Engine wear prediction and intelligent maintenance strategy maintenance method and device

Abstract

The present invention provides an engine wear prediction and intelligent maintenance strategy maintenance method and device, which relates to the field of artificial intelligence technology, and processes the operating state parameters of the engine through a pre-built feature extraction model, an autoencoder and a support vector machine. The parameter optimization algorithm of the feature extraction model dynamically adjusts the search strategy according to the network performance feedback, the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data, the support vector machine uses the whale optimization algorithm to determine the hyperparameters, and the quantum simulated annealing strategy is used to find the global optimal solution, which can more effectively find the optimal model parameters, and can flexibly adjust the model complexity according to the importance of the data, and also help to find the global optimal solution, which helps to achieve more accurate, efficient and reliable engine wear prediction and maintenance strategy formulation.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an engine wear prediction and intelligent maintenance strategy maintenance method and device. Background Art

[0002] With the rapid development of the automobile industry, the engine is the core component of the automobile, and its performance reliability directly affects the safety and economy of the automobile. Traditional engine maintenance methods mostly rely on regular inspections or experience-based repairs. This method is not only time-consuming, but also inefficient. It cannot accurately predict the degree of engine wear and easily leads to premature or late maintenance. In order to improve the accuracy and efficiency of engine maintenance, it is particularly important to adopt data-driven predictive maintenance methods. By collecting engine operation data in real time and using advanced data processing and machine learning algorithms, it is possible to not only accurately evaluate the engine condition, but also predict its future wear trend, thereby achieving a more intelligent and efficient engine maintenance strategy.

[0003] However, the operating state parameters of the engine are usually multi-dimensional, complex, time series, nonlinear, correlated and dependent, and there is also a large amount of data. Traditional models are usually static and cannot adapt to changes in data, which may lead to overly complex models and low computational efficiency. When processing complex and high-dimensional data, existing technologies are prone to falling into local optimal solutions, limiting their classification performance and scope of application. In addition, the prediction and generalization capabilities of existing technologies are limited, and they cannot accurately reflect the true state of the engine under different working conditions, resulting in inaccurate prediction results. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide an engine wear prediction and intelligent maintenance strategy maintenance method and device, which improves the accuracy of feature extraction, the flexibility of dimensionality reduction processing, and the classification and prediction performance.

[0005] In a first aspect, an embodiment of the present invention provides an engine wear prediction and intelligent maintenance strategy maintenance method, wherein the method includes: obtaining a preset engine operating state parameter; inputting the operating state parameter into a pre-built feature extraction model to output a target feature; wherein the neural network parameter of the feature extraction model is determined based on a preset parameter optimization algorithm; the search strategy of the parameter optimization algorithm performs performance feedback adjustment of feature extraction based on the feature extraction model; inputting the target feature into a preset autoencoder to perform dimensionality reduction processing on the target feature; wherein the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data; predicting the dimensionality reduced target feature by a pre-trained support vector machine to determine the engine wear prediction result indicated by the target feature; wherein the hyperparameters of the support vector machine are determined by a whale optimization algorithm, and the support vector machine uses a quantum simulated annealing strategy to find the global optimal solution; based on the engine wear prediction result, determining the maintenance strategy corresponding to the engine.

[0006] In combination with the first aspect, an embodiment of the present invention also provides a first implementation manner of the first aspect, wherein the step of predicting the target feature of the dimensionality reduction processing by a pre-trained support vector machine and determining the engine wear prediction result indicated by the target feature includes: predicting the target feature by the prediction function of the support vector machine, and outputting the prediction value corresponding to the target feature; comparing the predicted value with a preset wear state threshold to determine the engine wear prediction result; the engine wear prediction result includes light wear, moderate wear, and heavy wear, the wear state threshold of light wear is smaller than the wear state threshold of moderate wear, and the wear state threshold of moderate wear is smaller than the wear state threshold of heavy wear.

[0007] In combination with the first aspect, an embodiment of the present invention also provides a second implementation of the first aspect, wherein the parameter optimization algorithm includes an ant colony optimization algorithm; the search strategy of the parameter optimization algorithm is adjusted based on the performance feedback of feature extraction performed on the feature extraction model, and the step of determining the neural network parameters based on the parameter optimization algorithm includes: extracting features from a preset training sample set using a preset initial neural network parameter through a preset feature extraction model, and determining an evaluation error of the extracted features; based on the evaluation error, determining the amount of pheromone left by the preset ant optimization algorithm on the path corresponding to the initial neural network parameter; based on the pheromone amount, updating the ant pheromone concentration of the ant optimization algorithm; based on the updated ant pheromone concentration, determining an updated ant colony search strategy; and updating the initial neural network parameters based on the updated ant colony search strategy.

[0008] In combination with the first aspect, an embodiment of the present invention also provides a third implementation of the first aspect, wherein the method further includes: adjusting the node density of the network layer of the feature extraction model based on the evaluation error.

[0009] In combination with the first aspect, an embodiment of the present invention also provides a fourth implementation of the first aspect, wherein the autoencoder is constructed after training with a preset training sample set; the step of evaluating the importance of the input data to dynamically adjust the number of neurons of the autoencoder includes: mapping the preset training sample set to a compressed feature space through the encoder of the autoencoder to obtain a compressed feature vector; performing feature importance evaluation on the compressed feature vector through a preset importance evaluation function to obtain a feature importance evaluation result; wherein the importance evaluation function includes a parameter for controlling importance sensitivity, and the parameter is used to indicate the degree of influence of the feature activation value of the compressed feature vector on the importance judgment; based on the feature importance evaluation result, the number of neurons of the autoencoder is adjusted.

[0010] In combination with the first aspect, an embodiment of the present invention also provides a fifth implementation of the first aspect, wherein the step of using a quantum simulated annealing strategy to find the global optimal solution of a support vector machine includes: iteratively training a preset initial support vector machine through a preset training sample set, and calculating the interaction energy between preset quantum bits of the initial support vector machine; based on the interaction energy, updating the quantum state of the initial support vector machine through a quantum simulated annealing process; the quantum state is used to characterize the state of the solution space of the optimization problem of the initial support vector machine; based on the quantum state, determining the global optimal solution of the initial support vector machine.

[0011] In combination with the first aspect, an embodiment of the present invention also provides a sixth implementation of the first aspect, wherein a method for constructing a support vector machine includes: obtaining a hyperparameter combination of an initial support vector machine; using a whale optimization algorithm based on a spiral update strategy to simulate the encirclement and predation behavior of a whale, and adjusting the hyperparameter combination of the initial support vector machine; training the initial support vector machine with the adjusted hyperparameter combination until the initial support vector machine meets preset training requirements, and constructing a final support vector machine.

[0012] In combination with the first aspect, an embodiment of the present invention also provides a seventh implementation method of the first aspect, wherein the method also includes: obtaining operating status monitoring data of the engine under multiple operating conditions; based on the degree of engine wear indicated by the operating status monitoring data, data annotation of the operating status monitoring data is performed to construct an initial sample set; data expansion of the initial sample set is performed through a pre-trained generative adversarial network to generate expanded samples; wherein the feature weights of the generator of the generative adversarial network are adjusted using an adaptive temperature sensitivity adjustment mechanism; and a training sample set is constructed based on the initial sample set and the expanded samples.

