Data processing method and device for predicting replacement cycle of lubricating oil of vehicle
Through feature extraction and dimensionality reduction model combined with the gradient punishment mechanism of higher-order neural networks, the data processing speed and accuracy in vehicle lubricant replacement cycle prediction is solved, and more accurate lubricant replacement cycle prediction is achieved.
Patent Information
- Application Number
- CN202411578778.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-07
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2044-11-07
AI Technical Summary
The prior art is difficult to effectively process high-dimensional, time-serial and noisy vehicle operation data in vehicle lubricant replacement cycle prediction, resulting in improper grasp of the timing of lubricant replacement, affecting vehicle performance and increasing maintenance costs.
The feature extraction model of neural network parameters and the feature dimensionality reduction model of dynamic topological structure adjustment are updated using preset search strategies. Combined with higher-order neural networks and gradient punishment mechanisms, data dimensionality reduction and classification are carried out, and the lubricant replacement cycle prediction results are output.
Reduce data dimensions, improve data processing speed, enhance the identification ability of key features, improve classification accuracy and model stability, and enhance the accuracy and practical value of lubricant replacement cycle prediction.
Smart Images

Figure CN119067271A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of data processing, and in particular to a data processing method and device for predicting a vehicle lubricant replacement cycle. Background Art
[0002] In existing vehicle maintenance practices, lubricants are usually replaced according to the standard cycle recommended by the manufacturer, or based on the experience of the vehicle operator. This method ignores the actual impact of vehicle operating conditions on lubricant performance, which may cause the lubricant to be replaced before its optimal service life, or not be replaced in time when it should be replaced, thereby affecting vehicle performance and increasing maintenance costs. If the timing of lubricant replacement is not properly grasped, it will not only reduce vehicle performance, but may also cause problems such as accelerated component wear. With the development of data mining technology and machine learning, it has become possible to use vehicle operating data to predict the optimal time to replace lubricants. This method can more scientifically evaluate the use status and replacement needs of lubricants, and realize personalized maintenance based on actual vehicle conditions.
[0003] However, vehicle operation data, such as engine speed and vehicle speed, are basically collected through sensors, which are high-dimensional, time-series, heterogeneous, noisy and uncertain. In addition, the relationship between vehicle operating conditions and lubricant performance is highly nonlinear, and simple linear models or traditional algorithms may have difficulty capturing these complex relationships. Moreover, vehicle maintenance decisions require instant feedback, which means that data analysis algorithms need to have the ability to efficiently process large amounts of real-time data.
[0004] However, existing technologies are lacking in processing speed and timeliness. When processing large-scale vehicle operation data, it is often difficult to extract key information that affects lubricant performance through effective feature extraction methods, which reduces the prediction reliability of the model. When dealing with complex lubricant status prediction tasks, traditional classification algorithms are difficult to accurately capture the core features of highly nonlinear data, which reduces the classification effect and practical value. Summary of the invention In view of this, the purpose of the present invention is to provide a data processing method and device for predicting vehicle lubricant replacement cycle, which can reduce data dimensions, improve data processing speed, and enhance the recognition ability of key features and improve classification accuracy.
[0005] In a first aspect, an embodiment of the present invention provides a data processing method for predicting a vehicle lubricant replacement cycle, wherein the method comprises: obtaining maintenance records and vehicle monitoring parameters of a target vehicle, and assembling the maintenance records and vehicle monitoring parameters into vector data; inputting the vector data into a pre-constructed feature extraction model, and outputting key features of the vector data; wherein the feature extraction model updates neural network parameters using a preset search strategy, and dynamically adjusts key parameters in the preset search strategy by fusing historical iteration information; performing dimensionality reduction processing on key parameters using a pre-constructed feature dimensionality reduction model; wherein the number of network nodes and layer weights of the feature dimensionality reduction model are determined based on a dynamic topology structure adjustment mechanism; using a high-order neural network as a classification algorithm, classifying the key features after dimensionality reduction, and outputting classification results; wherein the classification results are used to characterize the lubricant replacement attributes of the target vehicle; the weight gradient of the high-order neural network is based on the mutual information between the input and output, and is adjusted using a high-order training mechanism of gradient penalty; based on the classification results, predicting the lubricant replacement cycle of the target vehicle.
[0006] In combination with the first aspect, an embodiment of the present invention also provides a first implementation method of the first aspect, wherein the classification results include normal replacement and early replacement; based on the classification results, the step of predicting the lubricant change cycle of the target vehicle includes: determining the classification result as the prediction result of the lubricant change cycle of the target vehicle.
[0007] In combination with the first aspect, an embodiment of the present invention also provides a second implementation of the first aspect, wherein the weight gradient of the high-order neural network is based on the mutual information between the input and output, and the step of adjusting the weight gradient using a high-order training mechanism with gradient penalty includes: obtaining the mutual information between the input and output of the high-order neural network, and using the chain rule to solve the weight gradient of the high-order neural network; updating the weight gradient based on the mutual information and a preset gradient penalty term.
[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: using a gradient-based method to update the model parameters of the high-order neural network.
[0009] In combination with the first aspect, an embodiment of the present invention also provides a fourth implementation of the first aspect, wherein the feature extraction model adopts a preset search strategy to update the neural network parameters, and the step of dynamically adjusting the key parameters in the preset search strategy by fusing historical iteration information includes: using a preset exponential decay term as a weight, calculating the parameter adjustment factor corresponding to the loss function value of the feature extraction model; based on the parameter adjustment factor and the iteration situation of the preset search strategy, adjusting the search step size of the search strategy to adjust the key parameters in the search strategy.
[0010] In combination with the first aspect, an embodiment of the present invention also provides a fifth implementation of the first aspect, wherein the search strategy includes a gray wolf optimization algorithm; wherein the key parameters of the gray wolf optimization algorithm include global search parameters and local search parameters.
