A Model-Driven Method and Device for Evaluating Equipment Health Level

Abstract

The present invention provides a method and device for evaluating the health level of a device based on model-driven, which relates to the technical field of device health level evaluation. After extracting features from a test sample corresponding to the operation monitoring data of a preset device through a pre-constructed feature extraction model, a classifier model based on a deep neural decision tree algorithm is used for classification prediction to determine the health state of the device. Among them, the network parameters of the feature extraction model are updated based on the competition strategy of species in the ecosystem simulated according to the allocated niche. The decision branches of the decision tree of the deep neural decision tree algorithm are determined based on the states of qubits of the nodes of the decision tree, and the states of the qubits are calculated by simulating the information gain during the decision tree splitting process. Based on this, the present invention can adaptively adjust the model parameters according to the change of data distribution, accurately capture the complex dependence relationship between features, and improve the classification accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment health level assessment, and in particular, to a method and device for equipment health level assessment based on model driving. Background Art

[0002] With the continuous increase in the scale and complexity of industrial equipment, the difficulty of equipment maintenance and management has also increased accordingly. The normal operation of large-scale equipment is crucial for the stability and efficiency of industrial production. However, equipment failures often have concealment, and serious economic losses will be caused when failures occur. To avoid the adverse effects brought by sudden equipment failures, preventive maintenance and equipment health assessment have become research hotspots in the industrial field. Traditional equipment health assessment mainly relies on expert experience or simple threshold judgment. However, with the increase in the complexity of equipment operation data, these methods are unable to cope and are difficult to accurately predict the health status of equipment.

[0003] When traditional methods process equipment operation data, they usually rely on pre-defined feature extraction methods, lacking adaptability and flexibility. This results in less than ideal feature extraction effects when dealing with different data distributions. Traditional feature extraction methods are also prone to falling into local optimal solutions and unable to globally optimize feature representations. In addition, when traditional classifiers process complex data structures, they are limited by their processing capabilities and expression capabilities and are difficult to comprehensively capture the complex relationships in the data. Based on this, the classification accuracy of traditional models on complex data is often not high. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide a method and device for equipment health level assessment based on model driving, which can adaptively adjust model parameters according to changes in data distribution, accurately capture the complex dependence relationships between features, and improve classification accuracy.

[0005] In a first aspect, an embodiment of the present invention provides a method for evaluating the health level of a device based on model driving. The method includes: obtaining operation monitoring data of a preset device, converting the text data in the operation monitoring data into a vector format, and generating a test sample of the preset device; the operation monitoring data is collected from the operation log and sensor data of the preset device; extracting features from the test sample through a pre-constructed feature extraction model to obtain target features corresponding to the test sample; wherein, the network parameters of the feature extraction model are updated based on the assigned niche, and the niche of the network parameters is determined by simulating the competition strategy of species in an ecosystem; inputting the target features into a preset classifier model, and making a classification decision on the target features through the classifier model to output a decision result; wherein, the classifier model is based on a deep neural decision tree algorithm as the classification algorithm, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the states of the qubits, and the states of the qubits are determined by simulating the information gain calculation in the decision tree splitting process; based on the decision result, evaluating the health state of the preset device.

[0006] In combination with the first aspect, an embodiment of the present invention further provides a first implementation manner of the first aspect. Among them, the step of evaluating the health state of the preset device based on the decision result includes: obtaining the decision label indicated by the decision result; determining the device state indicated by the decision label as the health state of the preset device.

[0007] In combination with the first aspect, an embodiment of the present invention further provides a second implementation manner of the first aspect. Among them, the step of determining the state of the qubit by simulating the information gain calculation in the decision tree splitting process includes: initializing the nodes of the decision tree of the preset deep neural decision tree algorithm as qubits; training through the preset training sample set with the deep neural decision tree algorithm, and determining the rotation parameters of the custom rotation gate corresponding to the training sample set through the exponential operation in the complex number domain; adjusting the entanglement and phase between the qubits based on the rotation parameters and the preset quantum logic gate operation; performing quantum measurement on the qubits based on the adjusted entanglement and phase between the qubits to determine the probability distribution corresponding to the qubits, so as to evaluate the state of the qubits.

[0008] In combination with the first aspect, an embodiment of the present invention further provides a third implementation manner of the first aspect. Among them, the step of generating the decision branches of the decision tree based on the states of the qubits includes: calculating the decision conditions corresponding to the training sample set according to the states of the qubits; controlling the node splitting operation of the decision tree through a preset decision function based on the decision conditions to generate the decision branches of the decision tree.

[0009] Combined with the first aspect, an embodiment of the present invention further provides a fourth implementation manner of the first aspect. Among them, the steps of determining the ecological niche based on the competition strategy of species in the simulated ecosystem and updating the network parameters of the feature extraction model based on the allocated ecological niche include: training a preset feature extraction model through a preset training sample set; wherein, the network parameters of the feature extraction model are allocated with initial ecological niches; calculating the loss function value corresponding to the network parameters of the feature extraction model, using the loss function value as the adaptability score corresponding to the network parameters, and calculating the energy regulation factor corresponding to the network parameters; based on the energy regulation factor, calculating the dynamic mutation rate corresponding to the network parameters; migrating the initial ecological niche based on the dynamic mutation rate, thereby updating the network parameters.

[0010] Combined with the first aspect, an embodiment of the present invention further provides a fifth implementation manner of the first aspect. Among them, the steps of calculating the dynamic mutation rate corresponding to the network parameters based on the energy regulation factor include: calculating the mutation energy corresponding to the network parameters according to the energy regulation factor and the preset maximum energy input; calculating the adaptability score based on the mutation energy and the preset migration and mutation regulation coefficient to determine the dynamic mutation rate corresponding to the network parameters.

[0011] Combined with the first aspect, an embodiment of the present invention further provides a sixth implementation manner of the first aspect. Among them, the method further includes: performing dimensionality reduction processing on the target feature through a pre-constructed dimensionality reduction model; wherein, the construction method of the dimensionality reduction model includes: inputting a preset training sample set into a preset autoencoder neural network, mapping the training sample set to an initial low-dimensional feature space through the encoder of the autoencoder neural network to obtain initial low-dimensional features; determining the feature importance of the initial low-dimensional features based on the loss function value corresponding to the initial low-dimensional features; recursively optimizing the feature weight vector of the autoencoder neural network based on the feature importance; determining new low-dimensional features based on the feature weight vector, and performing feature reconstruction on the new low-dimensional features through the decoder; until the autoencoder neural network meets the preset iteration conditions, constructing a dimensionality reduction model based on the autoencoder neural network.

[0012] Combined with the first aspect, an embodiment of the present invention further provides a seventh implementation manner of the first aspect. Among them, the method further includes: obtaining pre-collected device operation monitoring samples; the device operation monitoring samples include the operation logs and sensor data of a preset device; annotating the operation monitoring samples based on the device health state characterized by the operation monitoring samples, and converting the text data in the operation monitoring samples into a vector format to construct an initial sample set; using a pre-constructed data augmentation model to perform data augmentation on the initial sample set to generate augmented samples of the initial sample set; wherein, the data augmentation model includes a generative adversarial network algorithm, and the generator parameters and discriminator parameters of the generative adversarial network algorithm are adjusted based on the dynamic strategy of game theory; constructing a training sample set based on the augmented samples and the initial sample set.

[0013] Combined with the first aspect, an eighth implementation manner of the first aspect is further provided in the embodiments of the present invention. Among them, the steps of generating the generator parameters and discriminator parameters of the dynamic strategy adjustment generative adversarial network algorithm based on game theory include: adjusting the feature weights of a preset training sample set through an adaptive feature transformation strategy, and expanding the data of the training sample set with adjusted feature weights through the generator of the generative adversarial network algorithm to generate initial expanded samples; evaluating the initial expanded samples through the discriminator of the generative adversarial network algorithm to determine the initial loss value of the generator; calculating the loss function of the generator according to the initial loss value and the entropy of the initial expanded samples; and calculating the loss function of the discriminator based on a preset regularization term; adjusting the generator parameters of the generator and the discriminator parameters of the discriminator respectively based on the loss function of the generator, the loss function of the discriminator, and a preset dynamic game strategy.