[0013] In combination with the first aspect, an embodiment of the present invention also provides an eighth implementation of the first aspect, wherein the step of adjusting the feature weight using an adaptive temperature sensitivity adjustment mechanism includes: calculating the temperature sensitivity index corresponding to the training sample set based on the temperature characteristics of the training sample set and the average value of the temperature characteristics; adjusting the temperature characteristics of the training sample set based on the temperature sensitivity index to generate an adaptive temperature sensitivity adjustment item of the feature weight; and adjusting the feature weight of the generator based on the adaptive temperature sensitivity adjustment item.

[0014] In a second aspect, an embodiment of the present invention further provides an engine wear prediction and intelligent maintenance strategy maintenance device, the device comprising: a data acquisition module, used to obtain the operating status parameters of a preset engine; a feature extraction module, used to input the operating status parameters into a pre-built feature extraction model and output target features; wherein the neural network parameters of the feature extraction model are determined based on a preset parameter optimization algorithm; the search strategy of the parameter optimization algorithm performs performance feedback adjustment of feature extraction based on the feature extraction model; a data processing module, used to input the target features into a preset autoencoder and perform dimensionality reduction processing on the target features; wherein the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data; a prediction module, used to predict the target features of the dimensionality reduction processing through a pre-trained support vector machine, and determine the engine wear prediction result indicated by the target features; wherein the hyperparameters of the support vector machine are determined using a whale optimization algorithm, and the support vector machine uses a quantum simulated annealing strategy to find the global optimal solution; an output module, used to determine the maintenance strategy corresponding to the engine based on the engine wear prediction result.

[0015] The embodiments of the present invention bring the following beneficial effects: The embodiments of the present invention provide an engine wear prediction and intelligent maintenance strategy maintenance method and device, which, by combining advanced technologies such as ant colony optimization algorithm, autoencoder, whale optimization algorithm and quantum simulated annealing strategy, can not only more effectively find the optimal feature extraction model parameters, but also flexibly adjust the model complexity according to the importance of the data, and help find the global optimal solution. It solves the shortcomings of traditional methods in dealing with multi-dimensionality, complexity, time series characteristics, nonlinearity, correlation and dependence, and large amounts of data. These methods improve the accuracy of feature extraction, the flexibility of dimensionality reduction processing, classification and prediction performance, enhance the real-time and response speed of the system, improve the scientificity and effectiveness of the maintenance strategy, and ultimately achieve more accurate, efficient and reliable engine wear prediction and maintenance strategy formulation.

[0016] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0017] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0019] Figure 1 A flow chart of an engine wear prediction and intelligent maintenance strategy maintenance method provided by an embodiment of the present invention;

[0020] Figure 2 A flowchart of another engine wear prediction and intelligent maintenance strategy maintenance method provided by an embodiment of the present invention;

[0021] Figure 3 A flowchart of a method for constructing a training sample set provided by an embodiment of the present invention;

[0022] Figure 4 A schematic diagram of the structure of an engine wear prediction and intelligent maintenance strategy maintenance device provided by an embodiment of the present invention;

[0023] Figure 5 A schematic diagram of the structure of another engine wear prediction and intelligent maintenance strategy maintenance device provided by an embodiment of the present invention;

[0024] Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution of the present invention will be clearly and completely described in combination with the embodiments below. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] In order to solve the above technical problems, an embodiment of the present invention provides an engine wear prediction and intelligent maintenance strategy maintenance method and device, which improves the accuracy of feature extraction, the flexibility of dimensionality reduction processing, and the classification and prediction performance.

[0027] For ease of understanding, an engine wear prediction and intelligent maintenance strategy maintenance method provided by an embodiment of the present invention is first described. Figure 1 The flowchart of an engine wear prediction and intelligent maintenance strategy maintenance method provided by an embodiment of the present invention is shown. Figure 1 , the method comprises the following steps:

[0028] Step S102, obtaining preset engine operating state parameters.

[0029] The data of the present invention is collected from various sensors of the vehicle engine. The sensors monitor the operating status of the engine under different working conditions. The data is stored in a structured JSON format. In one embodiment, the attributes of the data include: engine speed , Throttle position , oil pressure , Coolant temperature , Intake pressure , fuel consumption rate , Exhaust temperature , Engine load , Engine running time , Number of engine starts It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds.

[0030] Step S104, inputting the operating state parameters into a pre-built feature extraction model, and outputting target features.

[0031] Feature extraction can extract meaningful features related to engine wear from raw sensor data, making the data easier to understand and analyze. By extracting key features, the model training process can be simplified and the model's generalization ability and prediction accuracy can be improved. Some solutions use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient exploding, or falling into local optimal solutions may occur, affecting the stability of training and the performance of the model.

[0032] The neural network parameters of the feature extraction model of the embodiment of the present invention are determined based on a preset parameter optimization algorithm. Among them, the parameter optimization algorithm can be an ant colony optimization algorithm, a particle swarm optimization algorithm, or a bee colony algorithm. These algorithms are heuristic search algorithms based on biological behaviors in nature, and can effectively search for global optimal solutions and avoid falling into local optimal solutions. Determining the neural network parameters of the feature extraction model by the above parameter optimization algorithm can improve the accuracy and comprehensiveness of feature extraction.

[0033] Furthermore, the search strategy of the parameter optimization algorithm of the embodiment of the present invention is based on the performance feedback adjustment of the feature extraction by the feature extraction model, so that the model can dynamically adjust the parameters according to the actual situation and adapt to different application scenarios and data changes. Moreover, the optimization algorithm can better jump out of the local optimum and find the global optimal solution. Taking the ant colony optimization algorithm as an example, the parameter space of the feature extraction model can be constructed into a graph model, in which the nodes represent different parameter combinations and the edges represent the path from one parameter combination to another parameter combination. The amount of pheromone on each edge reflects the quality of the path. Each ant selects a path (i.e., a set of parameters) according to the current pheromone distribution and runs the feature extraction model according to this set of parameters. Further, the performance indicators (such as accuracy, F1 score, etc.) of the feature extraction model corresponding to the path selected by each ant are calculated. The path with better performance will accumulate more pheromones, thereby attracting more ants to choose this path and update the pheromone amount on the corresponding path. After multiple iterations, the path with the highest pheromone amount is selected as the optimal solution, i.e., the optimal feature extraction model parameters.

[0034] In summary, the embodiments of the present invention adjust the search strategy through performance feedback, such as designing a pheromone mechanism for the ant colony optimization algorithm to guide the search direction of the ant colony. This method not only speeds up the search speed and improves the optimization efficiency. It can also effectively avoid falling into the local optimal solution and find the global optimal solution. Moreover, it can also enable the algorithm to better adapt to the characteristics of different data. The embodiments of the present invention can more effectively find the optimal feature extraction model parameters and improve the accuracy and comprehensiveness of feature extraction. This method not only overcomes the problem that traditional methods are prone to falling into local optimal solutions, but also improves the robustness and generalization ability of the model.

[0035] Step S106, input the target feature into a preset autoencoder to perform dimensionality reduction processing on the target feature.

[0036] Feature dimensionality reduction can eliminate the high correlation between features, avoid multicollinearity problems, and improve the stability and accuracy of the model. The data dimension after dimensionality reduction is lower and the calculation speed is faster, especially when processing large-scale data sets. The embodiment of the present invention adopts an autoencoder algorithm as a feature dimensionality reduction algorithm. The autoencoder algorithm includes two parts: an encoder and a decoder. The encoder is responsible for mapping the input data to a hidden layer, that is, a compressed representation of the features; the decoder reconstructs the original input from this compressed representation. Traditional feature dimensionality reduction methods lack an effective dynamic adjustment mechanism, which often leads to overly complex models, affecting the calculation efficiency and practicality of the model.