[0011] In combination with the first aspect, an embodiment of the present invention also provides a sixth implementation of the first aspect, wherein the feature dimensionality reduction model includes an autoencoder; the step of determining the number of network nodes and layer weights of the feature dimensionality reduction model based on the dynamic topology structure adjustment mechanism includes: training the autoencoder through a preset training sample set, and evaluating the importance of each feature of the training sample set in the decoding process through a sensitivity scoring function; based on the importance, adjusting the weight importance of each layer of the autoencoder network; using an approximate method to calculate the gradient of the loss function of the autoencoder with respect to the number of network nodes; based on the gradient, as well as the training progress and reconstruction error of the autoencoder, adjusting the number of network nodes of the autoencoder.
[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 a pre-constructed initial sample set; using a deep learning attention mechanism to perform feature distillation processing on the initial sample set to extract key feature vectors in the initial sample set; based on the key feature vectors, performing data expansion on the initial sample set to construct a training sample set.
[0013] In combination with the first aspect, an embodiment of the present invention also provides an eighth implementation of the first aspect, wherein data expansion is performed through a generative adversarial network, and the total loss function of the generative adversarial network includes generative adversarial loss, feature matching loss and quantum state fidelity loss; the method also includes: based on preset Euler angle parameters, performing quantum gate operations on the parameters of the generative adversarial network to initialize the parameters of the generative adversarial network.
[0014] In a second aspect, an embodiment of the present invention provides a data processing device for predicting a vehicle lubricant replacement cycle, wherein the device includes: a data acquisition module, used to acquire maintenance records and vehicle monitoring parameters of a target vehicle, and assemble the maintenance records and vehicle monitoring parameters into vector data; a feature extraction module, used to input the vector data into a pre-constructed feature extraction model, and output key features of the vector data; wherein the feature extraction model uses a preset search strategy to update neural network parameters, and dynamically adjusts key parameters in the preset search strategy by fusing historical iteration information; a feature dimension reduction module, used to reduce the dimension of key parameters through a pre-constructed feature dimension reduction model; wherein the number of network nodes and layer weights of the feature dimension reduction model are determined based on a dynamic topology structure adjustment mechanism; an execution module, used to use a high-order neural network as a classification algorithm to classify the key features after dimension reduction and output a classification result; wherein the classification result is used to characterize the lubricant replacement attribute of the target vehicle; the weight gradient of the high-order neural network is based on the mutual information between the input and output, and is adjusted using a high-order training mechanism of gradient penalty; an output module, used to predict the lubricant replacement cycle of the target vehicle based on the classification result.
[0015] The embodiments of the present invention bring the following beneficial effects: the embodiments of the present invention provide a data processing method and device for predicting the vehicle lubricant replacement cycle, which can reduce data dimensions, improve data processing speed, and enhance the recognition ability of key features, improve classification accuracy, enhance model stability, improve generalization ability, and better handle unbalanced data, thereby enhancing classification effect and practical value.
[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 1A flow chart of a data processing method for predicting a vehicle lubricant replacement cycle provided by an embodiment of the present invention; Figure 2 A flowchart of another data processing method for predicting a vehicle lubricant replacement cycle provided by an embodiment of the present invention; Figure 3 A flowchart of a method for constructing a training sample set provided by an embodiment of the present invention; Figure 4 A schematic diagram of the structure of a data processing device for predicting a vehicle lubricant replacement cycle provided by an embodiment of the present invention; Figure 5 A schematic diagram of the structure of another data processing device for predicting the lubricant replacement cycle of a vehicle provided by an embodiment of the present invention; Figure 6 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION
[0020] 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.
[0021] The embodiments of the present invention provide a data processing method and device for predicting the vehicle lubricant replacement cycle, which can reduce data dimensions, improve data processing speed, and enhance the recognition ability of key features and improve classification accuracy.
[0022] For ease of understanding, a data processing method for predicting a vehicle lubricant replacement cycle provided by an embodiment of the present invention is first described, wherein: Figure 1 A flow chart of a data processing method for predicting a vehicle lubricant replacement cycle provided by an embodiment of the present invention is shown. Figure 1 As shown, the method comprises the following steps: Step S102, obtaining the maintenance record and vehicle monitoring operation parameters of the target vehicle, and assembling the maintenance record and vehicle operation monitoring parameters into vector data.
[0023] First, the embodiment of the present invention collects relevant data of the target vehicle, assembles the data into data that can be recognized by the model, and then uses the pre-built model to process the data, so as to evaluate and predict the lubricant change cycle. The following method can be used for assembly: Maintenance record vectorization Assume that each maintenance record contains the following information: the time (date) of lubricant change, the type of lubricant used, and the notes of the maintenance technician. Convert this information into numerical features, for example: Time: It can be converted into the number of days since the last lubricant change. Lubricant type: It can be represented by One-Hot Encoding. Notes of maintenance technician: Text processing can be performed to extract keywords and convert them into numerical features. In specific implementation, the data collection sources of the present invention include internal sensors and maintenance records of the vehicle, and the collected data storage format is JSON. The maintenance records include the time of each lubricant change, the type of lubricant used, the notes of the maintenance technician, etc. Vehicle operation monitoring parameters: including mileage, engine operating time, oil temperature, oil pressure, engine speed, vehicle load, etc. In one embodiment, the attributes of the data include engine speed (RPM) expressed as R a1 , vehicle speed (km / h) is represented by R a2 , oil temperature (°C) is represented by R a3 , mileage (km) is represented by R a4 , fuel consumption (L / 100km) is expressed as R a5 , engine load (%) is expressed as R a6 , Lubricating oil viscosity (cP) is expressed as R a7 , Lubricating oil acid value (mgKOH / g) is expressed as R a8 , particle count (pieces / 100ml) is expressed as R a9 , moisture content (%) is expressed as R a10 .
[0024] It should be noted that this embodiment is only used to illustrate a data format and type of the present invention. In practical applications, the attributes of data are usually more than 10 attributes, and the number of attributes of data may reach dozens or even hundreds. Furthermore, the collected data is also preprocessed, including: cleaning: removing missing values, outliers and duplicate records. Standardization: standardizing data of different units and magnitudes to make them comparable.
