the target data expansion model, the present application can use quantum optimization algorithms to improve the diversity of the expanded operation data, and can also introduce quality assessment functions to further improve the quality of the expanded operation data. Among them, the quality assessment function refers to a function used to measure the quality or diversity of the expanded data; therefore, the present application can use the quality assessment function to evaluate the quality and diversity of the expanded operation data generated by the model to ensure that the expanded operation data can cover the diversity of the data while being close to the real operation data, thereby enhancing the generalization ability and accuracy of the subsequent training model.

[0087] In one embodiment, Figure 2 As shown, Figure 2 A flowchart of a target data expansion model determination process provided in an embodiment of the present application; Figure 2 In step S1212, determining the target data expansion model may include:

[0088] S2121: Initialize the network parameters of the generator and the discriminator in the generative adversarial network to form an initial data expansion model, and perform quantum state optimization on the initial data expansion model to generate quantum parameters.

[0089] S2122: Using a generator to generate predicted operation data of real operation data collected in advance based on quantum parameters, and using a discriminator to discriminate the authenticity of the real operation data and the predicted operation data to obtain a discrimination result.

[0090] S2123: Use a quality assessment function to perform quality assessment on the predicted operation data to obtain an assessment result, and based on the assessment result, use the predicted operation data and the discrimination result to optimize the network parameters of the initial data expansion model until the optimization round reaches a preset optimization round threshold, thereby generating a target data expansion model.

[0091] In this embodiment, when the computer device trains the target data expansion model, it can first initialize the network parameters of the generator and the discriminator in the generative adversarial network to form an initial data expansion model, and perform quantum state optimization on the initial data expansion model to generate quantum parameters. Then, the generator can be used to generate predicted operation data of the real operation data collected in advance based on the quantum parameters, and the discriminator can be used to distinguish the authenticity of the real operation data and the predicted operation data to obtain the discrimination result. Finally, the computer device can use the quality assessment function to perform quality assessment on the predicted operation data to obtain the assessment result, and based on the assessment result, the predicted operation data and the discrimination result are used to optimize the network parameters of the initial data expansion model until the optimization round reaches the preset optimization round threshold, and the target data expansion model is generated.

[0092] Among them, the generative adversarial network is a deep learning model consisting of a generator and a discriminator. The generator tries to generate realistic augmented samples, while the discriminator tries to distinguish the augmented samples generated by the generator from the real samples. The two compete with each other through adversarial training, and finally the generator can generate augmented samples similar to the real samples. As a part of the generative adversarial network, the generator is mainly responsible for generating realistic augmented samples from random noise. Its training goal is to generate augmented samples that can deceive the discriminator, so that the discriminator cannot accurately distinguish between real samples and augmented samples. As another part of the generative adversarial network, the discriminator is mainly responsible for judging whether the input sample is from the augmented sample generated by the generator or the real sample. Its training goal is to distinguish the augmented samples generated by the generator from the real samples as accurately as possible. In addition, the quantum parameters generated when the quantum state optimization is performed in the initial data augmentation model are random, which can prevent the model from falling into the local optimal solution in the process of generating augmented samples through quantum optimization, so as to find the global optimal solution, thereby improving the diversity of augmented generated samples.

[0093] Furthermore, when the computer device trains the generator and the discriminator, it can adopt an alternating training method, that is, iterative training is performed by first training the discriminator and then training the generator. Specifically, in each training cycle, the parameters of the generator can be fixed first, and the discriminator loss function can be used to train the discriminator and optimize the parameters, and then the parameters of the discriminator can be fixed again, and the generator loss function can be used to train the generator and optimize the parameters, so as to avoid the gradients of the two models from interfering with each other during the training process. Among them, the generator takes the similarity between the predicted running data and the real running data as not less than the preset similarity threshold as the training goal, and the discriminator takes the accuracy of the discrimination results of the predicted running data and the real running data as not less than the preset accuracy threshold as the goal. When the optimization rounds of the generator and the discriminator reach the preset optimization round threshold, the target data expansion model can be obtained.

[0094] Specifically, the training process of the target data augmentation model can be as follows:

[0095] First, the computer device can initialize the network parameters of the generator and the discriminator. Here, the network parameters can be initialized to randomly distributed small values ​​in the interval [0, 1]. Then, in the quantum state optimization phase, the generator can accept input from the quantum state combination optimization module, which can represent the potential data state through quantum bits (qubits) and use the Hamiltonian operator H to optimize the state, where the quantum state The calculation method can be expressed as:

[0096]

[0097] In the formula, represents the quantum state, represents the total number of quantum ground states, represents the i-th quantum ground state, represents the corresponding complex probability amplitude.

[0098] Complex probability amplitude The energy expectation value can be minimized by quantum algorithm optimization. The calculation method of the energy expectation value can be expressed as:

[0099]

[0100] In the formula, Represents the complex probability amplitude The complex conjugate of represents the Hamiltonian, and N is the total number of quantum ground states.

[0101] in, It can be decomposed into the calculation of the conversion elements of the system Hamiltonian H between different quantum ground states, which can be specifically expressed as:

[0102]

[0103] In the formula, and Represents quantum states and In Location The wave function expression of , H represents the Hamiltonian operator.

[0104] In the data expansion phase, the generator G can use the parameters obtained from quantum state optimization To generate extended running data, the extended running data The generation formula can be expressed as:

[0105]

[0106] In the formula, Represents the optimized quantum state The latent vector extracted from Representation Generator Parameters.

[0107] In the adversarial training phase, the discriminator D needs to distinguish between the generated augmented running data and the real running data, and the loss function used to train the discriminator is for:

[0108]

[0109] In the formula, The symbol for a particular distribution is represented by is the distribution of real running data, represents the distribution of latent variables.

[0110] Among them, each logarithmic term is expanded and calculated, which can be specifically expressed as:

[0111]

[0112]

[0113] In the formula, and They represent the distribution functions of the real running data and the expanded running data respectively, and They represent the evaluation output of the discriminator on the real running data and the augmented running data respectively.