[0037] The number of neurons of the autoencoder of the embodiment of the present invention is dynamically adjusted by evaluating the importance of the input data. The autoencoder can automatically learn the low-dimensional representation of the data, reduce the data dimension while retaining key information. Dynamically adjusting the number of neurons can flexibly adjust the model complexity according to the importance of the data, thereby improving computational efficiency and practicality.

[0038] Step S108, predicting the target features of the dimension reduction processing by using a pre-trained support vector machine to determine the engine wear prediction result indicated by the target features.

[0039] Classifiers are used to classify the status of the engine into different categories (such as normal, light wear, severe wear, etc.) for targeted maintenance and management. Based on current and historical data, the wear that may occur in the future is predicted, and preventive measures are taken in advance. In engine wear prediction, support vector machines can be used as an effective classifier. However, when dealing with complex and high-dimensional data, traditional support vector machines may fall into local optimal solutions due to improper hyperparameter selection, which limits the classification performance and application scope of the model.

[0040] The hyperparameters of the support vector machine of the embodiment of the present invention are determined by the whale optimization algorithm, which is an efficient global optimization algorithm that can effectively search for the hyperparameters of the support vector machine. In addition, the support vector machine uses a quantum simulated annealing strategy to find the global optimal solution. The quantum simulated annealing strategy is an optimization method based on the principle of quantum computing, which escapes the local optimal solution through the quantum tunneling effect. It can still maintain a certain search capability at extremely low temperatures, which helps to find the global optimal solution. The embodiment of the present invention combines the whale optimization algorithm and the quantum simulated annealing strategy to more accurately adjust the hyperparameters, so that the model performs well on both the training set and the test set, thereby improving the generalization ability and thus improving the accuracy of the prediction results.

[0041] Step S110, determining a maintenance strategy corresponding to the engine based on the engine wear prediction result.

[0042] The target feature can be predicted by the prediction function of the support vector machine, and the prediction value corresponding to the target feature can be output. Furthermore, the prediction value can be compared with the preset wear state threshold to determine the engine wear prediction result. The engine wear prediction results include light wear, moderate wear, and severe wear. The wear state threshold of light wear is less than the wear state threshold of moderate wear, and the wear state threshold of moderate wear is less than the wear state threshold of severe wear. Among them, the threshold of the wear state can be set according to historical data and domain knowledge. For example, when the degree of wear is less than a certain threshold, it is considered to be a normal state, and when it is greater than a certain threshold, it is considered to require maintenance.

[0043] Furthermore, the corresponding maintenance strategy is determined based on the predicted wear status. Common maintenance strategies may include: Normal status: Continue monitoring and regular inspection. Mild wear: It is recommended to perform routine maintenance, such as changing the oil and filter. Moderate wear: It is recommended to perform more detailed inspection and maintenance, such as checking the piston rings, valves and other components. Severe wear: It is recommended to conduct a comprehensive overhaul or replace key components, or even consider replacing the engine.

[0044] In summary, the embodiments of the present invention solve the shortcomings of traditional methods in dealing with multi-dimensionality, complexity, time series characteristics, nonlinearity, correlation and dependence, and large amounts of data by combining advanced technologies such as ant colony optimization algorithm, autoencoder, whale optimization algorithm, and quantum simulated annealing strategy. These methods improve the accuracy of feature extraction, the flexibility of dimensionality reduction processing, classification and prediction performance, enhance the real-time and response speed of the system, improve the scientificity and effectiveness of the maintenance strategy, and ultimately achieve more accurate, efficient and reliable engine wear prediction and maintenance strategy formulation.

[0045] Furthermore, based on the above-mentioned embodiments, the embodiments of the present invention also provide another method for engine wear prediction and intelligent maintenance strategy maintenance. In this embodiment, the parameter optimization algorithm includes an ant colony optimization algorithm. The embodiments of the present invention mainly describe the steps of adjusting the search strategy of the ant colony optimization algorithm based on the performance feedback of feature extraction based on the feature extraction model, and determining the parameters of the neural network based on the ant colony optimization algorithm. Among them, the present invention adopts a 6-layer fully connected neural network, and adopts a neural network algorithm based on reconstruction interactive ant colony optimization for feature extraction. Neural network feature feedback is used in the ant colony optimization algorithm, so that ants can adjust their search paths according to the network performance of the previous round, thereby improving the flexibility and adaptability of the model. Specifically, Figure 2 A flowchart of another engine wear prediction and intelligent maintenance strategy maintenance method provided by an embodiment of the present invention is shown, referring to Figure 2 , the method comprises the following steps:

[0046] Step S202, extracting features from a preset training sample set using a preset feature extraction model and using preset initial neural network parameters, and determining an evaluation error of the extracted features.

[0047] The neural network parameters are used to extract the features of the training data set, and the extraction effect of the extracted features is evaluated to determine the effectiveness of the features, which is expressed as:

[0048]

[0049]

[0050] In the formula, y p represents the predicted value of the output layer of the neural network, is the output of the neural network for the jth sample, and the predicted probability is obtained by the preset Softmax function; is the Sigmoid activation function; is the input feature of the neural network; It is the evaluation error of the neural network, which is calculated by the predicted probability obtained by the preset Softmax function and the true sample label; is the loss function of the neural network; is the true label of the jth sample; my is the number of samples input to the neural network in the current batch; is the one-hot encoding of the true label.

[0051] In one embodiment, the loss function of the neural network adopts cross entropy loss, and the calculation method is expressed as:

[0052]

[0053] In the formula, C is the number of sample categories.

[0054] Furthermore, the predicted probability of the jth sample is The calculation method using the preset Softmax function is expressed as:

[0055]

[0056] In the formula, is the output of the last layer of the neural network, is the output of the kth layer of the neural network.

[0057] In one embodiment, the parameters of the neural network are initialized as follows:

[0058]

[0059]

[0060] In the formula, Represents the weight matrix of the kth layer of the neural network; Represents the bias vector of the kth layer of the neural network; n p and n p-1 Respectively represent the number of nodes in the k-th layer and the k-1-th layer of the neural network; is a function that generates Gaussian distributed random numbers; zeros() is a function that generates a zero matrix.

[0061] Furthermore, the ant colony in the ant colony optimization algorithm is initialized. In one embodiment, the initialization method is expressed as follows:

[0062]

[0063]

[0064] In the formula, Represents the weight path corresponding to the i-th ant in the ant colony optimization algorithm; represents the pheromone concentration of the i-th ant in the ant colony optimization algorithm; is a function that generates uniformly distributed random numbers.

[0065] Step S204, based on the evaluation error, determining the amount of pheromone left by the preset ant optimization algorithm on the path corresponding to the initial neural network parameters.

[0066] In one embodiment, It is the amount of pheromone left by the i-th ant on its path in the ant colony optimization algorithm. It is calculated based on the path quality Q and is expressed as:

[0067]

[0068] Where Q is a constant used to adjust the influence of pheromone in the ant colony optimization algorithm; is the path length of the ith ant in the ant colony optimization algorithm, specifically the loss function value of the neural network corresponding to the ith ant. Preferably, Q is set to 3.

[0069] Step S206: based on the pheromone quantity, the ant pheromone concentration of the ant optimization algorithm is updated.

[0070] Step S208, determining an updated ant colony search strategy based on the updated ant pheromone concentration.