[0025] Step S104, inputting the vector data into a pre-built feature extraction model, and outputting key features of the vector data.
[0026] Step S106, performing dimensionality reduction processing on key parameters using a pre-built feature dimensionality reduction model.
[0027] Feature extraction is to extract the most useful information from the original data to reduce the dimension of the data and improve the performance of the model. Feature dimensionality reduction is to further reduce the number of features on the basis of feature extraction to reduce the computational complexity and improve the generalization ability of the model. Existing feature extraction techniques usually rely on fixed parameter settings, which may perform poorly when processing complex or changing data. Even if the parameter search strategy is involved to determine its network parameters, the historical iteration influence of the training process is not considered in the parameter tuning, resulting in the model's poor feature extraction ability to extract key information from a large amount of data, making the model unable to accurately reflect the actual situation of the data, affecting the reliability of the classification results. Correspondingly, the present invention introduces an adaptive feature extraction method, adopts a preset search strategy to update the neural network parameters, and dynamically adjusts the key parameters in the search strategy by fusing historical iteration information. This method combines parameter optimization strategy, historical iteration information and adaptive adjustment mechanism, which helps to improve the accuracy and reliability of the model, and can significantly improve the adaptability and performance of the model.
[0028] Furthermore, the feature dimensionality reduction processing of existing technologies (such as PCA, LDA, etc.) projects high-dimensional data into low-dimensional space through linear or nonlinear mapping. In this process, some feature information that is important to the task may be lost, resulting in failure to maintain key information while reducing the data dimension, causing information loss and affecting the execution efficiency and accuracy of the model. Moreover, many existing dimensionality reduction models use fixed network structures and parameter settings, which cannot be dynamically adjusted according to changes in data, and thus cannot adapt well to different data distributions. If the dimensionality reduction model is too simple, it may not be able to capture complex patterns in the data; if the model is too complex, it may lead to overfitting, that is, it performs well on training data but poorly on unseen data.
[0029] In this regard, the feature dimensionality reduction model proposed in the embodiment of the present invention solves the above problem by introducing a dynamic topology structure adjustment mechanism, wherein the number of network nodes and layer weights of the feature dimensionality reduction model are determined based on the dynamic topology structure adjustment mechanism. The model can dynamically adjust the number of network nodes and layer weights according to the characteristics of the input data. This adaptability enables the model to find the optimal network structure on different data sets, thereby better retaining key information.
[0030] Step S108, using a high-order neural network as a classification algorithm to classify the key features after dimensionality reduction, and outputting the classification results.
[0031] Step S110: predicting the lubricant replacement cycle of the target vehicle based on the classification result.
[0032] Among them, the embodiment of the present invention uses a trained high-order neural network as a classification algorithm to classify new dimensionality reduction features and output classification results. The classification result is used to characterize the replacement attribute of the lubricant of the target vehicle, which can be a binary classification (need to be replaced / no need to be replaced) or a multi-classification (different levels of replacement requirements). Furthermore, the current lubricant status of the target vehicle is determined based on the classification result. For example, if the classification result is "need to be replaced", it means that the current lubricant is in a bad state and needs to be replaced as soon as possible.
[0033] Traditional classifiers have the problem of gradient vanishing or gradient exploding, and the training process is unstable. In addition, in high-dimensional data, it is difficult to find the most discriminative feature combination, and the key features in the data cannot be accurately identified and utilized, resulting in poor classification results in practical applications and failure to meet the needs of practical applications. In this regard, the weight gradient of the high-order neural network of the present invention is based on the mutual information between the key features and the output, and is adjusted using a high-order training mechanism of gradient penalty. By measuring the dependency between two random variables, the key features that are most helpful for the classification task can be selected, reducing the interference of irrelevant or redundant features and improving the classification accuracy of the model. This helps the model better capture important patterns in the data. Moreover, by adding gradient-related penalty terms to the loss function, the model is encouraged to produce smoother gradients, thereby improving the stability and generalization ability of the model.
[0034] In summary, the data processing method for predicting the vehicle lubricant change cycle provided by the embodiment of the present invention can reduce data dimensions, improve data processing speed, and enhance the recognition ability of key features, improve classification accuracy, enhance model stability, improve generalization ability, and better handle unbalanced data, thereby enhancing classification effect and practical value.
[0035] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another data processing method for predicting the vehicle lubricant replacement cycle. Figure 2 A flowchart showing another data processing method for predicting vehicle lubricant replacement cycle provided by an embodiment of the present invention is shown. Figure 2 , the method comprises the following steps: Step S202: Obtain the maintenance record and vehicle monitoring operation parameters of the target vehicle, and assemble the maintenance record and vehicle operation monitoring parameters into vector data.
[0036] Step S204: input the vector data into a pre-built feature extraction model, and output key features of the vector data.
[0037] The present invention uses a 3-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, they may encounter problems such as gradient vanishing, gradient explosion, or falling into local optimal solutions, which affects the stability of training and the performance of the model. The present invention uses a neural network based on the gray wolf optimization algorithm influenced by historical iterations. On the basis of the traditional gray wolf optimization algorithm, an adaptive mechanism is used to adjust the search strategy according to the previous optimization results, thereby improving the algorithm's search efficiency and accuracy in high-dimensional data space.
[0038] The feature extraction model of the embodiment of the present invention is trained by the following steps: 1) Initializing the parameters of the neural network. The parameters of the neural network include weights and biases. In one embodiment, the initialization method is expressed as follows:
[0039] In the formula, Represents the initial value of the neural network parameters; is the initialization standard deviation; G t (0,I) represents a Gaussian distribution with a mean of 0 and a covariance of the identity matrix. Preferably, Set to 0.01.
[0040] 2) Use the preset search strategy to update the neural network parameters, and dynamically adjust the key parameters in the search strategy by integrating historical iteration information.