[0114] Next, the generator can self-correct based on the feedback from the discriminator. Specifically, the feedback adjustment of the generator can be achieved by minimizing the loss function To achieve this, specifically:

[0115]

[0116] Therefore, the generator can be trained according to the loss function , using gradient descent to adjust the parameters To generate more realistic data.

[0117] In addition, the computer device can periodically evaluate the quality and diversity of the augmented operation data to ensure that the generated augmented operation data is not only close to the real operation data, but also covers the diversity of the data and enhances the generalization ability of the model. The quality and diversity of the augmented operation data are evaluated by using the quality evaluation function Q, which can be expressed as:

[0118]

[0119] In the formula, and represents the weight factor for adjusting diversity and authenticity, represents the variance of the augmented running data, represents the diversity measure of the extended running data. In this application, and Can be set to 0.3 and 0.7 respectively.

[0120] Among them, the variance The calculation method can be expressed as:

[0121]

[0122] Diversity The calculation method can be expressed as:

[0123]

[0124] In the formula, It is a diversity metric function. and Represents data samples generated from different latent vectors. The function is calculated as the inverse of the L2 norm.

[0125] Finally, the above steps can be repeated until the preset stop iteration condition is met, indicating that the model training is completed. In this application, the preset stop iteration condition can be reaching a preset maximum number of iterations, where the preset maximum number of iterations is set to 1000 times, and there is no limit here.

[0126] For example, after the target data expansion model is trained, the computer device uses the trained target data expansion model to increase the number of samples. If the original collected real operation data is 800, and the expanded operation data generated by the target data expansion model is 200, then the sample operation data formed after expansion includes 1000 sample data.

[0127] In one embodiment, Figure 3 As shown, Figure 3 A schematic diagram of a process for determining a target feature extraction model provided in an embodiment of the present application; Figure 3 In step S120, determining the target feature extraction model may include:

[0128] S1221: Initialize the network parameters and topological structure of the pre-trained neural network to form an initial feature extraction model.

[0129] S1222: Input the pre-acquired sample operation data into the initial feature extraction model to obtain the predicted data features output by the initial feature extraction model.

[0130] S1223: With the goal of making the predicted data features close to the real data features of the sample running data, the network parameters or topological structure of the initial feature extraction model are optimized.

[0131] S1224: When the optimization round of the initial feature extraction model reaches a preset optimization round threshold, the optimized initial feature extraction model is used as the target feature extraction model.

[0132] In this embodiment, when determining the target feature extraction model, the computer device may first initialize the network parameters and topological structure of the pre-trained neural network to form an initial feature extraction model, so that the pre-acquired sample operation data can be input into the initial feature extraction model to obtain the predicted data features output by the initial feature extraction model; then the computer device may optimize the network parameters or topological structure of the initial feature extraction model with the real data features that the predicted data features approach as the goal, and when the optimization rounds of the initial feature extraction model reach a preset optimization round threshold, the computer device may use the optimized initial feature extraction model as the target feature extraction model.

[0133] Specifically, the computer device can select a corresponding neural network to initialize data and form an initial feature extraction model, which can then be improved and trained through the initial feature extraction model. During the training process, the computer device can first obtain sample operation data, and each sample data in the sample operation data is marked with a corresponding sample label, and the sample label is the real data feature of the sample data. After the sample operation data is input into the initial feature extraction model, the predicted data features output by the initial feature extraction model can be obtained. Then, the computer device can optimize the network parameters or topological structure of the initial feature extraction model with the goal of making the predicted data features approach the real data features of the sample operation data. When the optimization round of the initial feature extraction model reaches the preset optimization round threshold, the model training can be regarded as completed, and the trained initial feature extraction model can be used as the target feature extraction model.

[0134] In one embodiment, the optimization of the network parameters or topology structure of the initial feature extraction model in step S1223 may include:

[0135] S2231: Determine the loss function and topology adjustment index of the initial feature extraction model based on the network parameters and the topology structure, and determine the optimization direction of the initial feature extraction model based on the loss function and the topology adjustment index.

[0136] S2232: When the optimization direction is parameter optimization, the loss value of the initial feature extraction model is calculated based on the predicted feature vector, and the network parameters of the initial feature extraction model are optimized according to the loss value.

[0137] S2233: When the optimization direction is topology optimization, determine the topology reconstruction strategy of the initial feature extraction model, and optimize the topology structure of the initial feature extraction model according to the topology reconstruction strategy, and use the topology reconstruction feedback mechanism to evaluate the performance of the optimized initial feature extraction model, and adjust the topology reconstruction strategy according to the evaluation results.

[0138] In this embodiment, when the computer device optimizes the feature extraction model, it can determine the loss function and topology adjustment index of the initial feature extraction model based on the network parameters and the topology structure, and determine the optimization direction of the initial feature extraction model based on the loss function and the topology adjustment index. The optimization direction here may include parameter optimization and topology optimization. When the optimization direction is parameter optimization, the computer device can calculate the loss value of the initial feature extraction model based on the predicted feature vector, and optimize the network parameters of the initial feature extraction model according to the loss value; when the optimization direction is topology optimization, the computer device can determine the topology reconstruction strategy of the initial feature extraction model, and optimize the topology structure of the initial feature extraction model according to the topology reconstruction strategy, and use the topology reconstruction feedback mechanism to evaluate the performance of the optimized initial feature extraction model, and adjust the topology reconstruction strategy according to the evaluation results.

[0139] Among them, the topology reconstruction strategy refers to the adjustment objects and adjustment steps used when optimizing and reconstructing the topology structure of the model, which can be set based on the model's computational efficiency, model robustness, model generalization ability and other goals. In the application, every time the topology structure of the model is reconstructed, the computer device can also re-evaluate the topology structure of the initial feature extraction model, understand its advantages and disadvantages, identify potential improvement points, and adjust the topology reconstruction strategy to improve the accuracy and efficiency of the model.