[0071] The pheromone update rule takes into account the quality of the path, and the calculation method is expressed as:

[0072]

[0073] In the formula, is the pheromone evaporation coefficient in the ant colony optimization algorithm; preferably, Set to 0.3. is the amount of pheromone left by the i-th ant on its path.

[0074] Step S210, updating the initial neural network parameters based on the updated ant colony search strategy.

[0075] According to the performance feedback of feature extraction, the ant colony updates the search strategy and adjusts the weight search path. The ants use pheromones and the historical optimal path to update their positions in the weight space. Correspondingly, the adjustment of the weight adjustment amount is affected by the ant's pheromone update rule, and the ant's pheromone update rule is affected by the loss function value of the neural network corresponding to the ant. In traditional ant colony optimization, pheromone update is usually based on path length or global fitness. The embodiment of the present invention is aimed at real-time feedback of feature extraction performance. Among them, the search strategy of the ant colony is dynamically adjusted based on the performance feedback of the feature extraction model. After the ants complete the path exploration, the pheromone distribution is updated in real time according to the evaluation error of feature extraction, thereby effectively guiding the subsequent ants to explore in the direction of the optimal path. Combined with the above steps S204-S208, the improved pheromone update rule adopted by the present invention not only considers the path quality, but also introduces the pheromone evaporation coefficient and dynamic concentration adjustment mechanism to prevent excessive pheromones from accumulating on the suboptimal path, thereby reducing the risk of the algorithm falling into the local optimum. Based on this adjustment, it can be flexibly updated according to the model performance, thereby enhancing the adaptability of the model to complex data.

[0076] Updating the neural network parameters (i.e. weights) is expressed as:

[0077]

[0078]

[0079] In the formula, ΔW p is the weight adjustment of the neural network; Nf is the number of ants in the ant colony optimization algorithm; α p is the learning rate of the neural network; is the dynamic regularization coefficient of the neural network. Preferably, α p Set to 0.01.

[0080] Furthermore, in traditional ant colony optimization algorithms, regularization is usually a fixed value, lacking a mechanism for dynamic adjustment of feature importance, and the generalization ability of the model is poor. In order to improve generalization ability and reduce the risk of overfitting, the embodiment of the present invention also adaptively adjusts the strength of the regularization parameter according to the statistical importance of each feature. The feature importance is calculated by the partial derivative of the loss function, which helps to highlight key features and avoid overfitting of the model to irrelevant features. Specifically, the present invention adopts feature-aware dynamic regularization to adaptively adjust the regularization parameters, and dynamically adjusts the regularization strength according to the statistical importance of each feature, thereby improving the model's sensitivity to key features and overall prediction performance. Specifically, the calculation method of the dynamic regularization coefficient is expressed as:

[0081]

[0082] In the formula, is the dynamic regularization coefficient of the neural network; is the baseline value of the regularization strength of the neural network; is the hyperbolic tangent function; is the scaling factor of the neural network. Preferably, Set to 0.1, Set to 3.

[0083] Defining the feature importance index of neural network I p , the calculation method is expressed as:

[0084]

[0085] In the formula, I p Represents the feature importance index of the neural network and is the loss function of the neural network Weight W p The partial derivative of .

[0086] Furthermore, the embodiment of the present invention also optimizes the model structure by dynamically adjusting the node density of the network layer, so that the weight update of the ant path can automatically match the network complexity requirements. By adjusting the number of nodes based on performance feedback, the adaptability of the model in different data sets and feature spaces is further improved, and the generalization ability is enhanced. The network structure is slightly adjusted to dynamically adjust the node density of the network layer of the neural network. The adjustment method is expressed as:

[0087]

[0088] Where n p,new is the adjusted number of nodes of the neural network; E p,old and E p are the errors of the neural network before and after adjustment, respectively; is the adjustment step size of the neural network; is a symbolic function that returns E p,old -E p The symbol indicates whether the performance of the neural network is improved. Preferably, Set to 3, and E p,old ≥E p。

[0089] In summary, the above steps are repeated until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0090] The embodiment of the present invention adopts a fully connected neural network and a reconstruction interactive ant colony optimization algorithm. The ant colony algorithm dynamically adjusts the search strategy according to network performance feedback, thereby solving the problem of gradient disappearance, explosion or falling into local optimum in the feature extraction process.

[0091] Further, in combination with the above embodiments, the embodiments of the present invention also illustrate the training method of the autoencoder. The present invention adopts an autoencoder algorithm as a feature dimension reduction algorithm. The autoencoder algorithm includes two parts: an encoder and a decoder. Among them, the autoencoder of the present invention is constructed after training with a preset training sample set, and the training sample set can be used as the training sample set of the autoencoder after feature extraction by the above-mentioned feature extraction model. In addition, the embodiment of the present invention adopts a dynamic adjustment mechanism to dynamically adjust the number of neurons in the hidden layer according to the changes in the input data, so as to achieve adaptive compression of the feature space, so as to effectively suppress unnecessary feature dimensions and improve the quality of feature dimension reduction and the generalization ability of the model. Specifically, the training method of the autoencoder of the embodiment of the present invention is as follows:

[0092] 1) Initialize the parameters of the autoencoder algorithm, including the weight and bias parameters of the autoencoder algorithm, and adaptively set the initial number of hidden layer neurons according to the input data. The initialization method is expressed as:

[0093]

[0094]

[0095]

[0096] Where W r represents the weight parameter of the autoencoder, b r represents the bias parameter of the autoencoder; represents the initialized autoencoder weight parameters, is the standard deviation of the autoencoder parameters, N(0, I) represents a normal distribution with a mean of 0 and a covariance matrix of the identity matrix; I is the identity matrix; Represents the initialized autoencoder bias parameter, which is initialized to a zero vector; N r Represents the number of neurons in the hidden layer of the autoencoder; N input Represents the number of neurons in the input layer of the autoencoder; α r is the dynamic adjustment coefficient of the autoencoder. Preferably, Set to 0.1, α r Set to 1.5.

[0097] 2) During each forward propagation process, the data is mapped to a compressed feature space through the encoder.

[0098] For each input sample xr , the forward propagation process of the encoder and decoder is expressed as:

[0099]

[0100]

[0101] In the formula, x r is the input feature vector of the autoencoder; z r is the compressed feature vector after encoding of the autoencoder; is the reconstructed feature vector after decoding of the autoencoder; and are the weights and biases of the decoder of the autoencoder; Re is the ReLU activation function.

[0102] 3) The decoder attempts to reconstruct the input data from the compressed feature space.

[0103] The present invention transmits the compressed feature vector based on the residual connection method to obtain the reconstructed feature corresponding to the compressed feature vector; based on the reconstruction error of the reconstructed feature, the parameters of the autoencoder are updated. The residual connection method is used to ensure the effective transmission of gradients in the deep network, support the deep expansion of the network without losing the training stability, and define the calculation method of the residual module as follows:

[0104]

[0105] In the formula, h r is the output of the residual module, R(W r , x r ) represents the function of residual connection.

[0106] In one embodiment, the implementation of the residual connection is expressed as:

[0107]

[0108] Where W r,res and b r,res are the weights and biases on the residual path.

[0109] Furthermore, back propagation is used to update the weights and biases of the autoencoder. The calculation method of the partial derivatives in the back propagation process is expressed as:

[0110]

[0111]

[0112] Where Ln is the loss function of the autoencoder, R(W r , xr ) represents the function of residual connection, is the regularization parameter.