[0041] Among them, the search strategy includes a gray wolf optimization algorithm. The gray wolf optimization algorithm (GWO, Grey Wolf Optimizer) is a meta-heuristic optimization algorithm based on the predation behavior of gray wolves in nature. In an embodiment of the present invention, the key parameters of the gray wolf optimization algorithm include global search parameters and local search parameters. The gray wolf optimization algorithm based on historical iteration influence adjusts the parameters of the neural network by simulating the social behavior of gray wolves. Each wolf represents a set of potential solutions of the neural network, and the social behavior of gray wolves helps explore the solution space and find the optimal solution. The gray wolf position update method is expressed as:
[0042] In the formula, and are the positions of the wolves at iterations t and t+1 respectively; It is the individual best position among gray wolves; is the global optimal position; and is a parameter that balances exploration and exploitation. Preferably, and Set them to 0.3 and 0.5 respectively.
[0043] The embodiment of the present invention introduces parameters and To balance the exploration and development behaviors in the algorithm. The algorithm is allowed to flexibly switch between global search and local search according to the characteristics of the problem, thereby improving the ability to solve complex optimization problems. In particular, when set to 0.3 and 0.5 respectively, it tends to make more use of the good solution information found to perform a more detailed search. Compared with the traditional gray wolf optimization algorithm, the embodiment of the present invention not only defines fixed values as the weights of exploration and development, but also proposes a method to dynamically adjust one of these weights (i.e. ), which can enhance the adaptability of the algorithm to different optimization scenarios. In the specific implementation, the preset exponential decay term is used as the weight to calculate the parameter adjustment factor corresponding to the loss function value of the feature extraction model; based on the parameter adjustment factor and the iteration of the preset search strategy, the search step size of the search strategy is adjusted to adjust the key parameters in the search strategy.
[0044] Specifically, in each algorithm iteration, the search strategy is adjusted according to the optimization results of the previous generation. The adaptive adjustment method is expressed as:
[0045] In the formula, It is updated ; is the adjustment factor used to adjust size; Represents the Euclidean distance, also known as the L2 norm.
[0046] In one embodiment, the adjustment factor Adjustments are made based on feedback from historical optimization performance, and the calculation method is expressed as:
[0047] Where, T ve is the number of historical iterations considered, indicating how many generations of past performance affect the current adjustment; is the attenuation factor, is the decay rate; is the loss function value at the tth iteration; is the neural network parameter of the t-th iteration, that is, the individual best position x of the gray wolf in the t-th iteration cp The corresponding neural network parameters. Preferably, Set to 3.
[0048] In summary, the embodiment of the present invention uses the best position x of the previous generations cp and the global optimal position x cg , and the current individual's position To calculate the new position, this method effectively integrates past successful experience and current state information. In addition, the adjustment factor The calculation takes into account the changes in the loss function value within the historical number of iterations, further improving the accuracy of selecting the search direction.
[0049] Moreover, the embodiment of the present invention considers the influence of individual optimal positions and global optimal positions at the same time and assigns different weights. It not only retains the advantages of the traditional gray wolf optimization algorithm and can quickly locate the optimal area, but also strengthens the directionality of specific target points, so that the algorithm can be fine-tuned more accurately when approaching the optimal solution. The exponential decay term is used when calculating the adjustment factor. , which means that as time goes by (that is, the number of iterations increases), the importance of early iteration results gradually decreases, which helps the algorithm respond faster to the latest good solutions and avoids stagnation in search due to over-reliance on information from earlier stages. Update the weights and biases of the neural network, and the update method is expressed as:
[0050] In the formula, and are the neural network parameters of the t-th iteration and the t+1-th iteration respectively; is the learning rate of the neural network; It is in the parameter The gradient of the loss function at . Preferably, Set to 0.01.
[0051] Furthermore, the gradient The calculation method is expressed as:
[0052] In the formula, is the number of samples in each batch of training data; It is The input data of samples; It is The true labels of samples; is the cross entropy loss function; is the predicted output of the neural network model.
[0053] 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.
[0054] The embodiment of the present invention adopts a three-layer fully connected neural network for feature extraction, and achieves a better solution by improving the performance of the Grey Wolf Optimization Algorithm in handling complex, high-dimensional or multimodal optimization problems, more intelligently controlling the exploration and development ratio in the search process, and using historical information to guide future decisions. This enables the model to extract key information from the data, overcoming the poor performance of previous models due to insufficient feature extraction.
[0055] Step S206, performing dimensionality reduction processing on key parameters using a pre-built feature dimensionality reduction model.
[0056] The present invention adopts an autoencoder algorithm for feature dimensionality reduction, and the autoencoder algorithm includes two parts: an encoder and a decoder. The encoder is responsible for converting high-dimensional input data into low-dimensional feature representations, and the decoder attempts to reconstruct the original input from these low-dimensional features to ensure that as little information as possible is lost during the dimensionality reduction process. The present invention adopts a dynamic topology adjustment mechanism, which enables the network to automatically adjust its internal structure according to the complexity of the input data, such as increasing or decreasing the number of neural network layers or the number of neurons, optimizing the sparse representation ability of the network, and improving the efficiency and effect of feature dimensionality reduction.
[0057] In the specific implementation, the training steps are as follows: 1) Initialize the network parameters of the autoencoder. In one embodiment, the network parameters are initialized using the He initialization method, and the initialization method is expressed as:
[0058] In the formula, represents the weight of the lth layer of the autoencoder; Indicates the number of nodes in the l-1th layer; Represents the standard normal distribution The random value to draw.
[0059] 2) Determine the number of network nodes and layer weights of the feature dimensionality reduction model based on the dynamic topology structure adjustment mechanism.
[0060] a-Train the autoencoder using a preset training sample set.
[0061] In the forward propagation of data, the data is compressed into low-dimensional features by the encoder and then reconstructed by the decoder. The output of the encoder is and decoder reconstruction The calculation method is:
[0062]
[0063] In the formula, ReLU activation function. , Represent the weights of the encoder and decoder respectively; , Represent the bias of the encoder and decoder respectively; Represents input data.