[0140] Specifically, the training process of the target feature extraction model can be as follows:

[0141] First, the computer device can initialize the network parameters and basic topology of the pre-trained neural network. The network parameters here can include weights w and biases b, and the initial weight matrix can be defined and the bias vector for:

[0142]

[0143]

[0144] In the formula, and Respectively represent the input dimension and output dimension of the network layer, Indicates randomly selecting a value within a specific interval.

[0145] Before each iteration, the pre-trained neural network can evaluate the requirements of the current training task for the network structure through the built-in topology sensor. Here, the topology sensor can be defined as a function T, which outputs the topology adjustment index based on the current loss function L and the network structure S. , which can be expressed as:

[0146]

[0147]

[0148] In the formula, Represents the Sigmoid activation function.

[0149] The gradient The calculation method can be expressed as:

[0150]

[0151] In the formula, Represents the prediction error, which is calculated by the preset Softmax function.

[0152] Next, the computer device can calculate the loss function L under the current network topology configuration. The loss function L is composed of two parts, namely the prediction error E and the topology adaptability penalty P. The calculation method can be expressed as:

[0153]

[0154]

[0155]

[0156] In the formula, represents the topology weight hyperparameter, represents the jth structural element, such as the change in connection weight. Can be set to 0.1.

[0157] Further, The calculation method can be expressed as:

[0158]

[0159] In the formula, and They represent the structural states in consecutive iterations respectively.

[0160] according to and L, the computer device can decide whether the optimization direction of the model is network parameter adjustment or topology structure adjustment. The adjustment method is realized through the decision function D, which can be expressed as:

[0161]

[0162] In the formula, represents the threshold value, which can be set to 0.5.

[0163] According to the adjustment of the execution network parameters or topology structure, the parameter adjustment rule can be expressed as:

[0164]

[0165]

[0166] In the formula, and Respectively represent the weights and biases in the adjusted network parameters, and Respectively represent the weights and biases in the network parameters before adjustment, represents the learning rate. Can be set to 0.01.

[0167] Furthermore, the gradient Calculated by the chain rule, it can be expressed as:

[0168]

[0169] Similarly, It can be calculated by the chain rule, which is expressed as:

[0170]

[0171] In the formula, represents the real data characteristics of the i-th sample data, Represents the predicted data features of the i-th sample data output by the preset Softmax function.

[0172] The computer device can then optimize the effectiveness of topology and parameter tuning through a feedback loop, specifically, adjusting or To optimize the learning process, it can be expressed as:

[0173]

[0174]

[0175] In the formula, and denote the adjusted learning rate and topology weight hyperparameters, respectively. and are the learning rate and topology weight hyperparameters before adjustment, and are adjustment factors. In this application, and They can be set to 0.9 and 0.01 respectively.

[0176] In addition, by adopting an adaptive topology reconstruction feedback mechanism, by monitoring network performance feedback and dynamically adjusting the model's topology reconstruction strategy, more sophisticated network optimization can be achieved. Specifically, the computer device can define a performance monitoring function R as a measure of the stability and accuracy of the model output, which can be expressed as:

[0177]

[0178] In the formula, Represents output stability, which is obtained by calculating the output variance, represents accuracy, measured by the inverse loss function value. and The influence weights of stability and accuracy can be adjusted. In this application, and Can be set to 0.6 and 0.4 respectively.

[0179] in, and The calculation method can be expressed as:

[0180]

[0181]

[0182] In the formula, for The average value of Indicates the maximum possible error. In this application, Can be set to 0.5.

[0183] Furthermore, an adaptive adjustment factor is used , dynamically adjust the topology according to the value of R, which can be expressed as:

[0184]

[0185] In the formula, Indicates the upper limit of the adjustment rate. Indicates the sensitivity of the adjustment. In this application, and Can be set to 0.1 and 5 respectively.

[0186] Furthermore, according to Adjust the topology, which can be expressed as:

[0187]

[0188] Where S is the topology before adjustment, For the adjusted topology, It is the topology adjustment direction derived based on gradient or other optimization algorithms.

[0189] Finally, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. Among them, the preset stop iteration condition is to reach the preset maximum number of iterations. The preset maximum number of iterations of this application can be set to 1000 times, which is not limited here.

[0190] In one embodiment, Figure 4 As shown, Figure 4 A schematic diagram of a process for determining a target feature dimensionality reduction model provided in an embodiment of the present application; Figure 4 The initial feature dimensionality reduction model includes an autoencoder, a microadjustable layer, and a decoder; wherein the process of determining the target feature dimensionality reduction model in step S120 may include:

[0191] S1231: Initialize the network parameters of the initial feature dimensionality reduction model, and use the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features.

[0192] S1232: Dimensionally adjust the initial low-dimensional features through a micro-adjustable layer to obtain final low-dimensional features, and input the final low-dimensional features into the decoder to obtain predicted high-dimensional features output by the decoder.

[0193] S1233: A dual loss function is used to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features, and the network parameters of the initial feature dimensionality reduction model are optimized based on the dimensionality loss value until the optimization round reaches a preset optimization round threshold, thereby generating a target feature dimensionality reduction model.

[0194] In this embodiment, when determining the target feature dimensionality reduction model, the computer device can first initialize the network parameters of the initial feature dimensionality reduction model, and use the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features. Then, the initial low-dimensional features can be dimensionally adjusted through a micro-adjustable layer to obtain the final low-dimensional features, and the final low-dimensional features can be input into the decoder to obtain the predicted high-dimensional features output by the decoder. Finally, the computer device can use the dual loss function to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features, and optimize the network parameters of the initial feature dimensionality reduction model based on the dimensionality loss value until the optimization round reaches the preset optimization round threshold to generate the target feature dimensionality reduction model.