[0113] Furthermore, the calculation method of the loss function of the autoencoder is expressed as:

[0114]

[0115] For the partial derivative calculation of the weight parameters of the autoencoder, substitute the loss function and get:

[0116]

[0117] Where N is the number of samples; represents the Frobenius norm; is the regularization coefficient.

[0118] 4) Evaluate the importance of input data and dynamically adjust the number of neurons in the autoencoder.

[0119] In the specific implementation, combined with the above step 2, the preset training sample set is mapped to the compressed feature space through the encoder of the autoencoder to obtain a compressed feature vector. Further, the feature importance of the compressed feature vector is evaluated by a preset importance evaluation function to obtain a feature importance evaluation result. Among them, the importance evaluation function includes a parameter for controlling the importance sensitivity, and the parameter is used to indicate the degree of influence of the feature activation value of the compressed feature vector on the importance judgment. In one embodiment, the importance evaluation function Im(z r,i ) is calculated as:

[0120]

[0121] In the formula, is a parameter that controls the sensitivity of importance and determines the feature activation value z r,i The degree of influence on the judgment of importance.

[0122] Furthermore, based on the feature importance evaluation results, the number of neurons in the autoencoder is adjusted. In each training cycle, the number of neurons in the encoder and decoder is dynamically adjusted according to the loss function of the autoencoder and the feature importance. Specifically, the number of neurons in the hidden layer is automatically adjusted by evaluating the importance of each feature. The adjustment method is expressed as:

[0123]

[0124] In the formula, is the characteristic activation value output by the i-th neuron; N r is the number of neurons; ΔN r represents the change in the number of neurons after adjustment; is a dynamic regulatory factor; Im(z r,i ) represents the importance evaluation function of the i-th feature. Preferably, Set to 2.

[0125] Furthermore, the weight and bias parameters of the autoencoder are updated, and the update method is expressed as:

[0126]

[0127]

[0128] In the formula, is the weight parameter of the autoencoder at the t+1th iteration, is the autoencoder bias parameter for the t+1th iteration, is the weight parameter of the autoencoder at the tth iteration, is the autoencoder bias parameter of the tth iteration, is the learning rate of the autoencoder.

[0129] In one embodiment, the learning rate The update method of is set by adaptive adjustment, which is expressed as:

[0130]

[0131] In the formula, is the decay rate of the autoencoder; is the learning rate of the autoencoder at the t+1th iteration, is the learning rate of the autoencoder at the tth iteration; int(t) represents the current iteration number. Preferably, Set to 0.95.

[0132] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0133] In summary, the embodiments of the present invention adopt an autoencoder algorithm and a dynamic adjustment mechanism to adjust the number of hidden layer neurons according to changes in input data, thereby effectively compressing feature dimensions, reducing model complexity, and improving computational efficiency.

[0134] Furthermore, the embodiment of the present invention also describes a training method for a support vector machine. In combination with the above embodiment, a traditional support vector machine is prone to encounter a local optimal solution when processing complex data. The present invention uses a support vector machine algorithm based on quantum simulated annealing to classify data, and uses a quantum simulated annealing strategy to simulate the annealing process in quantum computing to find a global optimal solution and improve classification accuracy.

[0135] Specifically, the training process of the support vector machine algorithm based on quantum simulated annealing is as follows:

[0136] 1) Set the initial parameters of the support vector machine and initialize the position parameters of the quantum state and the whale algorithm. The initialization method is expressed as:

[0137] C u =rand(1,100)

[0138]

[0139]

[0140]

[0141] In the formula, C u is the penalty coefficient of the support vector machine; rand is a random function; Is the kernel function type; RBF is the radial basis function, R ucx is the input of the radial basis function; represents the initial quantum state; randn(D) is the function for randomly generating quantum states; Dg is the dimension of the feature space input to the support vector machine; Initial position parameters for the whale optimization algorithm.

[0142] 2) Use quantum simulated annealing strategy to find the global optimal solution of support vector machine.

[0143] Among them, in the process of quantum simulated annealing, the quantum state is encoded by quantum bits, and these quantum bits represent a state in the solution space of the optimization problem. Quantum simulated annealing is a technology that uses the characteristics of quantum mechanics to solve combinatorial optimization problems. The core idea is to start from an initial quantum state and gradually reduce the "temperature" of the system (that is, reduce the change of the Hamiltonian), so that the system evolves from a high-energy state to a low-energy state, and finally reaches or approaches the global optimal solution. The embodiment of the present invention iteratively trains the preset initial support vector machine through a preset training sample set, and calculates the interaction energy between the preset quantum bits of the initial support vector machine. Based on the interaction energy, the quantum state of the initial support vector machine is updated through the quantum simulated annealing process; based on the quantum state, the global optimal solution of the initial support vector machine is determined. The quantum state is used to characterize the state of the solution space of the optimization problem of the initial support vector machine.

[0144] Specifically, at the beginning of each iteration, the quantum state is updated through the quantum simulated annealing process, and the update method is expressed as:

[0145]

[0146]

[0147] In the formula, is the quantum state of the tth iteration; is the quantum state of the t+1th iteration; is the quantum annealing rate parameter; E u is the quantum state energy corresponding to the current quantum state; T u is the current simulated annealing temperature; T u,init is the initial temperature; is the annealing reduction coefficient, and Rd(t) is the current iteration number of the support vector machine.

[0148] In one embodiment, the calculation method of quantum state energy is expressed as:

[0149]

[0150]

[0151] In the formula, is the quantum state vector The i-th component of ; is the quantum state vector The jth component of ; Dw is the dimension of the quantum state vector; V ij is the interaction energy between different qubits, which is calculated by calculating the difference in Hamiltonian between different qubits; α u and βu is a parameter for adjusting the interaction strength and width. Preferably, α u and β u Set to 3 and 2 respectively.

[0152] 3) Obtain the initial support vector machine hyperparameter combination.

[0153] 4) The whale optimization algorithm is used to simulate the whale's encirclement and predation behavior based on the spiral update strategy, and the hyperparameter combination of the initial support vector machine is adjusted.

[0154] The whale optimization algorithm is used to adjust the hyperparameters of the support vector machine. The parameter configuration is optimized by simulating the whale's encirclement and predation behavior and spiral update strategy. Specifically, the method of adjusting the hyperparameters of the support vector machine using the whale optimization algorithm is expressed as follows:

[0155]

[0156]

[0157] b u =-1

[0158] l u =rand(-1,1)

[0159] In the formula, The whale position at the t+1th iteration in the whale optimization algorithm; The whale position at the tth iteration in the whale optimization algorithm; The best position so far in the whale optimization algorithm; a u Optimize the adjustment coefficient for the first whale, b u Optimize the adjustment coefficients for the second whale, constants that define the spiral shape. The third whale optimization adjustment coefficient is a number randomly drawn from the interval [-1, 1], which represents the randomness of the spiral behavior; T max The maximum number of iterations for the whale optimization algorithm.

[0160] Furthermore, bu⋅lu The term affects the spiral renewal strategy, which simulates the approximation process of whale predation behavior. The calculation method is expressed as:

[0161]

[0162] In the formula, is the Euclidean distance, also known as the L2 norm.

[0163] 5) Use the adjusted hyperparameter combination to train the initial support vector machine until the initial support vector machine meets the preset training requirements and construct the final support vector machine.