[0064] b- The importance of each feature of the training sample set in the decoding process is evaluated through the sensitivity scoring function.
[0065] c-Based on the importance, the weight importance of each layer of the autoencoder is adjusted.
[0066] In one embodiment, an adaptive feature feedback adjustment mechanism is used to calculate the adaptive feature feedback adjustment increment of the encoder, which is used to adjust the weight of the feature representation in real time, and dynamically adjust the weight importance of each layer of the encoder according to the error output by the decoder. The calculation method is expressed as:
[0067] In the formula, represents the learning rate of the adaptive feature feedback regulation mechanism; represents the loss function of the decoder; represents the partial derivative of the loss function with respect to the encoder output; is the sensitivity scoring function; represents element-wise multiplication. Preferably, Set to 0.01.
[0068] Moreover, in this embodiment, the sensitivity scoring function is used to evaluate the importance of each feature in the decoding process, and the calculation method is expressed as:
[0069] Use the gradient descent method to update the network parameters to minimize the loss function. The update method is expressed as:
[0070] In the formula, represents the learning rate; Represents the partial derivative of the loss function with respect to the weight; Represents the encoder's adaptive feature feedback adjustment increment.
[0071] d-Use an approximate method to calculate the gradient of the loss function of the autoencoder with respect to the number of network nodes.
[0072] In the early stage of training, the network hierarchy is automatically adjusted according to the network training progress and reconstruction error. The calculation method of the adjustment function is expressed as:
[0073] In the formula, The adjustment function is used to adjust the number of network nodes. If its value exceeds the threshold, the number of neurons in the middle layer is increased; if its value does not exceed the threshold, the number of neurons in the middle layer is reduced. Indicates the number of nodes in the lth layer; represents the step length of structural adjustment; Represents the current loss function; represents the gradient of the loss function with respect to the number of nodes in the lth layer. Furthermore, considering that the number of nodes directly affects the output dimension of the network layer, an approximate method is used to calculate the loss function About the number of nodes The gradient of is calculated as:
[0074] In the formula, Represents the standard normal distribution The random value to draw.
[0075] e- Based on the gradient, as well as the training progress and reconstruction error of the autoencoder, adjust the number of network nodes of the autoencoder.
[0076] Specifically, the loss function of the autoencoder is calculated. The present invention adopts the mean square error loss based on L1 regularization as the loss function of the autoencoder, and the calculation method is expressed as:
[0077] In the formula, is the loss function of the autoencoder; represents the reconstruction error; represents the regularization coefficient; represents the L1 regularization term of the weight. Preferably, Set to 0.3.
[0078] Furthermore, the calculation method of the reconstruction error is expressed as:
[0079] Where m is the number of samples input to the autoencoder in the current batch, and are the kth elements of the actual input and reconstructed output respectively.
[0080] Moreover, the L1 regularization term is used to increase the sparsity of the weights, and the calculation method is expressed as:
[0081] Here, n r is the dimension of the weight vector, is the i-th element of the weight vector.
[0082] Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is to reach a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times. In summary, the autoencoder algorithm of the embodiment of the present invention is used to perform feature dimensionality reduction, and the encoder converts high-dimensional data into low-dimensional data while maintaining important information of the data, thereby optimizing the processing speed and model efficiency.
[0083] Step S208, using a high-order neural network as a classification algorithm to classify the key features after dimensionality reduction, and output the classification results.
[0084] The present invention adopts a high-order neural network as the classification algorithm and a high-order training mechanism with gradient penalty to adjust the weight gradient of each layer during the training process, thereby ensuring that the network pays more attention to the features that have the greatest impact on the final classification decision during the learning process, thereby improving the classification accuracy.
[0085] Specifically, the training process of the high-order neural network classification algorithm is as follows: 1) Initialize the parameters of the high-order neural network. In one embodiment, the weights of the high-order neural network and bias The initialization method uses He initialization, which is expressed as:
[0086]
[0087] In the formula, represents the weight of the lth layer of the high-order neural network; Represents the bias of the lth layer of the high-order neural network; Represents the number of nodes in the l-1th layer of the high-order neural network; Represents a random value drawn from a normal distribution with mean 0 and standard deviation 1.
[0088] 2) Based on the mutual information between input and output, a high-order training mechanism with gradient penalty is used to adjust the weight gradient of the high-order neural network.
[0089] Specifically, the mutual information between the input and output of the high-order neural network is obtained, and the weight gradient of the high-order neural network is solved using the chain rule; the weight gradient is updated based on the mutual information and a preset gradient penalty term.
[0090] In the process of forward propagation, the input reduced-dimensional data passes through a multi-layer neural network, and each layer uses the ReLU activation function to transform the linearly weighted input. The calculation of each layer output is expressed as:
[0091] In the formula, represents the output of the lth layer; represents the output of the l-1th layer. For the first layer, is the input feature; Represents the ReLU activation function.
[0092] The loss function is used to calculate the difference between the model output and the true label. The calculation method is expressed as:
[0093] In the formula, Indicates the total number of categories; Represents the i-th element in the one-hot encoding of the true label; is the probability of the i-th element after the Softmax function of the model output, where the purpose of the Softmax function is to convert the output into a probability distribution.
[0094] Calculate the gradient of the loss function with respect to each parameter, and adjust the gradients of those parameters that are sensitive to loss changes through the gradient penalty mechanism, so that the model training pays more attention to important features. The calculation method of the weight update increment is expressed as:
[0095] In the formula, is the input feature and output The mutual information between Represents the learning rate of the high-order neural network; Represents the gradient penalty coefficient; is the sign function of the weight. Preferably, Set to 0.01, Set to 3.
[0096] Furthermore, the gradient The calculation of is solved by the chain rule, and the calculation method is expressed as:
[0097] In the formula, is the partial derivative of the loss function with respect to the output of the last layer of the network, is the partial derivative of the last layer output with respect to the weight.