[0014] In a second aspect, an apparatus for evaluating the health level of a device based on model driving is further provided in the embodiments of the present invention. The apparatus includes: a data acquisition module, configured to acquire the operation monitoring data of a preset device, convert the text data in the operation monitoring data into a vector format, and generate a to-be-tested sample of the preset device; the operation monitoring data is collected from the operation logs and sensor data of the preset device; a data processing module, configured to extract features from the to-be-tested sample through a pre-constructed feature extraction model to obtain target features corresponding to the to-be-tested sample; wherein, the network parameters of the feature extraction model are updated based on the assigned niche, and the niche of the network parameters is determined by simulating the competition strategy of species in an ecosystem; an execution module, configured to input the target features into a preset classifier model, make a classification decision on the target features through the classifier model, and output a decision result; wherein, the classifier model is based on a deep neural decision tree algorithm as a classifier algorithm, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the states of the qubits, and the states of the qubits are determined by simulating the information gain calculation in the decision tree splitting process; an output module, configured to evaluate the health state of the preset device based on the decision result.

[0015] The embodiments of the present invention bring the following beneficial effects: The embodiments of the present invention provide a method and device for evaluating the health level of a device based on model driving. After extracting features from a test sample corresponding to the operation monitoring data of a preset device through a pre-constructed feature extraction model, a classifier model is used for classification prediction to determine the health status of the device. Among them, the network parameters of the feature extraction model are updated based on the allocated ecological niche, and the ecological niche of the network parameters is determined by simulating the competition strategy of species in an ecosystem. The classifier model is based on the deep neural decision tree algorithm as the classifier algorithm. The nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the states of the qubits. The states of the qubits are determined by simulating the information gain calculation in the decision tree splitting process. The embodiments of the present invention can adaptively adjust the model parameters according to the changes in data distribution, accurately capture the complex dependence relationships between features, and improve the classification accuracy.

[0016] Other features and advantages of the present invention will be described in the following description, and some of them will become obvious from the description or be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the description and the drawings.

[0017] To make the above objectives, features, and advantages of the present invention more obvious and understandable, the following specifically provides preferred embodiments and, in conjunction with the accompanying drawings, the detailed description is as follows. Brief Description of the Drawings

[0018] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0019] Figure 1 It is a flowchart of a method for evaluating the health level of a device based on model driving provided by an embodiment of the present invention;

[0020] Figure 2 It is a flowchart of another method for evaluating the health level of a device based on model driving provided by an embodiment of the present invention;

[0021] Figure 3 It is a schematic diagram of the training process of a dimensionality reduction model provided by an embodiment of the present invention;

[0022] Figure 4 It is a flowchart of a method for constructing a training sample set provided by an embodiment of the present invention;

[0023] Figure 5Schematic diagram of a device health level evaluation device based on model-driven provided by an embodiment of the present invention;

[0024] Figure 6 Schematic diagram of another device health level evaluation device based on model-driven provided by an embodiment of the present invention;

[0025] Figure 7 Schematic diagram of an electronic device provided by an embodiment of the present invention. Detailed implementation manners

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following uses specific specific examples to illustrate the implementation manners of the present disclosure. Those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all the embodiments. The present disclosure can also be implemented or applied through other different specific implementation manners. The details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. Based on the embodiments in the present disclosure, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present disclosure.

[0027] It should be noted that the following describes various aspects of the embodiments within the scope of the appended claims. It should be obvious that the aspects described in the present invention can be embodied in a wide variety of forms, and any specific structure and / or function described in the present invention is illustrative only. Based on the present disclosure, those skilled in the art should understand that one aspect described in the present invention can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects described in the present invention can be used to implement a device and / or practice a method. Additionally, this device can be implemented and this method can be practiced using other structures and / or functionality in addition to one or more of the aspects described in the present invention.

[0028] It also should be noted that the drawings provided in the following embodiments only illustrate the basic concept of the present disclosure in a schematic manner. The drawings only show the components related to the present disclosure and are not drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be an arbitrary change, and the component layout type may also be more complex. Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0029] An embodiment of the present invention provides a model-driven method and device for evaluating the health level of a device, which can adaptively adjust model parameters according to changes in data distribution, accurately capture complex dependencies between features, and improve classification accuracy.

[0030] For ease of understanding, first, a model-driven method for evaluating the health level of a device provided by an embodiment of the present invention will be described. Figure 1 The flowchart of a model-driven method for evaluating the health level of a device provided by an embodiment of the present invention is shown, as Figure 1 shown, the method includes:

[0031] Step S102, obtain the operation monitoring data of a preset device, convert the text data in the operation monitoring data into a vector format, and generate a test sample of the preset device.

[0032] To evaluate the health status of the device, an embodiment of the present invention obtains the operation monitoring data of the device and processes the data. Among them, the operation monitoring data of an embodiment of the present invention is collected from the operation logs and sensor data of a preset device. In one implementation, the data is collected from the operation logs and sensor data of large equipment, continuously collected by a variety of sensors (such as temperature sensors, pressure sensors, vibration sensors, etc.), and transmitted to the central data storage system through an encryption protocol. When storing, the data is stored in a structured format, specifically in JSON format.

[0033] In one embodiment, the collected data attributes include: temperature R a1 , pressure R a2 , vibration R a3 , current R a4 , usage duration R a5 , maintenance history R a6 , energy efficiency ratio R a7 , failure history R a8 , environmental factors R a9 , device configuration R a10 .

[0034] In this embodiment, five of the data are as follows:

[0035] {70°C, 150 psi, 0.3 mm, 15 A, 10000 h, none, 1.2, none, mild, standard type};

[0036] {85°C, 180 psi, 0.5 mm, 20 A, 15000 h, slight, 1.0, yes, cold, high-performance type};

[0037] {65℃, 140 psi, 0.2 mm, 10 A, 8000 h, none, 1.5, none, hot, standard type};

[0038] {90℃, 200 psi, 0.6 mm, 25 A, 20000 h, severe, 0.8, yes, cold, heavy type};

[0039] {75℃, 165 psi, 0.4 mm, 18 A, 12000 h, slight, 1.1, yes, mild, high - performance type}.

[0040] It should be noted that this embodiment is only to illustrate a data format and type of the present invention. In practical applications, the attributes of data are usually more than 10, and the number of data attributes may reach dozens or even hundreds. In one embodiment, if there is text - type data in the collected data, let the text be , where is the i - th word in the text, nw is the length of the text word vector. The embodiment of the present invention also converts the collected data into a vector form to adapt to the input format of the subsequent machine learning model. In one implementation, a pre - trained word embedding model E wb is used to realize the conversion between words and vectors. Each word is mapped to a vector space with a fixed dimension, that is:

[0041]

[0042] In the formula, E wb is the word embedding model, is the word in the i - th text, is the i - th converted word vector. Based on this, the text - type T wb is converted into a vector sequence { , , …, , }.

[0043] Step S104, perform feature extraction on the sample to be tested through a pre - constructed feature extraction model to obtain the target feature corresponding to the sample to be tested.