[0164] In each iteration, the effectiveness of the classification boundary is evaluated and adjusted according to the error, so the support vector machine parameter update method is expressed as:

[0165]

[0166]

[0167] In the formula, is the parameter vector of the support vector machine; is the parameter vector of the support vector machine for the tth iteration; is the parameter vector of the support vector machine for the t+1th iteration; is the learning rate of the support vector machine; J is the loss function with regular term, N is the number of samples, is the regularization parameter; is the prediction function of the support vector machine model, X u is the feature set input to the support vector machine, y u is the label set of the sample input to the support vector machine, x i is the i-th sample input to the support vector machine, y i is the label of the i-th sample input to the support vector machine; max is the maximum value function.

[0168] Furthermore, the calculation method of the prediction function of the support vector machine model is expressed as:

[0169]

[0170] In the formula, the prediction function of the support vector machine model is A probability vector will be obtained, which is the probability value corresponding to different categories, and the category corresponding to the maximum probability value is taken as the final predicted category.

[0171] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0172] In summary, the embodiment of the present invention adopts a support vector machine algorithm based on quantum simulated annealing to simulate the annealing process of quantum computing, optimize the global search capability, and improve the classification accuracy.

[0173] Furthermore, the embodiment of the present invention provides another engine wear prediction and intelligent maintenance strategy maintenance method, which describes a method for constructing a training sample set. Figure 3A flowchart of a method for constructing a training sample set provided by an embodiment of the present invention is shown. Figure 3 , including the following steps:

[0174] Step S10, acquiring engine operating status monitoring data under various operating conditions.

[0175] Step S11, based on the engine wear degree indicated by the running status monitoring data, the running status monitoring data is labeled to construct an initial sample set.

[0176] The running state monitoring data of the embodiment of the present invention refers to the above embodiment. In one implementation, the collected data is annotated by manual annotation. In one implementation, the annotated categories include: light wear, moderate wear, and heavy wear, a total of 3 categories.

[0177] Step S12, performing data expansion on the initial sample set through a pre-trained generative adversarial network to generate expanded samples.

[0178] It is understandable that in the task of the present invention, the collection, acquisition, labeling and preprocessing of training data are time-consuming and labor-intensive, and insufficient training samples can easily lead to poor generalization ability of the model, while affecting the accuracy of the model. The present invention adopts a generative adversarial network algorithm based on fusion derivative features to generate samples, thereby achieving data expansion. The generative adversarial network algorithm consists of two parts: a generator and a discriminator. The generator is responsible for generating new data similar to real data, and the task of the discriminator is to distinguish the generated data from the real data. In the process of sample generation by the generator, the present invention adopts a dynamic feature decoupling module for adaptive learning, and adjusts the weights of each feature generation according to the characteristics of different data, so that the generated data is more diversified and more adaptable. Specifically, the generator of the generative adversarial network includes multiple generation modules, each of which performs data expansion on the initial sample set based on the corresponding activation function. In each training batch, data is first randomly selected from the real data set, and then the generator generates corresponding false data based on the current parameters, expressed as:

[0179]

[0180]

[0181] In the formula, G c is the generator function, Z c is the input noise of the generator, is a normal distribution, I is the unit matrix, Fake data generated for the generator.

[0182] In one embodiment, the generator G cBy adopting a multi-module collaborative working mechanism and using different activation functions to propagate data forward in the sample generation process, the authenticity of the generated samples can be improved, which can be expressed as:

[0183]

[0184] In the formula, concat is the characteristic cascade function; N c and M c They are two different generation modules, processing the same noise input Z through different activation functions c , N c is a generation module based on Sigmoid function, M c is a generation module based on the hyperbolic tangent activation function, tanh is the hyperbolic tangent function, Sig is the Sigmoid activation function; N c The parameters of the module, M c Parameters of the module.

[0185] In one embodiment, the generator G of the generative adversarial network is initialized c and the discriminator D c The parameters of , including weights and bias parameters. The parameter initialization method is expressed as:

[0186]

[0187]

[0188] In the formula, are the parameters of the generator, are the parameters of the discriminator, is the initial parameter of the generator, is the initial parameter of the discriminator, Indicates that it obeys a specific distribution. Uniform(-0.05,0.05) means randomly generating values ​​in the interval (-0.05,0.05).

[0189] Furthermore, the feature weights of the generator in the embodiment of the present invention are adjusted using an adaptive temperature sensitivity adjustment mechanism. In the specific implementation, the following steps are included:

[0190] 1) Based on the temperature features of the training sample set and the average value of the temperature features, calculate the temperature sensitivity index corresponding to the training sample set.

[0191] In the task of the present invention, due to the exhaust temperature R a7 This property has the most critical impact on engine wear prediction. In one embodiment, an adaptive temperature sensitivity adjustment mechanism is used to improve the accuracy of generated data. For the exhaust temperature parameters of the engine, the expression of temperature-related features in the generated data is adjusted so that the generated data can better reflect the impact of temperature changes on engine wear. Specifically, let T c is the temperature feature in the input data, is the adaptive adjustment factor, then the temperature sensitivity index The calculation method is expressed as:

[0192]

[0193] In the formula, β c is the sensitivity adjustment coefficient, T c,avg is the average value of the temperature feature in the input data, T c is the temperature feature in the input data.

[0194] 2) Based on the temperature sensitivity index, the temperature characteristics of the training sample set are adjusted to generate an adaptive temperature sensitivity adjustment item of the feature weight.

[0195] Adjust the temperature characteristics of the generated data, the adjusted temperature characteristics The calculation method is expressed as:

[0196]

[0197] In the formula, is the temperature feature in the original generated data, is the adjusted temperature characteristic.

[0198] The adaptive temperature sensitivity adjustment item The collection in Represents the temperature feature set in the original generated data, the set Represents the adjusted temperature feature set.

[0199] 3) Adjust the feature weights of the generator based on the adaptive temperature sensitivity adjustment item.

[0200] Dynamically adjust the feature weight α in the generator based on the difference between the generated data and the real data in specific attributes c , the adjustment method is expressed as:

[0201]

[0202]

[0203] In the formula, Sig is the Sigmoid function, X c represents the real data, W c is the weight adjustment matrix, represents element-wise multiplication, α c is the feature weight in the generator, mean is the averaging function, It is an adaptive temperature sensitivity adjustment item.

[0204] In one embodiment, the weight matrix W c Dynamically adjust according to data characteristics to enhance the adaptability of the model. The calculation method is expressed as:

[0205]

[0206] In the formula, softmax is the Softmax function, V c and U c They are the feature importance vector obtained from prior knowledge and the feature influence vector dynamically adjusted from the current data, Represents vector V c The transpose of .

[0207] Furthermore, for the generated data and the real data, a composite loss function is calculated. The embodiment of the present invention not only considers the basic adversarial loss, but also introduces a feature matching loss term. This helps the generator to better capture the statistical characteristics of the real data and improve the quality of the generated samples. The calculation method of the loss function of the generator and the discriminator is expressed as:

[0208]

[0209]

[0210] In the formula, Express expectations, is the feature matching loss weight, Represents feature matching loss calculation; represents the loss function of the discriminator; Represents the loss function of the generator.

[0211] In one embodiment, the feature matching loss term is calculated using a similarity metric, which is expressed as:

[0212]

[0213] In the formula, Ng represents the total number of generated samples, Xc,i represents the i-th real data, The fake data generated by the i-th generator; fg Represents a deep feature extraction function, which contains a 3-layer neural network structure and is used to more accurately calculate the similarity of data features.