[0098] Furthermore, an adaptive feature association feedback mechanism is used to dynamically adjust the connection weights between the layers in the neural network, and the adjustment is made according to the mutual information between the features, so as to improve the model's ability to identify the complex relationship between the factors affecting the lubricant replacement cycle. and output The calculation method of mutual information between is expressed as:
[0099] In the formula, is the input feature and output The mutual information between is the joint probability density function, and is the marginal probability density function.
[0100] Furthermore, the embodiment of the present invention adopts a gradient-based method to update the model parameters of the high-order neural network to improve the optimization efficiency and convergence speed. The update method is expressed as:
[0101]
[0102] In the formula, is the gradient of the bias to the loss. 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 to reach a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times. In summary, the embodiment of the present invention uses a high-order neural network and a gradient penalty mechanism to train the classifier model, which strengthens the model's learning of key features and improves the accuracy and reliability of classification.
[0103] Step S210: determining the classification result as the prediction result of the lubricant replacement cycle of the target vehicle.
[0104] The classification results include normal replacement and early replacement; "normal replacement" refers to the standard replacement cycle based on the manufacturer's recommendations and operating conditions, and "early replacement" refers to the need to replace the lubricant in advance due to harsh operating conditions (such as high load, high temperature).
[0105] Furthermore, based on the above embodiment, the embodiment of the present invention further designs a training sample set. Correspondingly, the embodiment of the present invention also provides another data processing method for predicting the vehicle lubricant replacement cycle. This embodiment mainly describes the method for constructing the training sample set. Figure 3 , the method comprises the following steps: Step S10, obtaining a pre-built initial sample set.
[0106] Among them, the initial sample set refers to the maintenance records and vehicle monitoring parameters of the above-mentioned embodiment, which will not be repeated here. In the process of constructing the training sample set, the embodiment of the present invention also prepares a label for each sample to mark the collected data, and the label indicates whether the lubricant needs to be replaced. In one embodiment, the labeled categories include: "normal replacement" and "early replacement". For example, "need to be replaced" can be marked as 1, and "early replacement" can be marked as 0. Among them, the labels can be generated by manual labeling. It should be noted that "normal replacement" refers to the standard replacement cycle recommended by the manufacturer and under operating conditions, and "early replacement" refers to the need to replace the lubricant in advance due to harsh operating conditions (such as high load, high temperature).
[0107] Step S11, using the attention mechanism of deep learning to perform feature distillation processing on the initial sample set, and extracting the key feature vectors in the initial sample set.
[0108] 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 model generalization ability and affect the accuracy of the model. The embodiment of the present invention uses a data expansion network algorithm based on adaptive feature distillation to generate samples. Before each training cycle, key feature vectors are extracted from the real data set through a feature distillation unit. Furthermore, these feature vectors are used as additional guidance information to enable the data expansion algorithm to generate expanded samples, which assists the data expansion algorithm to more accurately capture and simulate the real data distribution, expressed as:
[0109]
[0110] In the formula, is the distilled feature vector; It is a real data set; The false data generated by the generator; represents the feature distillation function; Represents a function that generates fake data.
[0111] In one embodiment, the feature distillation process uses the attention mechanism in deep learning to extract important features. The calculation method is expressed as:
[0112] Where, Q c , K c and V c From the input data The query matrix, key matrix, and value matrix learned in is the Softmax function, d kis the dimension of the key vector. Preferably, the dimension of the key vector is d k Set to 64.
[0113] Furthermore, the query, key and value matrices are obtained through linear transformation, and the calculation method is expressed as:
[0114]
[0115]
[0116] In the formula, is the query weight matrix, is the key weight matrix, is the value weight matrix. Moreover, the query weight matrix, the key weight matrix and the value weight matrix are determined by learning. In this embodiment, the learning method is to perform back propagation calculation by gradient descent.
[0117] Step S12: Based on the key feature vectors, the initial sample set is expanded to construct a training sample set.
[0118] In one embodiment, the present invention performs data augmentation by generating adversarial networks. The generative adversarial network algorithm based on adaptive feature distillation includes two parts: a generator ( ) and the discriminator ( ), the generator is responsible for generating new data instances, and the discriminator is responsible for evaluating the authenticity of these data instances. In order to enhance the data generation capability of the model and optimize the diversity and authenticity of the generated data, the present invention adopts a feature distillation unit to learn and distill key features from the real data set, guide the generator to more accurately simulate the data distribution, and then enhance the exploration capability of the generated samples to achieve a wider coverage of the data space. In the specific implementation, the embodiment of the present invention performs quantum gate operations on the parameters of the generative adversarial network based on the preset Euler angle parameters to perform parameterization on the generative adversarial network (generator G c and the discriminator D c In one embodiment, the parameters of the generator and the parameters of the discriminator The initialization method uses the initialization function of quantum gate conversion, which allows more flexible adjustment space for parameters during initialization. Taking the generator as an example, when the initialization function of quantum gate conversion initializes the parameters of the generator, the parameter initialization method is expressed as:
[0119] In the formula, Represented by Euler angles , and Parameterized quantum gate operations, is the first Euler angle parameter, is the second Euler angle parameter, is the third Euler angle parameter.
[0120] Furthermore, the Euler angle parameter intuitively describes the rotation of the object in three-dimensional space and determines the initialization direction of the qubit. When initializing the parameters of the generated network, the randomness of the parameter initialization is increased, and a variety of angle parameters are introduced to make the parameters have more flexible adjustment directions. The calculation method is expressed as:
[0121]
[0122]
[0123] Where rand(0, 1) is a function that generates uniformly distributed random numbers between 0 and 1.