[0195] Specifically, the training process of the target feature dimensionality reduction model can be as follows:

[0196] First, in the initialization stage, all weights w and biases b in the initial feature dimensionality reduction model can be initialized with random values. In this application, the computer device can be initialized with a Gaussian distribution, which can be specifically expressed as:

[0197]

[0198]

[0199] In the formula, represents the connection weight from the jth neuron to the ith neuron in the autoencoder; represents the bias of the i-th neuron; is the standard deviation, which is used to control the dispersion of the initialization. In this application, Can be set to 0.01.

[0200] Then, the computer device can extract the high-dimensional target data features extracted by the target feature extraction model. Input to the encoder, propagated through the activation function and multi-layer network structure, generating low-dimensional features of low-dimensional representation, each layer of the encoder outputs It can be defined as:

[0201]

[0202] In the formula, represents the output of the encoder layer l, is the Sigmoid activation function.

[0203] Furthermore, after the low-dimensional features enter the decoder, the decoder will try to reconstruct the original high-dimensional target data features, and the decoder output reconstruction for:

[0204]

[0205] In the formula, Represents the k-th layer input of the decoder.

[0206] Suppose the last layer output of the encoder is , a differentiable adjustment layer is used to adjust the encoder output for , which can be specifically expressed as:

[0207]

[0208] In the formula, represents element-wise multiplication; is a learnable adjustment factor; is a parameter adjusted according to the gradient information. In this application, Can be set to 0.3.

[0209] In a specific embodiment, the adjustment factor The calculation method can be expressed as:

[0210]

[0211] In the formula, Represents the feature after encoding from the kth to the weight of the adjustment factor; represents the bias of the adjustment factor; Represents the ReLU activation function.

[0212] The computer equipment can then calculate the total loss , which includes reconstruction loss and dual loss. For reconstruction loss, the computer device can use mean square error to measure the difference between the reconstructed data and the original data. For dual loss, it includes evaluating the matching degree between the encoded features and the predetermined low-dimensional target features to optimize the representation ability of the features. Specifically, the total loss The calculation method can be expressed as:

[0213]

[0214] In the formula, N represents the number of training samples; represents the i-th original data point; represents the i-th reconstructed data point; is the dual loss function; Represents the target low-dimensional space; represents the weight factor of the dual loss. In this application, Set to 0.4.

[0215] Among them, the dual loss function Can be used to measure the encoded features and the target low-dimensional space The degree of matching between them can be calculated as:

[0216]

[0217] In the formula, Represents the i-th feature dimension after encoding; Represents the i-th feature dimension in the target low-dimensional space.

[0218] Secondly, the computer device can use the gradient descent method to update the network parameters according to the total loss function. In particular, adjusting the parameters according to the gradient information in the micro-adjustable layer can achieve better information compression and feature expression. Specifically, let the gradient and Used to update parameters, the update method of the autoencoder parameters can be expressed as:

[0219]

[0220]

[0221] In the formula, and represents the updated autoencoder weights and bias parameters, and Represents the autoencoder weights and bias parameters before updating; Indicates the learning rate. In this application, Can be set to 0.01.

[0222] Among them, the gradient of the weight The calculation method can be expressed as:

[0223]

[0224] In the formula, represents the partial derivative of the loss function with respect to the model output, Represents the partial derivative of the model output with respect to the weight.

[0225] Similarly, the gradient of the bias The calculation method can be expressed as:

[0226]

[0227] In the formula, is a constant term and can be set to 1 in this application.

[0228] Finally, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. Among them, the preset stop iteration condition is to reach the preset maximum number of iterations. In this application, the preset maximum number of iterations is set to 1000 times, which is not limited here.

[0229] In one embodiment, the process of determining the target classifier in step S140 may include:

[0230] S141: Initialize the network parameters of the extreme learning machine and determine the minimization loss function of the extreme learning machine.

[0231] S142: Input the pre-acquired sample training data into the extreme learning machine to obtain the predicted value output by the extreme learning machine.

[0232] S143: Based on the predicted value, the network parameters of the extreme learning machine are optimized by minimizing the loss function until the optimization round reaches a preset optimization round threshold, thereby generating a target classifier.

[0233] In this embodiment, when determining the target classifier, the computer device can first initialize the network parameters of the extreme learning machine and determine the minimized loss function of the extreme learning machine, and then input the pre-acquired sample training data into the extreme learning machine to obtain the predicted value output by the extreme learning machine, so that the computer device can optimize the network parameters of the extreme learning machine based on the predicted value by minimizing the loss function until the optimization round reaches a preset optimization round threshold, thereby generating a target classifier.

[0234] Specifically, the training process of the target classifier can be as follows:

[0235] First, obtain the target feature vector obtained by reducing the dimension of the target dimensionality reduction model, and further process the target feature vector using variational mode decomposition to decompose multiple modal components as input data for the extreme learning machine. Each component obtained by variational mode decomposition represents a specific feature or trend of the data. To elaborate, the present application can regard the target feature vector after feature dimensionality reduction as a time series signal, and use variational mode decomposition to decompose the target feature vector q into a series of modal components , where k represents the number of modes, and each modal component It can be obtained by solving the following optimization problem:

[0236]

[0237] In the formula, represents the Hilbert transform, j represents the imaginary unit, represents the center frequency of the kth modal component, represents the Lagrange multiplier.

[0238] Among them, the center frequency It can be calculated by analyzing the instantaneous frequency of the signal, and the calculation method is:

[0239]

[0240] In the formula, represents the instantaneous frequency, calculated as the derivative of the signal after the Hilbert transform, and T represents the total time length of the signal.

[0241] Secondly, initialize the weights and biases of the extreme learning machine. In this application, the weights W and biases b of the extreme learning machine can be initialized to random values, and the dimensions of the weight matrix and the bias vector are adapted to the number of nodes in the input layer and the hidden layer, respectively, which can be specifically expressed as:

[0242]

[0243]

[0244] Where N represents normal distribution and U represents uniform distribution.