[0214] According to the result of the loss function, the parameters of the generator and the discriminator are updated through the back propagation algorithm. The parameter update method is expressed as:

[0215]

[0216]

[0217] In the formula, is the parameter of the generator for the tth iteration, is the parameter of the discriminator at the tth iteration, is the parameter of the generator for the t+1th iteration, is the parameter of the discriminator at the t+1th iteration, is the learning rate of the generative adversarial network, is the loss function of the generator About parameters The gradient of is the loss function of the discriminator About parameters gradient.

[0218] Repeat the above steps until the preset stop iteration condition is met, which means that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0219] Step S13: construct a training sample set based on the initial sample set and the expanded samples.

[0220] In one embodiment, the initial sample set and the expanded sample are combined to construct a training sample set. The embodiment of the present invention adopts a generative adversarial network based on fusion derived features through a dynamic feature decoupling module for adaptive learning, adjusts feature generation weights, and increases data diversity and coverage by generating high-quality synthetic data to address the scarcity and sample imbalance problems of engine monitoring data.

[0221] Based on the above embodiments, the present invention also provides an engine wear prediction and intelligent maintenance strategy maintenance device. Figure 4 The schematic diagram of the structure of the device is shown in FIG. Figure 4The device includes a data acquisition module 100, which is used to obtain the operating state parameters of a preset engine; a feature extraction module 200, which is used to input the operating state parameters into a pre-built feature extraction model and output the target features; wherein the neural network parameters of the feature extraction model are determined based on a preset parameter optimization algorithm; the search strategy of the parameter optimization algorithm is based on the performance feedback adjustment of feature extraction by the feature extraction model; a data processing module 300, which is used to input the target features into a preset autoencoder and reduce the dimension of the target features; wherein the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data; a prediction module 400, which is used to predict the target features of the reduced dimension processing through a pre-trained support vector machine, and determine the engine wear prediction result indicated by the target features; wherein the hyperparameters of the support vector machine are determined using a whale optimization algorithm, and the support vector machine uses a quantum simulated annealing strategy to find the global optimal solution; an output module 500, which is used to determine the corresponding maintenance strategy for the engine based on the engine wear prediction result.

[0222] An engine wear prediction and intelligent maintenance strategy maintenance device provided in an embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0223] Furthermore, based on the above device embodiment, the present invention also provides another engine wear prediction and intelligent maintenance strategy maintenance device. Figure 5 The schematic diagram of the structure of the device is shown in FIG. Figure 5 The above-mentioned prediction module 400 is also used to predict the target feature through the prediction function of the support vector machine, and output the prediction value corresponding to the target feature; compare the prediction value with the preset wear state threshold to determine the engine wear prediction result; the engine wear prediction result includes light wear, moderate wear, and severe wear, the wear state threshold of light wear is smaller than the wear state threshold of moderate wear, and the wear state threshold of moderate wear is smaller than the wear state threshold of severe wear.

[0224] Furthermore, the parameter optimization algorithm includes an ant colony optimization algorithm; the above-mentioned feature extraction module 200 is also used to extract features from a preset training sample set using a preset initial neural network parameter through a preset feature extraction model, and determine an evaluation error of the extracted features; based on the evaluation error, determine the amount of pheromone left by the preset ant optimization algorithm on the path corresponding to the initial neural network parameter; based on the pheromone amount, update the ant pheromone concentration of the ant optimization algorithm; based on the updated ant pheromone concentration, determine an updated ant colony search strategy; based on the updated ant colony search strategy, update the initial neural network parameters.

[0225] The feature extraction module 200 is also used to adjust the node density of the network layer of the feature extraction model based on the evaluation error.

[0226] Furthermore, the autoencoder is constructed after training with a preset training sample set; the above-mentioned data processing module 300 is also used to map the preset training sample set to a compressed feature space through the encoder of the autoencoder to obtain a compressed feature vector; the feature importance of the compressed feature vector is evaluated through a preset importance evaluation function to obtain a feature importance evaluation result; wherein the importance evaluation function includes a parameter for controlling the importance sensitivity, and the parameter is used to indicate the degree of influence of the feature activation value of the compressed feature vector on the importance judgment; based on the feature importance evaluation result, the number of neurons of the autoencoder is adjusted.

[0227] The above-mentioned prediction module 400 is also used to iteratively train a preset initial support vector machine through a preset training sample set, calculate the interaction energy between preset quantum bits of the initial support vector machine; based on the interaction energy, update the quantum state of the initial support vector machine through a quantum simulated annealing process; the quantum state is used to characterize the state of the solution space of the optimization problem of the initial support vector machine; based on the quantum state, determine the global optimal solution of the initial support vector machine.

[0228] The above-mentioned prediction module 400 is also used to obtain the hyperparameter combination of the initial support vector machine; use the whale optimization algorithm based on the spiral update strategy to simulate the surrounding predation behavior of the whale, and adjust the hyperparameter combination of the initial support vector machine; use the adjusted hyperparameter combination to train the initial support vector machine until the initial support vector machine meets the preset training requirements, and construct the final support vector machine.

[0229] The device also includes a construction module 600, which is used to obtain operating status monitoring data of the engine under various working conditions; based on the degree of engine wear indicated by the operating status monitoring data, the operating status monitoring data is data labeled to construct an initial sample set; the initial sample set is data expanded by a pre-trained generative adversarial network to generate expanded samples; wherein the feature weights of the generator of the generative adversarial network are adjusted using an adaptive temperature sensitivity adjustment mechanism; and a training sample set is constructed based on the initial sample set and the expanded samples.

[0230] The above-mentioned construction module 600 is also used to calculate the temperature sensitivity index corresponding to the training sample set based on the temperature characteristics of the training sample set and the average value of the temperature characteristics; adjust the temperature characteristics of the training sample set based on the temperature sensitivity index to generate an adaptive temperature sensitivity adjustment item of the feature weight; and adjust the feature weight of the generator based on the adaptive temperature sensitivity adjustment item.

[0231] An embodiment of the present invention further provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above Figures 1 to 3 The embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to perform the above Figures 1 to 3 Any of the steps of the method shown.

[0232] The embodiment of the present invention also provides a schematic diagram of the structure of an electronic device, such as Figure 6 FIG. 1 is a schematic diagram of the structure of the electronic device, wherein the electronic device includes a processor 61 and a memory 60, the memory 60 stores computer executable instructions that can be executed by the processor 61, and the processor 61 executes the computer executable instructions to implement the above Figures 1 to 3 Any of the methods shown.

[0233] exist Figure 6 In the illustrated embodiment, the electronic device further includes a bus 62 and a communication interface 63, wherein the processor 61, the communication interface 63 and the memory 60 are connected via the bus 62. The memory 60 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 realized through at least one communication interface 63 (which may be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. may be used. The bus 62 may be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. It may also be an AMBA (Advanced Microcontroller Bus Architecture) bus, where AMBA defines three types of buses, including an APB (Advanced Peripheral Bus) bus, an AHB (Advanced High-performance Bus) bus, and an AXI (Advanced Xtensible Interface) bus. The bus 62 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6Only one bidirectional arrow is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0234] The processor 61 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 61. The above processor 61 can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can 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 gates or transistor logic devices, discrete hardware components. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor. The steps of the method disclosed in the embodiment of the present application can be directly embodied as a hardware decoding processor to execute, or the hardware and software modules in the decoding processor can be executed. The software module can be located in a mature storage medium in the field 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 61 reads the information in the memory and combines its hardware to complete the above Figures 1 to 3 Any of the methods shown.