[0044] Through feature extraction, the key features that can best represent the device state can be extracted from high-dimensional data, and the information most valuable for evaluating the device health status can be retained, thereby improving the prediction accuracy and generalization ability of the model. When traditional neural network models perform feature extraction, they are easily affected by gradient vanishing, gradient explosion, and local optimal solutions, making it difficult to ensure the training stability and global search ability of the model. To solve this problem, the network parameters of the feature extraction model in the embodiments of the present invention are updated based on the allocated niches, and the niches of the network parameters are determined by simulating the competition strategies of species in an ecosystem. By defining the niches for the position and role of each neuron or parameter group in the feature space in the model and simulating the competition of species in nature, the niches with lower fitness will be replaced by the niches with higher fitness. Moreover, multiple niches can co-evolve to jointly solve complex problems. Through continuous competition and cooperation, the network parameters of the model are gradually optimized, and ultimately better performance is achieved. Dynamically adjusting the network parameters based on the above method can improve the training stability of the model, reduce the risks of gradient vanishing and explosion. It can also help the model better explore the solution space and avoid falling into local optimal solutions. Moreover, the model parameters can be adaptively adjusted according to the changes in data distribution to improve the generalization ability of the model.

[0045] Step S106, input the target feature into a preset classifier model, and the classifier model makes a classification decision on the target feature and outputs a decision result.

[0046] An appropriate classifier model can be selected according to the nature of the problem and the characteristics of the data, and the processed target feature is input into the pre-trained classifier model. The classifier model will output the health status categories of the device, such as "normal", "minor fault", "severe fault", etc.

[0047] In the embodiments of the present invention, the classifier model is based on the deep neural decision tree algorithm as the classifier algorithm. Traditional classifier models usually rely on fixed feature representation methods and lack flexibility. Moreover, when dealing with complex data structures, limited by their processing ability and expression ability, it is difficult to comprehensively capture the complex relationships in the data, resulting in insufficient classification accuracy. For example, although the decision tree model can handle non-linear relationships, it is prone to overfitting when dealing with high-dimensional data, and the depth and number of nodes of the tree are limited.

[0048] To solve the above problems, embodiments of the present invention introduce the concept of qubits to improve the processing and expressive capabilities of the model. Among them, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the states of the qubits. The states of the qubits can represent complex data relationships through characteristics such as quantum state superposition and entanglement. Further, the states of the qubits in embodiments of the present invention are determined by simulating the information gain calculation during the decision tree splitting process, and the model can automatically adjust the states of the qubits and the structure of the decision tree to optimize the classification performance. Based on this, the classifier model in embodiments of the present invention combines the strong expressive ability of the deep neural network and the interpretability of the decision tree, and can more accurately capture the complex dependence relationships between features through the states of the qubits, improve the calculation accuracy of the information gain, and improve the decision-making ability of the classifier.

[0049] Step S108, based on the decision result, evaluate the health status of the preset device.

[0050] The health status of the device can be explained according to the output result of the classifier. In one implementation, the decision label indicated by the decision result can be obtained, and the device state indicated by the decision label is determined as the health status of the preset device. Further, the result can be further analyzed by combining business knowledge and domain experience. Corresponding maintenance plans or preventive measures can also be formulated based on the health status. For example, if the model predicts that the device is about to fail, maintenance or component replacement can be carried out in advance.

[0051] Further, based on the above embodiments, embodiments of the present invention also provide another model-driven device health level evaluation method, which mainly describes the steps of determining the ecological niche based on the competition strategies of species in the simulated ecosystem and updating the network parameters of the feature extraction model based on the allocated ecological niche. In embodiments of the present invention, a 5-layer fully connected neural network is used for feature extraction. Among them, the neural network algorithm based on adaptive ecological optimization is used as the feature extraction model. Inspired by the ecological niche concept in the natural ecosystem, each species has its specific ecological niche, which specifically refers to the position occupied by a species in the ecosystem, the functions performed, and the interactions with the environment. The survival and prosperity of each species depend on its ability to effectively adapt to changes in its ecological niche. This principle is applied to the optimization of the weights and biases of the neural network, simulating the process of organisms continuously adapting to environmental changes to improve their survival rate in their ecological niche. During the training process of the neural network, each set of parameters (weights and biases) is regarded as a species in the high-dimensional parameter space, and each species tries to optimize its ecological niche in the problem space.

[0052] Figure 2The flowchart of another model-driven device health level evaluation method provided by an embodiment of the present invention is shown. Refer to Figure 2 , the method includes the following steps:

[0053] Step S202, training a preset feature extraction model through a preset training sample set.

[0054] Among them, the network parameters of the feature extraction model are assigned initial niches. First, an initial neural network parameter set is generated. Each parameter set is regarded as a species in the ecosystem, and an initial niche is randomly assigned, corresponding to the parameter initialization of the neural network. In one embodiment, the initialization method is expressed as:

[0055]

[0056]

[0057] In the formula, and respectively represent the initial weight and initial bias of the i-th species, that is, the initial values of the weight and bias corresponding to the neural network; W p,i and b p,i respectively represent the weight and bias of the i-th species; is the initialization standard deviation of the neural network, used to control the initial distribution range of the parameters; is a normal distribution with a mean of 0 and a variance of 1; represents the normal distribution; represents being subject to a specific distribution.

[0058] Furthermore, a random assignment is made to the niche of each group of parameters in the ecosystem, expressed as: L p,i ~rand(0,1). L p,i represents the niche of the i-th species, that is, the niche corresponding to the neural network parameters; rand(0,1) represents a random value between 0 and 1.

[0059] Step S204, calculating the loss function value corresponding to the network parameters of the feature extraction model, taking the loss function value as the fitness score corresponding to the network parameters, and calculating the energy regulation factor corresponding to the network parameters.

[0060] Evaluate each species, calculate the loss function value under the current neural network architecture as its fitness score, and the fitness evaluation method is expressed as:

[0061]

[0062] In the formula, S p,i is the fitness score of the i-th species; is an adaptive scoring hyperparameter used to adjust the influence of the loss function on the adaptive score S p,i ; Loss(W p,i , b p,i ) represents the network loss calculated using the weight W p,i and the bias b p,i ; Loss represents the loss function of the neural network. Preferably, is set to 5.

[0063] In one embodiment, considering the model complexity and prediction error, the calculation method of the loss function of the neural network is expressed as:

[0064]

[0065] In the formula, is the weight coefficient for adjusting the prediction error and model complexity; MSE is the mean square error of the neural network; Com is the model complexity function. Preferably, is set to 0.5.

[0066] Furthermore, the calculation method of the mean square error of the neural network is expressed as:

[0067]

[0068] In the formula, is the feature vector output by the neural network for the j-th sample, and the model prediction value obtained through the preset Softmax function; y j is the true label of the j-th sample; n MSE is the number of samples input to the neural network in the current batch.

[0069] Furthermore, the calculation method of the model complexity function is expressed as:

[0070]

[0071] In the formula, W p,ik and represent the k-th parameter in the weight parameter combination of the i-th species.

[0072] In one embodiment, the embodiment of the present invention adopts an adaptive energy adjustment mechanism to dynamically adjust the mutation energy of each species (parameter set) according to its historical performance. By adjusting the energy input during each mutation, species with lower adaptability are given a greater chance of adjustment, while species with higher adaptability maintain a relatively stable state, thereby increasing the global search ability of the algorithm and avoiding local optima. Specifically, the calculation method of the energy adjustment factor is defined as:

[0073]

[0074] Wherein, is the energy regulation factor of the i-th species; H p,i represents the historical adaptability of the i-th species, that is, the cumulative adaptability score value before this iteration; is the sensitivity parameter of energy regulation, is the average value of the cumulative adaptability scores before this iteration. Preferably, is set to 0.1.

[0075] Step S206, calculate the dynamic mutation rate corresponding to the network parameters based on the energy regulation factor.

[0076] Step S208, migrate the initial niche based on the dynamic mutation rate, thereby updating the network parameters.