[0245] In addition, the computer device can also set an initial dynamic forgetting factor. Used to adjust the learning rate during training, based on the error Dynamic update can be specifically expressed as:

[0246]

[0247] In the formula, is the adjustment factor and e(t) is the classification error at the tth iteration.

[0248] Among them, the adjustment factor of the dynamic forgetting factor The setting is optimized by gradient descent method to ensure fast convergence while avoiding excessive oscillation. Its calculation method can be expressed as:

[0249]

[0250] In the formula, represents the loss function, Indicates the learning rate. In this application, Can be set to 0.01.

[0251] Then, the computer device inputs the target feature vector after the variational mode decomposition into the extreme learning machine for training. The output y of the extreme learning machine can be calculated by the activation function g, which is specifically expressed as:

[0252]

[0253] In the formula, Represents a Sigmoid activation function based on steepness.

[0254] Among them, the calculation method of the Sigmoid activation function based on steepness can be expressed as:

[0255]

[0256] In the formula, Indicates the parameter for adjusting the steepness of the curve. In this application, Can be set to 0.3.

[0257] Furthermore, the computer device can continuously optimize the classifier parameters, including weights and biases, during the training process by minimizing the loss function Lu:

[0258]

[0259] In the formula, represents the target output of the i-th training sample, N represents the number of samples, represents the regularization parameter.

[0260] Among them, the regularization parameter It can be used to control model complexity and prevent model overfitting. In this application, its value is usually selected based on the results of cross-validation and can be expressed as:

[0261]

[0262] In the formula, represents the average error calculated by k-fold cross validation, Represents the minimum value function.

[0263] Finally, the above steps are repeated until the preset stop iteration condition is met, indicating that the model training is completed. Among them, the preset stop iteration condition is to reach the preset maximum number of iterations. The preset maximum number of iterations in this application is set to 1000 times, which is not limited here.

[0264] The following is a description of the equipment operation efficiency evaluation device provided in an embodiment of the present application. The equipment operation efficiency evaluation device described below and the equipment operation efficiency evaluation method described above can be referenced to each other.

[0265] In one embodiment, Figure 5 As shown, Figure 5 A schematic diagram of the structure of a device operation efficiency evaluation device provided in an embodiment of the present application; the present application also provides a device operation efficiency evaluation device, including a data acquisition module 210, a model determination module 220, a feature extraction module 230 and an efficiency evaluation module 240, specifically including the following:

[0266] The data acquisition module 210 is used to acquire the operating data of the device to be evaluated.

[0267] The model determination module 220 is used to determine the target feature extraction model and the target feature dimensionality reduction model; wherein the target feature extraction is obtained by training the preset initial feature extraction model based on the sample operation data obtained by expanding the quantum optimization algorithm, using the topological dynamic adjustment mechanism; the target feature dimensionality reduction model is obtained by training the preset initial feature dimensionality reduction model using the dual loss function.

[0268] The feature extraction module 230 is used to extract features from the operating data using the target feature extraction model to obtain target data features, and input the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model.

[0269] The performance evaluation module 240 is used to decompose the target feature vector through a variational model, and input the decomposed target feature vector into a predetermined target classifier to obtain the operating performance evaluation result of the device to be evaluated output by the target classifier.

[0270] In the above embodiment, after obtaining the operation data of the device to be evaluated, the target feature extraction model and the target feature dimensionality reduction model can be determined. Among them, the target feature extraction model is based on the sample operation data obtained by the quantum optimization algorithm, and is trained by the topological dynamic adjustment mechanism. The sample operation data obtained by the quantum optimization algorithm is diverse, and the topological dynamic adjustment mechanism can better adapt to the differences between different samples during the model training process, and can identify and delete redundant connections in the model. Therefore, the feature extraction efficiency of the model trained in this way is high, and the generalization ability on unseen samples can be improved, thereby improving the accuracy of the evaluation results; and the target feature dimensionality reduction model is trained by the dual loss function. Compared with the traditional single loss function, the dual loss function may evaluate the performance of the model from multiple angles or levels, thereby providing a more comprehensive and accurate evaluation result. Therefore, using these two models, it is possible to efficiently generate a target feature vector with high quality in the operation data. Finally, the present application can decompose the target feature vector through the variational mode to mine the potential information of the target feature vector, so that when the target classifier is used to evaluate the equipment operation efficiency of the decomposed target feature vector, the accuracy and timeliness of the evaluation results can be greatly improved.

[0271] In one embodiment, the model determination module 220 may include:

[0272] The data collection submodule is used to collect the actual operation data generated by multiple devices to be evaluated during operation.

[0273] The model determination submodule is used to determine the target data expansion model; the target data expansion model is obtained by training the preset initial data expansion model using the quantum optimization algorithm and the quality evaluation function.

[0274] The model expansion submodule is used to input the real operating data into the target data expansion model to obtain the expanded operating data output by the target data expansion model.

[0275] The type labeling submodule is used to determine the performance type of each real operation data, and to label each real operation data and the corresponding extended operation data based on each performance type to form sample operation data.

[0276] In one embodiment, the model determination submodule may include:

[0277] The parameter initialization unit is used to initialize the network parameters of the generator and the discriminator in the generative adversarial network to form an initial data expansion model, and to perform quantum state optimization on the initial data expansion model to generate quantum parameters.

[0278] The data prediction unit is used to generate predicted operation data of the real operation data collected in advance based on the quantum parameters by using the generator, and to use the discriminator to distinguish the authenticity of the real operation data and the predicted operation data to obtain a distinction result.

[0279] The parameter optimization unit is used to use a quality evaluation function to perform quality evaluation on the predicted operation data, obtain evaluation results, and based on the evaluation results, use the predicted operation data and the discrimination results to optimize the network parameters of the initial data expansion model until the optimization round reaches a preset optimization round threshold, thereby generating a target data expansion model.