[0235] A computer program product of an engine wear prediction and intelligent maintenance strategy maintenance method and device provided in an embodiment of the present invention includes a computer-readable storage medium storing a program code, and the instructions included in the program code can be used to execute the method described in the previous method embodiment. The specific implementation can refer to the method embodiment, which will not be repeated here. A person skilled in the art can clearly understand that for the convenience and simplicity of description, the specific working process of the system described above can refer to the corresponding process in the above method embodiment, which will not be repeated here. For those skilled in the art, the specific meaning of the above terms in the present invention can be understood in specific circumstances. If the function is implemented in the form of a software function 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 is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), magnetic disk or optical disk and other media that can store program code. In the description of the present invention, it should be noted that the terms "first", "second" and "third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0236] Finally, it should be noted that the above embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can still modify the technical solutions recorded in the above embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; 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 be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. An engine wear prediction and intelligent maintenance strategy maintenance method, characterized in that: The method comprises: Obtaining the running state parameters of the preset engine; The operating state parameters are input into a pre-built feature extraction model to output target features; wherein the neural network parameters of the feature extraction model are determined based on a preset parameter optimization algorithm; and the search strategy of the parameter optimization algorithm performs performance feedback adjustment of feature extraction based on the feature extraction model; Input the target feature into a preset autoencoder to reduce the dimension of the target feature; wherein the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data; The target feature processed by dimensionality reduction is predicted by a pre-trained support vector machine to determine the engine wear prediction result indicated by the target feature; wherein the hyperparameters of the support vector machine are determined by a whale optimization algorithm, and the support vector machine uses a quantum simulated annealing strategy to find a global optimal solution; Determining a maintenance strategy corresponding to the engine based on the engine wear prediction result; The parameter optimization algorithm includes an ant colony optimization algorithm; the search strategy of the parameter optimization algorithm is adjusted based on the performance feedback of feature extraction performed on the feature extraction model; and the step of determining the neural network parameters based on the parameter optimization algorithm includes: Extracting features from a preset training sample set using a preset feature extraction model and using preset initial neural network parameters to determine an evaluation error of the extracted features; Based on the evaluation error, determining the amount of pheromone left by the preset ant optimization algorithm on the path corresponding to the initial neural network parameter; Based on the pheromone amount, updating the ant pheromone concentration of the ant optimization algorithm to determine an updated ant colony search strategy; Based on the updated ant colony search strategy, the initial neural network parameters are updated.

2. The method according to claim 1, characterized in that The step of predicting the target feature processed by dimension reduction by a pre-trained support vector machine to determine the engine wear prediction result indicated by the target feature comprises: Predicting the target feature by using the prediction function of the support vector machine, and outputting the prediction value corresponding to the target feature; The predicted value is compared with a preset wear state threshold to determine the engine wear prediction result; the engine wear prediction result includes light wear, moderate wear, and severe wear, the wear state threshold of light wear is smaller than the wear state threshold of moderate wear, and the wear state threshold of moderate wear is smaller than the wear state threshold of severe wear.

3. The method according to claim 1, characterized in that The method further comprises: Based on the evaluation error, the node density of the network layer of the feature extraction model is adjusted.

4. The method according to claim 1, characterized in that The autoencoder is constructed after training with a preset training sample set; the step of evaluating the importance of input data and dynamically adjusting the number of neurons of the autoencoder comprises: Mapping a preset training sample set to a compressed feature space through an encoder of the autoencoder to obtain a compressed feature vector; Performing feature importance evaluation on the compressed feature vector by using a preset importance evaluation function to obtain a feature importance evaluation result; wherein the importance evaluation function includes a parameter for controlling importance sensitivity, and the parameter is used to indicate the degree of influence of the feature activation value of the compressed feature vector on the importance judgment; Based on the feature importance evaluation result, the number of neurons of the autoencoder is adjusted.

5. The method according to claim 1, characterized in that The steps of using quantum simulated annealing strategy to find the global optimal solution of support vector machine include: Iteratively training a preset initial support vector machine using a preset training sample set, and calculating the interaction energy between preset qubits of the initial support vector machine; Based on the interaction energy, the quantum state of the initial support vector machine is updated through a quantum simulated annealing process; the quantum state is used to characterize the state of the solution space of the optimization problem of the initial support vector machine; Based on the quantum state, a global optimal solution of the initial support vector machine is determined.

6. The method according to claim 5, characterized in that The method for constructing the support vector machine comprises: Obtaining a hyperparameter combination of the initial support vector machine; The whale optimization algorithm is used to simulate the encirclement and predation behavior of whales based on the spiral update strategy, and the hyperparameter combination of the initial support vector machine is adjusted; The initial support vector machine is trained using the adjusted hyperparameter combination until the initial support vector machine meets the preset training requirements, thereby constructing a final support vector machine.

7. The method according to claim 1, characterized in that The method further comprises: Obtain engine operating status monitoring data under various operating conditions; Based on the engine wear degree indicated by the operating status monitoring data, data annotation is performed on the operating status monitoring data to construct an initial sample set; Performing data expansion on the initial sample set through a pre-trained generative adversarial network to generate expanded samples; wherein the feature weights of the generator of the generative adversarial network are adjusted using an adaptive temperature sensitivity adjustment mechanism; A training sample set is constructed based on the initial sample set and the expanded samples.

8. The method according to claim 7, characterized in that The step of adjusting the feature weights by using an adaptive temperature sensitivity adjustment mechanism includes: Calculating a temperature sensitivity index corresponding to the training sample set based on the temperature feature of the training sample set and an average value of the temperature feature; Adjusting the temperature feature of the training sample set based on the temperature sensitivity index to generate an adaptive temperature sensitivity adjustment item of the feature weight; The feature weights of the generator are adjusted based on the adaptive temperature sensitivity adjustment item.

9. An engine wear prediction and intelligent maintenance strategy maintenance device, characterized in that: The device comprises: A data acquisition module, used to obtain preset engine operating status parameters; A feature extraction module, used to input the operating state parameters into a pre-built feature extraction model and output target features; wherein the neural network parameters of the feature extraction model are determined based on a preset parameter optimization algorithm; and the search strategy of the parameter optimization algorithm performs performance feedback adjustment of feature extraction based on the feature extraction model; A data processing module, used for inputting the target feature into a preset autoencoder to reduce the dimension of the target feature; wherein the number of neurons of the autoencoder is dynamically adjusted by evaluating the importance of the input data; A prediction module, used to predict the target feature processed by dimension reduction through a pre-trained support vector machine, and determine the engine wear prediction result indicated by the target feature; wherein the hyperparameters of the support vector machine are determined by a whale optimization algorithm, and the support vector machine uses a quantum simulated annealing strategy to find a global optimal solution; An output module, used for determining a maintenance strategy corresponding to the engine based on the engine wear prediction result; The parameter optimization algorithm includes an ant colony optimization algorithm; the feature extraction module is also used to extract features from a preset training sample set using a preset feature extraction model and preset initial neural network parameters, and determine an evaluation error of the extracted features; based on the evaluation error, determine the amount of pheromone left by the preset ant optimization algorithm on the path corresponding to the initial neural network parameters; based on the pheromone amount, update the ant pheromone concentration of the ant optimization algorithm to determine an updated ant colony search strategy; based on the updated ant colony search strategy, update the initial neural network parameters.

Images