[0077] Specifically, in the embodiment of the present invention, the mutation energy corresponding to the network parameters is calculated according to the energy regulation factor and the preset maximum energy input, and the adaptability score is calculated based on the mutation energy and the preset migration and mutation adjustment coefficients to determine the dynamic mutation rate corresponding to the network parameters.

[0078] The calculation of the mutation energy is expressed as:

[0079]

[0080] Wherein, E p,max is the maximum energy input, is the parameter orthogonality term. Preferably, E p,max is set to 0.4.

[0081] Furthermore, the parameter orthogonality of the environment-aware parameter orthogonality adjustment mechanism is used to ensure appropriate diversity between different parameter sets (species), thereby avoiding premature convergence to local optima and enhancing the global search ability. Specifically, by using the orthogonality adjustment of parameters during the niche adjustment process, it is ensured that the parameter update process not only responds to the current adaptability but also maintains the independence between parameters, thereby promoting the algorithm to explore unknown and potentially better solution spaces. The calculation method of the orthogonality energy is expressed as:

[0082]

[0083] Wherein, Ortho p,i represents the orthogonality energy of the i-th species; represents the vector dot product, represents the L2 norm; W p,j represents the weight and bias of the j-th species.

[0084] Further, based on the orthogonalized energy, calculate the orthogonal term of the parameters, and the calculation method is expressed as:

[0085]

[0086] In the formula, is the learning rate of orthogonal adjustment. Preferably, is set to 0.01.

[0087] The simulated species migrate and mutate the parameters according to their adaptability. The species with high adaptability have a smaller mutation rate for their parameters and remain stable; the species with low adaptability try larger mutations to seek improvement in the ecological niche. Specifically, the migration and mutation of the parameters depend on the adaptability score of the species, and the calculation method is expressed as:

[0088]

[0089]

[0090] In the formula, represents the weights of the neural network after migration and mutation, and serves as the weights of the neural network for the next iteration; is the dynamic mutation rate based on adaptability, which reflects the inverse of species adaptability. The lower the adaptability, the higher the mutation rate; represents the migration and mutation influence factor; is the migration and mutation adjustment coefficient; is the mutation energy. Preferably, is set to 0.1, is set to 0.3.

[0091] Simulate the ecological niche competition strategy, that is, competition occurs between species. The species with a low loss function (high adaptability) will occupy a better ecological niche, while the species with low adaptability may be forced to migrate to a worse ecological niche or be eliminated, thereby achieving the migration of the ecological niche, that is, corresponding to the change of the neural network weights and bias parameters, which is expressed as:

[0092]

[0093]

[0094] In the formula, and are the weights and biases of the updated neural network, ΔW p,i and Δb p,i are the update increments of the neural network weights and biases.

[0095] In one embodiment, the calculation method of the update increment of the neural network weights and biases is expressed as:

[0096]

[0097]

[0098] In the formula, is the learning rate of the neural network; is the gradient of the fitness score of the i-th species with respect to the weight parameter; is the gradient of the fitness score of the i-th species with respect to the bias parameter. Preferably, is set to 0.01.

[0099] Furthermore, the above steps are repeatedly iterated until a preset stopping iteration condition is met, which indicates that the model training is completed. In one embodiment, the preset stopping iteration condition is to reach a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0100] In summary, in the embodiments of the present invention, the weight and bias parameters are dynamically adjusted through the adaptive ecological optimization algorithm, and the parameter optimization process of the neural network is analogized to the ecological niche adaptation process of species in the ecosystem, thereby improving the stability and global search ability of the network, reducing the occurrence of gradient disappearance, gradient explosion and local optimum problems, and improving the accuracy and efficiency of feature extraction.

[0101] Furthermore, in the embodiments of the present invention, the target features are also subjected to dimensionality reduction processing through a pre-constructed dimensionality reduction model. After dimensionality reduction, the number of features is reduced, and the computing resources required for the training and inference processes of the model are significantly reduced, accelerating the training and prediction speeds of the model. The low-dimensional data after dimensionality reduction is easier to visualize, helping researchers and engineers to more intuitively understand the distribution and structure of the data. The features after dimensionality reduction can better reflect the essential structure of the data, helping the model to better generalize to unseen data and improving the robustness and accuracy of the model. Moreover, high-dimensional data is likely to cause the model to overfit, and dimensionality reduction can reduce the number of features and the complexity of the model, thereby reducing the risk of overfitting.

[0102] Traditional feature dimensionality reduction methods mainly rely on statistical methods or linear transformations, and it is difficult to effectively distinguish which features are important and which are redundant. This may lead to inaccurate feature representations after dimensionality reduction, loss of key information or introduction of irrelevant information. In addition, traditional dimensionality reduction methods are usually fixed and cannot be adjusted according to the characteristics of specific datasets, lacking flexibility and adaptability. In the embodiments of the present invention, an autoencoder neural network based on feature refinement is used as the dimensionality reduction model, and by recursively optimizing the feature weight vector, the importance of each feature can be dynamically adjusted. This method can flexibly determine the weight of each feature according to the actual distribution of the data and the relationship between features, and the autoencoder neural network can more accurately identify which features are important and which are redundant.

[0103] Among them, the autoencoder neural network based on feature refinement in the embodiments of the present invention consists of three parts: an encoder, a decoder, and a feature adjustment module. The encoder is responsible for mapping high-dimensional input data to a low-dimensional feature space, and the decoder is used to reconstruct the reduced-dimensional features back to the original space to ensure the reversibility of the dimensionality reduction process. The feature adjustment module dynamically adjusts the reduced-dimensional feature space through recursive feature adaptive optimization, enabling important features to be strengthened and secondary features to be gradually weakened, so that the reduced-dimensional feature representation has good simplicity while retaining data information. Correspondingly, Figure 3 FIG. shows a schematic diagram of the training process of the dimensionality reduction model.

[0104] Specifically, the construction method of the dimensionality reduction model is as follows:

[0105] 1) Input a preset training sample set into a preset autoencoder neural network, and map the training sample set to an initial low-dimensional feature space through the encoder of the autoencoder neural network to obtain initial low-dimensional features.

[0106] First, let the data input to the autoencoder neural network be X r , and the encoder adopts a multi-layer non-linear mapping structure to map the high-dimensional data to the initial low-dimensional feature space, which is expressed as:

[0107]

[0108] In the formula, Z r represents the initial low-dimensional feature, W r is the weight matrix of the encoder, b r is the bias vector of the encoder, and Sig enc is the multi-layer Sigmoid activation function of the encoder.

[0109] 2) Determine the feature importance of the initial low-dimensional features based on the loss function value corresponding to the initial low-dimensional features.

[0110] 3) Recursively optimize the feature weight vector of the autoencoder neural network based on the feature importance.

[0111] After the low-dimensional features are generated, the feature adjustment module automatically generates feature weights according to the feature importance in the current feature space. When the module is initialized, the same initial weight is assigned to all features for subsequent step-by-step adjustment according to the feature contribution. The calculation method of the initial weight matrix is expressed as:

[0112]

[0113] In the formula, A r is the initial weight matrix; is the feature weight vector, Each element in is initialized to the same value, indicating that all features have the same importance in the initial stage; diag is a function for extracting the diagonal elements of the extraction matrix.

[0114] Furthermore, the adjusted features can be expressed as:

[0115]

[0116] where is the feature representation after feature weight adjustment.

[0117] The feature adjustment module recursively optimizes the initially generated low-dimensional features. In each round of iteration, the module adjusts the weights of each feature according to the performance of the previous round of features, gradually enhancing the features with important influences and gradually weakening redundant or noisy features. Let the weight update rule in the t-th round of iteration be as follows:

[0118]

[0119] where represents the feature weight in the (t + 1)-th round of iteration, represents the feature weight in the t-th round of iteration, is the learning rate of the autoencoder neural network, L r is the loss function of the autoencoder neural network, Y r is the label data, represents the gradient of the loss function with respect to the feature weight. Preferably, the loss function of the autoencoder neural network is the reconstruction error loss function. Preferably, is set to 0.01.