[0280] In one embodiment, the model determination module 220 may further include:

[0281] The model initialization submodule is used to initialize the network parameters and topology structure of the pre-trained neural network to form an initial feature extraction model.

[0282] The model prediction submodule is used to input the pre-acquired sample operation data into the initial feature extraction model to obtain the predicted data features output by the initial feature extraction model.

[0283] The model optimization submodule is used to optimize the network parameters or topology structure of the initial feature extraction model with the goal of making the predicted data features close to the real data features of the sample data features.

[0284] The model generation submodule is used to use the optimized initial feature extraction model as the target feature extraction model when the optimization round of the initial feature extraction model reaches a preset optimization round threshold.

[0285] In one embodiment, the model optimization submodule may include:

[0286] The optimization direction determination unit is used to determine the loss function and topology adjustment index of the initial feature extraction model based on the network parameters and the topology structure, and to determine the optimization direction of the initial feature extraction model based on the loss function and the topology adjustment index.

[0287] The parameter optimization unit is used to calculate the loss value of the initial feature extraction model based on the predicted feature vector when the optimization direction is parameter optimization, and optimize the network parameters of the initial feature extraction model according to the loss value.

[0288] The topology optimization unit is used to determine the topology reconstruction strategy of the initial feature extraction model when the optimization direction is topology optimization, and optimize the topology structure of the initial feature extraction model according to the topology reconstruction strategy, and use the topology reconstruction feedback mechanism to evaluate the performance of the optimized initial feature extraction model, and adjust the topology reconstruction strategy according to the evaluation results.

[0289] In one embodiment, the model determination module 220 may further include:

[0290] The sub-data dimensionality reduction module is used to initialize the network parameters of the initial feature dimensionality reduction model, and use the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features.

[0291] The dimension adjustment submodule is used to adjust the dimension of the initial low-dimensional features through a micro-adjustment layer to obtain the final low-dimensional features, and input the final low-dimensional features into the decoder to obtain the predicted high-dimensional features output by the decoder.

[0292] The model optimization submodule is used to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features using the dual loss function, and optimize the network parameters of the initial feature dimensionality reduction model based on the dimensionality loss value until the optimization round reaches the preset optimization round threshold, thereby generating a target feature dimensionality reduction model.

[0293] In one embodiment, the performance evaluation module 240 may include:

[0294] The parameter initialization submodule is used to initialize the network parameters of the extreme learning machine and determine the minimized loss function of the extreme learning machine.

[0295] The learning machine prediction submodule is used to input the pre-acquired sample training data into the extreme learning machine to obtain the predicted value output by the extreme learning machine.

[0296] The classifier generation submodule is used to optimize the network parameters of the extreme learning machine based on the predicted value by minimizing the loss function until the optimization round reaches a preset optimization round threshold, thereby generating a target classifier.

[0297] In one embodiment, the present application also provides a storage medium, in which computer-readable instructions are stored. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the device operation efficiency evaluation method as described in any of the above embodiments.

[0298] In one embodiment, the present application also provides a computer device having computer-readable instructions stored therein. When the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the device operation efficiency evaluation method as described in any of the above embodiments.

[0299] Indicatively, Figure 6 As shown, Figure 6 This is a schematic diagram of the internal structure of a computer device provided in an embodiment of the present application. The computer device 300 may be provided as a server. Figure 6 , the computer device 300 includes a processing component 302, which further includes one or more processors, and a memory resource represented by a memory 301, for storing instructions executable by the processing component 302, such as an application. The application stored in the memory 301 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 302 is configured to execute instructions to perform the device operation performance evaluation method of any of the above embodiments.

[0300] The computer device 300 may further include a power supply component 303 configured to perform power management of the computer device 300, a wired or wireless network interface 304 configured to connect the computer device 300 to a network, and an input / output (I / O) interface 305. The computer device 300 may operate based on an operating system stored in the memory 301, such as Windows Server TM, Mac OS X TM, Unix TM, Linux TM, Free BSD TM, or the like.

[0301] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0302] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0303] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