[0124] Furthermore, the generator uses the current quantum state and the distilled feature vector to generate new data instances, and the discriminator evaluates the similarity between the generated data and the real data, and feeds back to the generator to adjust the parameters. The rule for alternating training of the generator and the discriminator to update the parameters is expressed as:
[0125]
[0126] In the formula, and are the learning rates of the generator and the discriminator respectively; is the total loss function of the generative adversarial network; and Respectively represent the gradients of the corresponding parameters. Preferably, and All settings are made using adaptive adjustment. The initial value of is set to 0.01. The initial value of is set to 0.03. In one embodiment, define and They are the loss change rates of the generator and the discriminator respectively, and the calculation method is expressed as:
[0127]
[0128] In the formula, and Represents the derivative of the generator and discriminator loss functions with respect to time, which is used to detect the speed of change of the loss.
[0129] Furthermore, in this embodiment, and The adaptive setting is expressed as:
[0130]
[0131] In the formula, and are the learning rates of the generator and discriminator at the tth iteration, and is the initial learning rate of the generator and discriminator; and are the first adjustment coefficient and the second adjustment coefficient respectively. Preferably, and are set to 0.01 and 0.03 respectively, and Set to 0.3 and 0.7 respectively.
[0132] Furthermore, the total loss function of the generative adversarial network includes the generative adversarial loss, feature matching loss, and quantum state fidelity loss. The total loss function of the generative adversarial network is calculated to continuously optimize the performance of the generator and the discriminator. The calculation method of the total loss function of the generative adversarial network is expressed as:
[0133]
[0134]
[0135]
[0136] In the formula, is the L2 norm, , , They are the weight coefficients for generating adversarial loss, feature matching loss, and quantum state fidelity loss respectively; , , They are generative adversarial loss, feature matching loss, and quantum state fidelity loss; represents the feature extraction function; represents the inner product of quantum states; is the absolute value symbol. Preferably, , , Set them to 1.0, 0.5, and 0.5 respectively.
[0137] In one embodiment, the adversarial loss is generated The log-likelihood function is used and the calculation method is expressed as:
[0138] In the formula, x i represents the i-th sample input to the generative adversarial network, specifically a sample from a real dataset or generated data; nc represents the number of samples generated by the current batch input to the generative adversarial network, y i is the sample label of the i-th input to the generative adversarial network (1 for real data and 0 for generated data), is the output of the discriminator.
[0139] 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 to reach the preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times. After the data expansion model training is completed, the trained data expansion model is used to increase the number of samples. In one embodiment, assuming that the original collected samples are 800, and the data expansion model expands and generates 200 samples, the expanded data set contains 1000 samples. Among them, the training sample set can be used to train feature extraction models, feature dimensionality reduction models and high-order neural networks. In one embodiment, the feature extraction model can be used to extract features from the training sample set, and then the feature dimensionality reduction model can be trained. Furthermore, the feature dimensionality reduction model can also be used to reduce the dimension of its training samples and then train the high-order neural network. In summary, the embodiment of the present invention effectively solves the problem of insufficient training data by adopting adaptive feature distillation to generate adversarial networks, learns and distills key features from real data sets, and guides the generator to simulate data distribution more accurately.
[0140] Furthermore, an embodiment of the present invention also provides a data processing device for predicting a vehicle lubricant replacement cycle. Figure 4 A schematic diagram of the structure of a data processing device for predicting the replacement cycle of vehicle lubricating oil provided by an embodiment of the present invention is shown. Figure 4The device includes: a data acquisition module 100, which is used to acquire the maintenance record and vehicle monitoring parameters of the target vehicle, and assemble the maintenance record and vehicle monitoring parameters into vector data; a feature extraction module 200, which is used to input the vector data into a pre-built feature extraction model and output the key features of the vector data; wherein the feature extraction model uses a preset search strategy to update the neural network parameters, and dynamically adjusts the key parameters in the search strategy by fusing historical iteration information; a feature dimension reduction module 300, which is used to reduce the dimension of the key parameters through a pre-built feature dimension reduction model; wherein the number of network nodes and layer weights of the feature dimension reduction model are determined based on a dynamic topology structure adjustment mechanism; an execution module 400, which is used to use a high-order neural network as a classification algorithm to classify the key features after dimension reduction and output a classification result; wherein the classification result is used to characterize the lubricant replacement attribute of the target vehicle; the weight gradient of the high-order neural network is based on the mutual information between the input and the output, and is adjusted using a high-order training mechanism of gradient penalty; an output module 500, which is used to predict the lubricant replacement cycle of the target vehicle based on the classification result.
[0141] An embodiment of the present invention provides a data processing device for predicting vehicle lubricant replacement cycles, and its implementation principle and technical effects are the same 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.
[0142] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another data processing device for predicting the vehicle lubricant replacement cycle. Figure 5 A schematic diagram of the structure of another data processing device for predicting the lubricant replacement cycle of a vehicle provided by an embodiment of the present invention is shown. The classification results include normal replacement and early replacement; the output module 500 is also used to determine the classification result as the prediction result of the lubricant replacement cycle of the target vehicle. The execution module 400 is also used to obtain the mutual information between the input and output of the high-order neural network, and to solve the weight gradient of the high-order neural network using the chain rule; based on the mutual information and the preset gradient penalty term, the weight gradient is updated. The execution module 400 is also used to update the model parameters of the high-order neural network using a gradient-based method.
[0143] The feature extraction module 200 is also used to use the preset exponential decay term as a weight to calculate the parameter adjustment factor corresponding to the loss function value of the feature extraction model; based on the parameter adjustment factor and the iteration of the preset search strategy, the search step of the search strategy is adjusted to adjust the key parameters in the search strategy. Wherein, the search strategy includes the gray wolf optimization algorithm; wherein the key parameters of the gray wolf optimization algorithm include global search parameters and local search parameters. Wherein, the feature dimensionality reduction model includes an autoencoder; the feature dimensionality reduction module 300 is also used to train the autoencoder through a preset training sample set, and evaluate the importance of each feature of the training sample set in the decoding process through a sensitivity scoring function; based on the importance, the weight importance of each layer of the network of the autoencoder is adjusted; using an approximate method, the gradient of the loss function of the autoencoder with respect to the number of network nodes is calculated; based on the gradient, as well as the training progress and reconstruction error of the autoencoder, the number of network nodes of the autoencoder is adjusted.