[0120] 4) Based on the feature weight vector, determine the new low-dimensional features, and perform feature reconstruction on the new low-dimensional features through the decoder.

[0121] 5) Until the autoencoder neural network meets the preset iteration conditions, construct a dimensionality reduction model based on the autoencoder neural network.

[0122] To ensure the effectiveness of the dimensionality reduction process, the decoder remaps the low-dimensional features back to the high-dimensional space to ensure that important information is not lost during the dimensionality reduction process. The reconstruction process of the decoder is expressed as:

[0123]

[0124] where is the reconstructed high-dimensional data, is the weight matrix of the decoder, is the bias vector of the decoder, Sig dec is the multi-layer Sigmoid activation function of the decoder.

[0125] Repeat the above steps iteratively until the preset stop iteration condition is met, which indicates that the model training is completed. In one embodiment, the preset stop iteration condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0126] In summary, the embodiment of the present invention adopts an autoencoder neural network based on feature refinement, maps high-dimensional data to a low-dimensional feature space, recursively optimizes feature weights, gradually strengthens important features, weakens redundant features, ensures effective dimensionality reduction while retaining key information, effectively reduces irrelevant and redundant information, makes the feature representation after dimensionality reduction more concise, and at the same time retains the key information of the data.

[0127] Furthermore, the embodiment of the present invention also describes the construction method of the classifier model. Combining the above embodiments, in the prior art, when dealing with complex data structures, limited by its processing ability and expression ability, it is difficult to comprehensively capture the complex relationships in the data, and the classification accuracy is insufficient. The embodiment of the present invention adopts a deep neural decision tree algorithm based on quantum entanglement states as the classifier model. The Deep Neural Decision Tree (DNDT) is a model that combines the advantages of traditional decision trees and deep neural networks. It embeds the structure of a decision tree into a neural network, so that the powerful representation ability and optimization method of the neural network can be used to train the model. The intermediate layer of the deep neural decision tree consists of multiple neurons, and each neuron corresponds to a node of the decision tree. These neurons perform non-linear transformations through activation functions (such as ReLU). At each internal node (decision node), according to the feature value and the weight of the node (similar to the threshold in a decision tree), it is determined which path the data will follow. Eventually, the data will reach a leaf node, and the output of this leaf node is the final prediction result. The embodiment of the present invention combines the entanglement state of qubits to simulate the multi-dimensional representation of information, thereby enhancing the ability of the classifier to process complex data structures.

[0128] Specifically, the construction method of the classifier is as follows:

[0129] 1) Simulate the calculation of information gain during the decision tree splitting process to determine the state of qubits.

[0130] a - Initialize the nodes of the decision tree of the preset deep neural decision tree algorithm as qubits.

[0131] Initialize the qubits of the model. Each qubit represents a node in the decision tree, and the initial state is the maximum superposition state to ensure that each feature dimension can be fairly considered at the initial stage of training. Specifically, set the initial state of the qubit to , and the calculation method is expressed as:

[0132]

[0133] In the formula, n u represents the number of qubits, is the ground state of the data.

[0134] b - Train through the depth neural decision tree algorithm of the preset training sample set. Determine the rotation parameters of the custom rotation gate corresponding to the training sample set through the exponential operation in the complex domain.

[0135] c - Based on the rotation parameters and the preset quantum logic gate operations, adjust the entanglement and phase between qubits.

[0136] During the training process, use the quantum logic gate to adjust the entanglement and phase between qubits according to the training data, and simulate the information gain calculation in the decision tree splitting process. Specifically, assume the quantum logic gate is , including the Pauli - X gate, the CNOT gate, and the custom rotation gate, to simulate the calculation of information gain, which is expressed as:

[0137]

[0138] In the formula, is the custom rotation gate; is the rotation parameter dynamically adjusted according to the data, representing the rotation angle on the Bloch sphere, used to adjust the entanglement and phase relationship between qubits and enhance the non - linear decision - making ability of the model. And, is the first rotation parameter, is the second rotation parameter, is the third rotation parameter; CNOT represents the CNOT gate; Pauli - X represents the Pauli - X gate.

[0139] In one embodiment, the custom rotation gate realizes the rotation of qubits through the exponential operation in the complex domain, allowing to simulate the information gain process of the decision tree in the high - dimensional feature space. The calculation method is expressed as:

[0140]

[0141] In the formula, X us and Z us are Pauli matrices representing the rotation operations on the X and Z axes; is the imaginary unit.

[0142] Perform quantum measurement on qubits based on the entanglement and phase between the adjusted qubits, determine the probability distribution corresponding to the qubits, and evaluate the state of the qubits.

[0143] After each iteration, measure the qubits to obtain the specific states of each node. This state information is used to determine the splitting conditions in the decision tree, such as the optimization of information gain or Gini impurity, expressed as:

[0144]

[0145] In the formula, M u is the quantum measurement operation. The quantum measurement operation extracts the projection probability of the quantum state on a specific basis state to evaluate whether the current node meets a certain splitting condition; represents the probability distribution obtained after the measurement in the t-th iteration, which is used to evaluate the state of the current qubit, so as to determine the splitting path that maximizes the information gain; is the qubit in the t-th iteration; is the conjugate of the qubit in the t-th iteration.

[0146] In one embodiment, the quantum measurement operation is a measurement operation based on specific decision rules. According to the logical rules of the decision tree nodes and the state superposition and entanglement characteristics in quantum computing, it can more accurately evaluate the splitting conditions, and its matrix representation is as follows:

[0147]

[0148] In the formula, and are the basis state projection operations. The purpose of the basis state projection operation is to determine whether the node state meets the logical conditions of the decision tree node through a specific basis state. For example, a node needs to check whether the data meets a certain feature threshold, which is expressed by the weight in the direction of a specific basis vector during the projection operation, and is the first basis state projection operation, is the second basis state projection operation; I is the identity matrix, and Z is the Pauli-Z matrix; is matrix multiplication.

[0149] 2) Generate the decision branches of the decision tree based on the state of the qubits.

[0150] Among them, according to the measurement results and training data, perform node splitting to generate new decision branches. In specific implementation, it includes the following steps:

[0151] a - Calculate the decision conditions corresponding to the training sample set according to the state of the qubits.

[0152] In one embodiment, the decision condition calculated based on the state of the qubits and the training data is based on a non-linear mapping to adapt to complex data characteristics, and the calculation method is expressed as:

[0153]

[0154] where D u is the decision condition calculated based on the state of the qubits and the training data, and is used for the node splitting decision. m u is the number of qubits, w uj is the quantum state weight coefficient of the j-th node, Sig is the Sigmoid activation function; x uj is the splitting data attribute of the j-th node; is the first rotation parameter of the j-th node.

[0155] b - Based on the decision condition, control the node splitting operation of the decision tree through a preset decision function to generate the decision branches of the decision tree.

[0156] The superposition and entanglement capabilities of the quantum state enable the algorithm to more effectively perform data partitioning in high-dimensional space. Specifically, the node splitting operation is controlled by a decision function, which is executed according to the probability threshold adjusted by the rotation gate, and is expressed as:

[0157]

[0158] where is the threshold determined after optimizing the information gain or Gini impurity; Split is the splitting behavior. Preferably, is set to 0.04.

[0159] 3) Conduct model evaluation.

[0160] Among them, after the node splitting is completed, evaluate the classification effect of the current model, and adjust the quantum gate parameters according to the evaluation results to optimize the entire model, which is expressed as:

[0161]

[0162]

[0163]

[0164] where is the learning rate of the deep neural decision tree, represents the gradient of the loss function with respect to the parameter; is the first rotation parameter of the t-th iteration, is the second rotation parameter for the t-th iteration, is the third rotation parameter for the t-th iteration; is the first rotation parameter for the (t + 1)-th iteration, is the second rotation parameter for the (t + 1)-th iteration, is the third rotation parameter for the (t + 1)-th iteration. Preferably, is set to 0.2.