[0304] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating equipment operation efficiency, characterized in that: The method comprises: Obtain the operating data of the equipment to be evaluated; Determine a target feature extraction model and a target feature dimensionality reduction model; wherein the target feature extraction model is obtained by training a preset initial feature extraction model using a topological dynamic adjustment mechanism based on sample operation data expanded by a quantum optimization algorithm; and the target feature dimensionality reduction model is obtained by training a preset initial feature dimensionality reduction model using a dual loss function; Using the target feature extraction model to extract features from the operating data to obtain target data features, and inputting the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model; Decomposing the target feature vector through a variational model, and inputting the decomposed target feature vector into a predetermined target classifier to obtain an operating efficiency evaluation result of the device to be evaluated output by the target classifier; The expansion process of the sample operation data includes: Collect the real operation data generated by multiple devices to be evaluated during operation; Determine a target data expansion model; the target data expansion model is obtained by training a preset initial data expansion model using a quantum optimization algorithm and a quality evaluation function; Inputting the real operation data into the target data expansion model to obtain the expanded operation data output by the target data expansion model; Determine the performance type of each real operating data, and perform type annotation on each real operating data and the corresponding extended operating data based on each performance type to form sample operating data; Wherein, determining the target data expansion model includes: Initializing network parameters of a generator and a discriminator in a generative adversarial network to form an initial data expansion model, and performing quantum state optimization on the initial data expansion model to generate quantum parameters; Using the generator to generate predicted operation data of the real operation data collected in advance based on the quantum parameters, and using the discriminator to discriminate the authenticity of the real operation data and the predicted operation data to obtain a discrimination result; Using a quality assessment function to perform quality assessment on the predicted operation data to obtain an assessment result, and based on the assessment result, using the predicted operation data and the discrimination result to optimize the network parameters of the initial data expansion model until the optimization round reaches a preset optimization round threshold, thereby generating a target data expansion model; The process of performing quantum state optimization on the initial data expansion model includes: The quantum state combination optimization module is used to represent the potential data state through quantum bits, and the Hamiltonian operator H is used to optimize the data state; wherein the quantum state The specific formula is: ; Where N represents the total number of quantum ground states, represents the i-th quantum ground state, represents the corresponding complex probability amplitude; Optimizing complex probability amplitudes via quantum algorithms To minimize the expected value of energy, the expected value of energy The specific formula is: ; In the formula, Represents the complex probability amplitude The complex conjugate of , H represents the Hamiltonian operator; represents the complex probability amplitude corresponding to the jth quantum ground state; The calculation process of decomposing the conversion elements of the system Hamiltonian operator H between different quantum ground states is as follows: ; In the formula, and Represents quantum states and The wave function expression at position r; The formula of the quality assessment function Q is specifically: ; In the formula, and represents the weight factor for adjusting diversity and authenticity, Var() represents the variance of the augmented running data, Diversity() represents the diversity measure of the augmented running data, and G(z) represents the augmented running data; The process of training the preset initial feature extraction model using the topology dynamic adjustment mechanism includes: Before each iteration, the network structure S of the current training task is evaluated through the function T in the topology sensor built into the initial feature extraction model, and the topology adjustment index is output based on the current loss function L and network structure S. , the specific formula is: ; ; In the formula, Represents the Sigmoid activation function; Identify the gradient of the loss function L and the network structure S; The loss function L includes the prediction error E and the topological adaptability penalty P. The specific formula is: ; ; ; In the formula, represents the topology weight hyperparameter, represents the jth structural element, including the change in connection weights; represents the real data characteristics of the i-th sample data, represents the predicted data feature of the i-th sample data output by the preset Softmax function; n represents the total number of sample data; m represents the total number of structural elements; Adjust the metrics based on the loss function L and topology Determine the optimization direction of the initial feature extraction model, the specific formula is: ; In the formula, D() represents the decision function; Indicates the threshold value; The network parameters or topology structure are adjusted according to the optimization direction, and the topology structure adjustment rules are as follows: The performance monitoring function R is defined as a measure of the stability and accuracy of the model output. The specific formula is: ; In the formula, Represents output stability, which is obtained by calculating the output variance; Indicates accuracy, measured by the inverse loss function value; and Indicates the influence weight of adjusting stability and accuracy; Adaptive adjustment factor , dynamically adjust the topology structure according to the value of the performance monitoring function R. The specific formula is: ; In the formula, Indicates the upper limit of the adjustment rate; Indicates the sensitivity of the adjustment; According to the adaptive adjustment factor Adjust the topology structure. The specific formula is: ; In the formula, This is the topology before adjustment. For the adjusted topology, It is the topological adjustment direction derived based on gradient or other optimization algorithms; Repeat the iteration until the preset stop iteration condition is met, indicating that the model training is completed; The initial feature dimensionality reduction model includes an autoencoder, a microadjustable layer and a decoder; the determination process of the target feature dimensionality reduction model includes: Initializing the network parameters of the initial feature dimensionality reduction model, and using the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features; Performing dimension adjustment on the initial low-dimensional features through the micro-adjustable layer to obtain final low-dimensional features, and inputting the final low-dimensional features into the decoder to obtain predicted high-dimensional features output by the decoder; A dual loss function is used to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features, and the network parameters of the initial feature dimensionality reduction model are optimized based on the dimensionality loss value until the optimization round reaches a preset optimization round threshold, thereby generating a target feature dimensionality reduction model.

2. The equipment operation efficiency evaluation method according to claim 1, characterized in that: The determining of the target feature extraction model comprises: Initialize the network parameters and topological structure of the pre-trained neural network to form an initial feature extraction model; Inputting the pre-acquired sample operation data into the initial feature extraction model to obtain the predicted data features output by the initial feature extraction model; With the goal of making the predicted data features approach the real data features of the sample running data, optimizing the network parameters or topological structure of the initial feature extraction model; When the optimization round of the initial feature extraction model reaches a preset optimization round threshold, the optimized initial feature extraction model is used as the target feature extraction model.

3. The equipment operation efficiency evaluation method according to claim 2 is characterized in that: The optimizing of the network parameters or topological structure of the initial feature extraction model comprises: Determining a loss function and a topology adjustment index of the initial feature extraction model based on the network parameters and the topology structure, and determining an optimization direction of the initial feature extraction model based on the loss function and the topology adjustment index; When the optimization direction is parameter optimization, a loss value of the initial feature extraction model is calculated based on the predicted feature vector, and a network parameter of the initial feature extraction model is optimized according to the loss value; When the optimization direction is topology optimization, the topology reconstruction strategy of the initial feature extraction model is determined, and the topology structure of the initial feature extraction model is optimized according to the topology reconstruction strategy. In addition, the performance of the optimized initial feature extraction model is evaluated using a topology reconstruction feedback mechanism, and the topology reconstruction strategy is adjusted according to the evaluation results.

4. The equipment operation efficiency evaluation method according to claim 1, characterized in that: The process of determining the target classifier includes: Initializing network parameters of an extreme learning machine and determining a minimized loss function of the extreme learning machine; Inputting the pre-acquired sample training data into the extreme learning machine to obtain a predicted value output by the extreme learning machine; Based on the predicted value, the network parameters of the extreme learning machine are optimized using the minimization loss function until the optimization round reaches a preset optimization round threshold, thereby generating a target classifier.