[0165] Repeat the above steps iteratively until the preset iteration stop condition is met, which indicates that the model training is completed. In one embodiment, the preset iteration stop condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 500 times.

[0166] In summary, the embodiments of the present invention use a deep neural decision tree algorithm based on quantum entanglement states for classification decision-making. By utilizing the superposition and entanglement characteristics of qubits to simulate the multi-dimensional representation of complex data, it can efficiently process multi-dimensional complex data. And by optimizing the decision tree node splitting conditions through quantum logic gate operations, the node splitting process is optimized, and the classification ability of the classifier under complex data structures can be improved.

[0167] Furthermore, based on the above embodiments, the embodiments of the present invention also provide another model-driven device health level evaluation method. This embodiment mainly describes the construction method of the training sample set. Figure 4 shows a flowchart of a construction method of a training sample set provided by the embodiments of the present invention, as Figure 4 shown, this method includes the following steps:

[0168] Step S10, obtain the pre-collected device operation monitoring samples.

[0169] Step S11, based on the device health status characterized by the operation monitoring samples, label the operation monitoring samples, and convert the text data in the operation monitoring samples into vector format to construct an initial sample set.

[0170] The device operation monitoring samples in the embodiments of the present invention include the operation logs and sensor data of the preset device, which can refer to the content of the above embodiments. Further, the collected data is labeled for training the model. In one embodiment, the labeling method can be manual labeling, and the labeled categories include: healthy, warning, and fault, a total of 3 categories.

[0171] Step S12, use the pre-constructed data augmentation model to augment the initial sample set to generate augmented samples of the initial sample set.

[0172] It can be understood that in the task of the present invention, the acquisition, annotation, and preprocessing of training data are time-consuming and laborious, and insufficient training samples are likely to lead to poor generalization ability of the model and affect the accuracy of the model. However, traditional generative adversarial networks often have difficulty generating accurate samples of relatively scarce categories when faced with skewed data distributions, resulting in data imbalance and affecting the generalization performance of the model. The data augmentation model of the embodiments of the present invention includes a generative adversarial network algorithm, and the generator parameters and discriminator parameters of the generative adversarial network algorithm are adjusted based on a dynamic strategy of game theory, which can achieve the balance between the generator and the discriminator during the adversarial process, making the data generated each time better simulate the real data distribution. Specifically, it includes the following steps:

[0173] 1) Adjust the feature weights of a preset training sample set through an adaptive feature transformation strategy.

[0174] In one embodiment, the quality of the generated samples is improved and the learning ability of the model for different data is enhanced through an adaptive feature transformation strategy. Based on the dynamic weighted sum and adjustment of the input features of the generator, feature weight learning is achieved, and the calculation method of the weight of the feature transformation is expressed as:

[0175]

[0176] In the formula, Sig is the Sigmoid activation function; V c is the parameter matrix of weight learning, specifically the weight parameter matrix of the generator.

[0177] 2) Through the generator of the generative adversarial network algorithm, data augmentation is performed on the training sample set with adjusted feature weights to generate initial augmented samples.

[0178] Let the generator be G c , and the discriminator be D c . Set the weight of the generator as W Gc , and the weight of the discriminator as W Dc . In one embodiment, the initialization method of the parameters of the generator and the discriminator is expressed as:

[0179]

[0180]

[0181] In the formula, is a normal distribution with a mean of 0 and a standard deviation of the identity matrix I; obeys a specific distribution.

[0182] In the generation stage, the generator generates a batch of new data samples according to the current network parameters. These samples initially attempt to mimic the distribution characteristics of real data, expressed as:

[0183]

[0184] where z c is a latent vector sampled from the prior distribution; G c is the generator function, is the generator weight at the t-th iteration; is the sample generated by the generator in the t-th iteration; W Tc represents the weight of the feature transformation; denotes element-wise multiplication.

[0185] 3) Evaluate the initial augmented samples through the discriminator of the generative adversarial network algorithm to obtain the evaluation results.

[0186] The discriminator evaluates the generated data samples and real data samples and outputs the probability that each sample is a real sample, expressed as:

[0187]

[0188] where represents combining the generated samples and real samples into a batch for input; D c is the discriminator function, is the discriminator weight at the t-th iteration; is the discrimination result of the discriminator in the t-th iteration.

[0189] 4) Calculate the loss function of the generator according to the evaluation results and the entropy of the initial augmented samples; and, calculate the loss function of the discriminator based on a preset regularization term.

[0190] The loss function of the generator in the embodiment of the present invention is comprehensively calculated in combination with the entropy of the generated data to more comprehensively evaluate the quality of the generated data and increase the diversity of the generated samples, expressed as:

[0191]

[0192] where H c represents the entropy function, is a regulation coefficient that controls the influence of the entropy term.

[0193] Furthermore, the loss function of the discriminator uses the regularization term to prevent overfitting, expressed as:

[0194]

[0195] In the formula, y c represents the true sample label (1 for true sample, 0 for generated sample); is the regularization coefficient for weight decay to reduce model complexity and overfitting risk; represents the square of the L2 norm of the discriminator weights.

[0196] 5) Based on the loss function of the generator and the discriminator, and a preset dynamic game strategy, adjust the generator parameters of the generator and the discriminator parameters of the discriminator respectively.

[0197] Use the dynamic game strategy to adjust the learning rates and update strategies of the generator and the discriminator, dynamically adjust the loss function according to the results of each round of confrontation, and give priority to improving the ability to generate scarce category data, expressed as:

[0198]

[0199]

[0200] In the formula, is the generator weight at the (t + 1)-th iteration; is the discriminator weight at the (t + 1)-th iteration; is the learning rate of the generator; is the learning rate of the discriminator; is the loss function of the generator; is the loss function of the discriminator; X rc is the sample generated by the generator; represents the gradient with respect to the generator weight; represents the weight with respect to the discriminator parameters. The goal of the generator is to maximize the loss function, so gradient ascent is used to update the parameters. The goal of the discriminator is to maximize its loss function, and gradient descent is used to update the parameters. The generator continuously adjusts its parameters to generate more realistic samples. In each iteration, the loss function of the generator changes with the quality of the samples generated by the generator. The goal of the generator is to make the loss function as large as possible, that is, to make the discriminator think that the generated samples are real. The loss function of the discriminator changes with the quality of the samples generated by the generator and the real samples. The goal of the discriminator is to maximize the loss function to correctly identify real samples and generated samples.

[0201] During the adversarial training process of the generator and the discriminator, the dynamic game strategy makes the data generated each time better simulate the real data distribution by dynamically adjusting the adjustment coefficient and regularization coefficient affected by the control entropy term, expressed as:

[0202]

[0203]

[0204] wherein, is the learning rate adjustment factor, used to control the dynamic adjustment of the learning rate; is the symbol of partial derivative; is the adjustment coefficient for controlling the influence of the entropy term in the t-th iteration, is the regularization coefficient in the t-th iteration; is the adjustment coefficient for controlling the influence of the entropy term in the t-th iteration, is the regularization coefficient in the t-th iteration. Preferably, is set to 0.01.

[0205] Repeat the above steps iteratively until the preset iteration stop condition is met, which indicates that the model training is completed. In one embodiment, the preset iteration stop condition is to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times. After the data augmentation model training is completed, the trained data augmentation model is used to increase the number of samples. In one embodiment, assuming the original collected samples are 800, and the data augmentation model generates 200 samples through augmentation, then the augmented dataset contains 1000 samples.

[0206] Step S13, construct a training sample set based on the augmented samples and the initial sample set.

[0207] In summary, the embodiment of the present invention adopts a generative adversarial network algorithm based on dynamic game. By adjusting the dynamic strategies of the generator and the discriminator, and combining the entropy function to evaluate the quality and diversity of the generated data, it can preferentially generate data of scarce categories, ensure the balance of data augmentation, improve the generalization ability and training accuracy of the model, and effectively solve the problem of insufficient training data.

[0208] Furthermore, on the basis of the above embodiments, the embodiment of the present invention further provides a model-driven device health level evaluation device, Figure 5 shows a schematic structural diagram of a model-driven device health level evaluation device provided by the embodiment of the present invention. Refer to Figure 5, the device includes: a data acquisition module 100, configured to acquire operation monitoring data of a preset device, convert text data in the operation monitoring data into a vector format, and generate a to-be-tested sample of the preset device; the operation monitoring data is collected from the operation log and sensor data of the preset device; a data processing module 200, configured to perform feature extraction on the to-be-tested sample through a pre-constructed feature extraction model to obtain a target feature corresponding to the to-be-tested sample; wherein, the network parameters of the feature extraction model are updated based on the allocated ecological niche, and the ecological niche of the network parameters is determined by simulating the competition strategy of species in an ecosystem; an execution module 300, configured to input the target feature into a preset classifier model, perform classification decision on the target feature through the classifier model, and output a decision result; wherein, the classifier model is based on a deep neural decision tree algorithm as a classification algorithm, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the states of the qubits, and the states of the qubits are determined by simulating the information gain calculation in the decision tree splitting process; an output module 400, configured to evaluate the health state of the preset device based on the decision result.

[0209] The device for evaluating the health level of a device based on model driving provided by an embodiment of the present invention has the same technical features as the method for evaluating the health level of a device based on model driving provided by the above embodiment, so it can also solve the same technical problems and achieve the same technical effects.

[0210] Further, an embodiment of the present invention also provides another device for evaluating the health level of a device based on model driving. Figure 6 The structural schematic diagram of another device for evaluating the health level of a device based on model driving provided by an embodiment of the present invention is shown, as Figure 6 shown, the above output module 400 is further configured to obtain a decision label indicated by the decision result; determine the device state indicated by the decision label as the health state of the preset device.

[0211] The above execution module 300 is further configured to initialize the nodes of the decision tree of the preset deep neural decision tree algorithm as qubits; train through the preset training sample set deep neural decision tree algorithm, and determine the rotation parameters of the custom rotation gate corresponding to the training sample set through the exponential operation in the complex domain; adjust the entanglement and phase between the qubits based on the rotation parameters and the preset quantum logic gate operation; perform quantum measurement on the qubits based on the adjusted entanglement and phase between the qubits to determine the probability distribution corresponding to the qubits, so as to evaluate the state of the qubits.

[0212] The above-mentioned execution module 300 is further configured to generate decision branches of a decision tree based on the states of qubits, including: calculating decision conditions corresponding to a training sample set according to the states of qubits; and controlling node splitting operations of the decision tree through a preset decision function based on the decision conditions to generate decision branches of the decision tree.

[0213] The above-mentioned data processing module 200 is further configured to train a preset feature extraction model through a preset training sample set; wherein, initial ecological niches are assigned to network parameters of the feature extraction model; calculating a loss function value corresponding to the network parameters of the feature extraction model, using the loss function value as an adaptability score corresponding to the network parameters, and calculating an energy adjustment factor corresponding to the network parameters; based on the energy adjustment factor, calculating a dynamic mutation rate corresponding to the network parameters; and migrating the initial ecological niche based on the dynamic mutation rate, thereby updating the network parameters.

[0214] The above-mentioned data processing module 200 is further configured to calculate mutation energy corresponding to network parameters according to the energy adjustment factor and a preset maximum energy input; calculating an adaptability score based on the mutation energy and a preset migration and mutation adjustment coefficient, and determining a dynamic mutation rate corresponding to the network parameters.

[0215] The above-mentioned data processing module 200 is further configured to perform dimensionality reduction processing on target features through a pre-constructed dimensionality reduction model; wherein, the construction method of the dimensionality reduction model includes: inputting a preset training sample set into a preset autoencoder neural network, mapping the training sample set to an initial low-dimensional feature space through an encoder of the autoencoder neural network to obtain initial low-dimensional features; determining the feature importance of the initial low-dimensional features based on the loss function value corresponding to the initial low-dimensional features; recursively optimizing the feature weight vector of the autoencoder neural network based on the feature importance; determining new low-dimensional features based on the feature weight vector, and performing feature reconstruction on the new low-dimensional features through a decoder; until the autoencoder neural network meets a preset iteration condition, constructing a dimensionality reduction model based on the autoencoder neural network.

[0216] The device further includes a construction module 500, configured to obtain pre-collected device operation monitoring samples; the device operation monitoring samples include operation logs and sensor data of a preset device; annotating the operation monitoring samples based on the device health status characterized by the operation monitoring samples, and converting text data in the operation monitoring samples into a vector format to construct an initial sample set; performing data augmentation on the initial sample set by using a pre-constructed data augmentation model to generate augmented samples of the initial sample set; wherein, the data augmentation model includes a generative adversarial network algorithm, and the generator parameters and discriminator parameters of the generative adversarial network algorithm are adjusted based on a dynamic strategy of game theory; constructing a training sample set based on the augmented samples and the initial sample set.

[0217] The above-mentioned building block 500 is further configured to adjust the feature weights of a preset training sample set through an adaptive feature transformation strategy, perform data augmentation on the training sample set with adjusted feature weights through the generator of the generative adversarial network algorithm to generate initial augmented samples; evaluate the initial augmented samples through the discriminator of the generative adversarial network algorithm to determine the initial loss value of the generator; calculate the loss function of the generator according to the initial loss value and the entropy of the initial augmented samples; and calculate the loss function of the discriminator based on a preset regularization term; adjust the generator parameters of the generator and the discriminator parameters of the discriminator respectively based on the loss function of the generator, the loss function of the discriminator, and a preset dynamic game strategy.

[0218] An embodiment of the present invention further provides an electronic device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, it implements the above Figures 1 to 4 steps of any of the methods shown. An embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is run by a processor, it executes the above Figures 1 to 4 steps of any of the methods shown. An embodiment of the present invention further provides a schematic structural diagram of an electronic device, as Figure 7 shown, which is the schematic structural diagram of the electronic device. Among them, the electronic device includes a processor 71 and a memory 70. The memory 70 stores computer-executable instructions that can be executed by the processor 71. The processor 71 executes the computer-executable instructions to implement the above Figures 1 to 4 steps of any of the methods shown. In Figure 7 the shown embodiment, the electronic device further includes a bus 72 and a communication interface 73. Among them, the processor 71, the communication interface 73, and the memory 70 are connected through the bus 72.

[0219] Among them, the memory 70 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. The communication connection between the system network element and at least one other network element is implemented through at least one communication interface 73 (which can be wired or wireless), and the Internet, wide area network, local area network, metropolitan area network, etc. can be used. The bus 72 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, etc., and can also be an AMBA (Advanced Microcontroller Bus Architecture) bus. Among them, AMBA defines three buses, including an APB (Advanced Peripheral Bus) bus, an AHB (Advanced High-performance Bus) bus, and an AXI (Advanced eXtensible Interface) bus. The bus 72 can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 7is represented only by a bidirectional arrow, but it does not mean that there is only one bus or one type of bus. The processor 71 may be an integrated circuit chip with signal processing capabilities. In the implementation process, the steps of the above method can be completed by the integrated logic circuit in the hardware of the processor 71 or instructions in the form of software. The above-mentioned processor 71 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present application can be directly embodied as being executed and completed by a hardware decoding processor, or executed and completed by a combination of hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. The storage medium is located in the memory, and the processor 71 reads the information in the memory and combines its hardware to complete the foregoing Figures 1 to 4 any of the shown methods.