5. A device for evaluating equipment operation efficiency, characterized in that: include: A data acquisition module, used to acquire the operating data of the equipment to be evaluated; A model determination module is used to determine a target feature extraction model and a target feature dimensionality reduction model; wherein the target feature extraction model is obtained by training a preset initial feature extraction model based on sample operation data expanded by a quantum optimization algorithm using a topological dynamic adjustment mechanism; and the target feature dimensionality reduction model is obtained by training a preset initial feature dimensionality reduction model using a dual loss function; A feature extraction module is used to extract features from the operating data using the target feature extraction model to obtain target data features, and input the target data features into the target feature dimensionality reduction model to obtain a target feature vector output by the target feature dimensionality reduction model; An efficiency evaluation module, used for decomposing the target feature vector through a variational mode, and inputting the decomposed target feature vector into a predetermined target classifier to obtain an operation efficiency evaluation result of the device to be evaluated output by the target classifier; Wherein, the model determination module includes: Collect the real operation data generated by multiple devices to be evaluated during operation; Determine a target data expansion model; the target data expansion model is obtained by training a preset initial data expansion model using a quantum optimization algorithm and a quality evaluation function; Inputting the real operation data into the target data expansion model to obtain the expanded operation data output by the target data expansion model; Determine the performance type of each real operating data, and perform type annotation on each real operating data and the corresponding extended operating data based on each performance type to form sample operating data; Wherein, the model determination module also includes: Initializing network parameters of a generator and a discriminator in a generative adversarial network to form an initial data expansion model, and performing quantum state optimization on the initial data expansion model to generate quantum parameters; Using the generator to generate predicted operation data of the real operation data collected in advance based on the quantum parameters, and using the discriminator to discriminate the authenticity of the real operation data and the predicted operation data to obtain a discrimination result; Using a quality assessment function to perform quality assessment on the predicted operation data to obtain an assessment result, and based on the assessment result, using the predicted operation data and the discrimination result to optimize the network parameters of the initial data expansion model until the optimization round reaches a preset optimization round threshold, thereby generating a target data expansion model; The process of the model determination module performing quantum state optimization on the initial data expansion model includes: The quantum state combination optimization module is used to represent the potential data state through quantum bits, and the Hamiltonian operator H is used to optimize the data state; wherein the quantum state The specific formula is: ; Where N represents the total number of quantum ground states, represents the i-th quantum ground state, represents the corresponding complex probability amplitude; Optimizing complex probability amplitudes via quantum algorithms To minimize the expected value of energy, the expected value of energy The specific formula is: ; In the formula, Represents the complex probability amplitude The complex conjugate of , H represents the Hamiltonian operator; represents the complex probability amplitude corresponding to the jth quantum ground state; The calculation process of decomposing the conversion elements of the system Hamiltonian operator H between different quantum ground states is as follows: ; In the formula, and Represents quantum states and The wave function expression at position r; The formula of the quality assessment function Q is specifically: ; In the formula, and represents the weight factor for adjusting diversity and authenticity, Var() represents the variance of the augmented running data, Diversity() represents the diversity measure of the augmented running data, and G(z) represents the augmented running data; The process of the model determination module training the preset initial feature extraction model by adopting the topology dynamic adjustment mechanism includes: Before each iteration, the network structure S of the current training task is evaluated through the function T in the topology sensor built into the initial feature extraction model, and the topology adjustment index is output based on the current loss function L and network structure S. , the specific formula is: ; ; In the formula, Represents the Sigmoid activation function; Identify the gradient of the loss function L and the network structure S; The loss function L includes the prediction error E and the topological adaptability penalty P. The specific formula is: ; ; ; In the formula, represents the topology weight hyperparameter, represents the jth structural element, including the change in connection weights; represents the real data characteristics of the i-th sample data, represents the predicted data feature of the i-th sample data output by the preset Softmax function; n represents the total number of sample data; m represents the total number of structural elements; Adjust the metrics based on the loss function L and topology Determine the optimization direction of the initial feature extraction model, the specific formula is: ; In the formula, D() represents the decision function; Indicates the threshold value; The network parameters or topology structure are adjusted according to the optimization direction, and the topology structure adjustment rules are as follows: The performance monitoring function R is defined as a measure of the stability and accuracy of the model output. The specific formula is: ; In the formula, Represents output stability, which is obtained by calculating the output variance; Indicates accuracy, measured by the inverse loss function value; and Indicates the influence weight of adjusting stability and accuracy; Adaptive adjustment factor , dynamically adjust the topology structure according to the value of the performance monitoring function R. The specific formula is: ; In the formula, Indicates the upper limit of the adjustment rate; Indicates the sensitivity of the adjustment; According to the adaptive adjustment factor Adjust the topology structure. The specific formula is: ; In the formula, This is the topology before adjustment. For the adjusted topology, It is the topological adjustment direction derived based on gradient or other optimization algorithms; Repeat the iteration until the preset stop iteration condition is met, indicating that the model training is completed; The initial feature dimensionality reduction model includes an autoencoder, a micro-adjustable layer and a decoder; the process of determining the target feature dimensionality reduction model by the model determination module includes: Initializing the network parameters of the initial feature dimensionality reduction model, and using the autoencoder to reduce the dimension of the pre-acquired sample data features to generate initial low-dimensional features; Performing dimension adjustment on the initial low-dimensional features through the micro-adjustable layer to obtain final low-dimensional features, and inputting the final low-dimensional features into the decoder to obtain predicted high-dimensional features output by the decoder; A dual loss function is used to calculate the dimensionality loss value between the predicted high-dimensional features and the sample data features, and the network parameters of the initial feature dimensionality reduction model are optimized based on the dimensionality loss value until the optimization round reaches a preset optimization round threshold, thereby generating a target feature dimensionality reduction model.

6. A storage medium, characterized in that: The storage medium stores computer-readable instructions, and when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the device operation efficiency evaluation method as described in any one of claims 1 to 4.

7. A computer device, characterized in that: include: one or more processors, and memory; The memory stores computer-readable instructions, and when the computer-readable instructions are executed by the one or more processors, the steps of the device operation efficiency evaluation method according to any one of claims 1 to 4 are performed.

Citation Information

Patent Citations

  • Recommendation method and system based on adaptive noise reduction training, electronic equipment and medium

    CN114254187A

  • Equipment state prediction method and device based on artificial intelligence

    CN118171180A

  • Equipment health degree assessment method and equipment based on artificial intelligence, and medium

    CN118296511A