[0220] A computer program product of a method and device for evaluating the health level of a device based on model driving provided by an embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the method described in the foregoing method embodiment. For specific implementation, reference can be made to the method embodiment and will not be elaborated here. Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working process of the above-described system can refer to the corresponding process in the foregoing method embodiment and will not be elaborated here. In addition, in the description of the embodiments of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations. If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program code. In the description of the present invention, it should be noted that the orientation or positional relationship indicated by the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation to the present invention. In addition, the terms "first", "second", and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.Finally, it should be noted that the above embodiments are only specific embodiments of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments, or can easily think of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A model-driven equipment health level assessment method, characterized in that: The method comprises: Acquire operation monitoring data of a preset device, convert text data in the operation monitoring data into a vector format, and generate a sample to be tested of the preset device; the operation monitoring data is collected from the operation log and sensor data of the preset device; Extracting features of the sample to be tested by a pre-built feature extraction model to obtain target features corresponding to the sample to be tested; wherein the network parameters of the feature extraction model are updated based on the assigned ecological niche, and the ecological niche of the network parameters is determined by simulating the competition strategies of species in the ecosystem; The target feature is input into a preset classifier model, and a classification decision is made on the target feature through the classifier model, and a decision result is output; wherein the classifier model is based on a deep neural decision tree algorithm as a classifier algorithm, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the state of the qubit, and the state of the qubit is determined by calculating the information gain in the process of simulating the splitting of the decision tree; Based on the decision result, evaluating the health status of the preset device; The steps of determining the ecological niche based on the competition strategies of species in the simulated ecosystem and updating the network parameters of the feature extraction model based on the assigned ecological niche include: The preset feature extraction model is trained by using a preset training sample set; wherein the network parameters of the feature extraction model are assigned with initial ecological niches; Calculating the loss function value corresponding to the network parameter of the feature extraction model, using the loss function value as the adaptability score corresponding to the network parameter, and calculating the energy adjustment factor corresponding to the network parameter; Based on the energy adjustment factor, calculating the dynamic mutation rate corresponding to the network parameter; The initial ecological niche is migrated based on the dynamic mutation rate, thereby updating the network parameters.

2. The method according to claim 1, characterized in that: Based on the decision result, the step of evaluating the health status of the preset device includes: Obtaining a decision label indicated by the decision result; The device state indicated by the decision tag is determined as the health state of the preset device.

3. The method according to claim 1, characterized in that The information gain calculation in the simulation decision tree splitting process determines the steps of the state of the quantum bit, including: Initialize the nodes of the decision tree of the preset deep neural decision tree algorithm as qubits; Training is performed using a preset deep neural decision tree algorithm for a training sample set, and determining the rotation parameters of a custom revolving door corresponding to the training sample set through exponential operations in a complex domain; Adjusting the entanglement and phase between the qubits based on the rotation parameters and a preset quantum logic gate operation; Based on the adjusted entanglement and phase between the qubits, quantum measurements are performed on the qubits to determine the probability distribution corresponding to the qubits to evaluate the states of the qubits.

4. The method according to claim 3, characterized in that: The step of generating a decision branch of the decision tree based on the state of the qubit comprises: Calculating a decision condition corresponding to the training sample set according to the state of the qubit; Based on the decision condition, the node splitting operation of the decision tree is controlled by a preset decision function to generate a decision branch of the decision tree.

5. The method according to claim 1, characterized in that The step of calculating the dynamic mutation rate corresponding to the network parameter based on the energy adjustment factor includes: Calculate the variation energy corresponding to the network parameter according to the energy adjustment factor and the preset maximum energy input; Based on the mutation energy and preset migration and mutation adjustment coefficients, the adaptability score is calculated to determine the dynamic mutation rate corresponding to the network parameter.

6. The method according to claim 1, characterized in that The method further comprises: Performing dimensionality reduction processing on the target feature through a pre-built dimensionality reduction model; The method for constructing the dimensionality reduction model comprises: Inputting a preset training sample set into a preset autoencoding neural network, mapping the training sample set to an initial low-dimensional feature space through an encoder of the autoencoding neural network to obtain an initial low-dimensional feature; Determining the feature importance of the initial low-dimensional feature based on the loss function value corresponding to the initial low-dimensional feature; recursively optimizing the feature weight vector of the autoencoder neural network based on the feature importance; Based on the feature weight vector, determine a new low-dimensional feature, and perform feature reconstruction on the new low-dimensional feature through a decoder; Until the autoencoding neural network meets the preset iteration condition, a dimensionality reduction model is constructed based on the autoencoding neural network.

7. The method according to claim 1, characterized in that The method further comprises: Acquire pre-collected equipment operation monitoring samples; the equipment operation monitoring samples include operation logs and sensor data of preset equipment; Based on the equipment health status represented by the operation monitoring sample, the operation monitoring sample is annotated, and the text data in the operation monitoring sample is converted into a vector format to construct an initial sample set; Performing data expansion on the initial sample set using a pre-built data expansion model to generate expanded samples of the initial sample set; wherein the data expansion model includes a generative adversarial network algorithm, and the generator parameters and the discriminator parameters of the generative adversarial network algorithm are adjusted based on a dynamic strategy of game theory; A training sample set is constructed based on the expanded samples and the initial sample set.

8. The method according to claim 7, characterized in that The steps of adjusting the generator parameters and discriminator parameters of the generative adversarial network algorithm based on the dynamic strategy of game theory include: The feature weights of the preset training sample set are adjusted through the adaptive feature transformation strategy; Using the generator of the generative adversarial network algorithm, data expansion is performed on the training sample set after feature weight adjustment to generate initial expanded samples; Evaluate the initial expanded sample by using the discriminator of the generative adversarial network algorithm to obtain an evaluation result; Calculating the loss function of the generator according to the evaluation result and the entropy of the initial expanded sample; and calculating the loss function of the discriminator based on a preset regularization term; Based on the loss function of the generator and the loss function of the discriminator, and a preset dynamic game strategy, the generator parameters of the generator and the discriminator parameters of the discriminator are adjusted respectively.

9. A model-driven equipment health level assessment device, characterized in that: The device comprises: A data acquisition module, used to acquire operation monitoring data of a preset device, convert text data in the operation monitoring data into a vector format, and generate a sample to be tested of the preset device; the operation monitoring data is collected from the operation log and sensor data of the preset device; A data processing module, used to extract features from the sample to be tested by using a pre-built feature extraction model to obtain target features corresponding to the sample to be tested; wherein the network parameters of the feature extraction model are updated based on the assigned ecological niche, and the ecological niche of the network parameters is determined by simulating the competition strategies of species in the ecosystem; An execution module, used for inputting the target feature into a preset classifier model, performing classification decision on the target feature through the classifier model, and outputting a decision result; wherein the classifier model is based on a deep neural decision tree algorithm as a classifier algorithm, the nodes of the decision tree of the deep neural decision tree algorithm are initialized as qubits, and the decision branches of the decision tree are determined based on the state of the qubit, and the state of the qubit is determined by calculating the information gain in the process of simulating the splitting of the decision tree; An output module, used for evaluating the health status of the preset device based on the decision result; The data processing module is also used to train a preset feature extraction model through a preset training sample set; wherein the network parameters of the feature extraction model are assigned an initial ecological niche; the loss function value corresponding to the network parameters of the feature extraction model is calculated, the loss function value is used as the fitness score corresponding to the network parameter, and the energy adjustment factor corresponding to the network parameter is calculated; based on the energy adjustment factor, the dynamic mutation rate corresponding to the network parameter is calculated; based on the dynamic mutation rate, the initial ecological niche is migrated, thereby updating the network parameters.

Images