[0144] The device also includes a construction module 600 for obtaining a pre-constructed initial sample set; using a deep learning attention mechanism to perform feature distillation processing on the initial sample set to extract key feature vectors in the initial sample set; based on the key feature vectors, data expansion is performed on the initial sample set to construct a training sample set. Wherein, data expansion is performed through a generative adversarial network, and the total loss function of the generative adversarial network includes generative adversarial loss, feature matching loss, and quantum state fidelity loss; the above-mentioned construction module 600 is also used to perform quantum gate operations on the parameters of the generative adversarial network based on preset Euler angle parameters to initialize the parameters of the generative adversarial network.
[0145] 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 The embodiment of the present invention also provides a structural diagram 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. Figure 6In 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 6 Only 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.
[0146] 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.
[0147] A computer program product of a data processing method and device for predicting a vehicle lubricant replacement cycle provided by 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 previous method embodiment, which will not be repeated here. 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 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 several instructions for 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: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a disk or an optical disk, and other media that can store program codes. 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. A data processing method for predicting vehicle lubricant replacement cycle, characterized in that: The method comprises: Acquire maintenance records and vehicle monitoring parameters of a target vehicle, and assemble the maintenance records and the vehicle monitoring parameters into vector data; Input the vector data into a pre-built feature extraction model, and output key features of the vector data; wherein the feature extraction model uses a preset search strategy to update neural network parameters, and dynamically adjusts key parameters in the search strategy by fusing historical iteration information; The key parameters are subjected to dimensionality reduction processing by a pre-built feature dimensionality reduction model; wherein the number of network nodes and layer weights of the feature dimensionality reduction model are determined based on a dynamic topology structure adjustment mechanism; A high-order neural network is used as a classification algorithm to classify the key features after dimensionality reduction and output classification results; wherein the classification results are used to characterize the lubricant replacement attributes of the target vehicle; the weight gradient of the high-order neural network is adjusted based on the mutual information between the input and output, using a high-order training mechanism with gradient penalty; Based on the classification result, a lubricant replacement period of the target vehicle is predicted.
2. The method according to claim 1, characterized in that: The classification results include normal replacement and early replacement; Based on the classification result, the step of predicting the lubricant replacement cycle of the target vehicle includes: The classification result is determined as a prediction result of a lubricant replacement period of the target vehicle.
3. The method according to claim 1, characterized in that The weight gradient of a high-order neural network is based on the mutual information between input and output, and the steps of adjusting it using a high-order training mechanism with gradient penalty include: Obtaining mutual information between input and output of the high-order neural network, and solving the weight gradient of the high-order neural network using the chain rule; The weight gradient is updated based on the mutual information and a preset gradient penalty term.
4. The method according to claim 3, characterized in that The method further comprises: A gradient-based method is used to update the model parameters of the high-order neural network.
5. The method according to claim 1, characterized in that The step of dynamically adjusting key parameters in the preset search strategy by fusing historical iteration information in the feature extraction model includes: Using a preset exponential decay term as a weight, calculating a parameter adjustment factor corresponding to the loss function value of the feature extraction model; Based on the parameter adjustment factor and the iteration status of the preset search strategy, the search step size of the search strategy is adjusted to adjust the key parameters in the search strategy.
6. The method according to claim 5, characterized in that The search strategy includes a gray wolf optimization algorithm; Among them, the key parameters of the gray wolf optimization algorithm include global search parameters and local search parameters.
7. The method according to claim 1, characterized in that The feature dimensionality reduction model includes an autoencoder; the step of determining the number of network nodes and layer weights of the feature dimensionality reduction model based on a dynamic topology structure adjustment mechanism includes: The autoencoder is trained using a preset training sample set, and the importance of each feature of the training sample set in the decoding process is evaluated using a sensitivity scoring function; Based on the importance, adjusting the weight importance of each layer of the autoencoder network; Using an approximate method, the gradient of the loss function of the autoencoder with respect to the number of network nodes is calculated; Based on the gradient, as well as the training progress and reconstruction error of the autoencoder, the number of network nodes of the autoencoder is adjusted.
8. The method according to claim 1, characterized in that The method further comprises: Get a pre-built initial sample set; Using a deep learning attention mechanism, the initial sample set is subjected to feature distillation processing to extract key feature vectors from the initial sample set; Based on the key feature vector, data expansion is performed on the initial sample set to construct a training sample set.
9. The method according to claim 8, characterized in that Data augmentation is performed by generating an adversarial network, wherein the total loss function of the generated adversarial network includes a generated adversarial loss, a feature matching loss, and a quantum state fidelity loss; the method further comprises: Based on the preset Euler angle parameters, quantum gate operations are performed on the parameters of the generative adversarial network to initialize the parameters of the generative adversarial network.
10. A data processing device for predicting the replacement cycle of vehicle lubricating oil, characterized in that: The device comprises: A data acquisition module, used to acquire the maintenance record and vehicle monitoring parameters of the target vehicle, and assemble the maintenance record and the vehicle monitoring parameters into vector data; A feature extraction module, used to input the vector data into a pre-built feature extraction model and output key features of the vector data; wherein the feature extraction model uses a preset search strategy to update neural network parameters, and dynamically adjusts key parameters in the search strategy by fusing historical iteration information; A feature dimension reduction module, used for reducing the dimension of the key parameters through a pre-built feature dimension reduction model; wherein the number of network nodes and layer weights of the feature dimension reduction model are determined based on a dynamic topology structure adjustment mechanism; An execution module is used to classify the key features after dimensionality reduction by using a high-order neural network as a classification algorithm, and output a classification result; wherein the classification result is used to characterize the replacement attribute of the lubricant oil of the target vehicle; the weight gradient of the high-order neural network is adjusted based on the mutual information between the input and the output, using a high-order training mechanism of gradient penalty; An output module is used to predict the lubricant replacement cycle of the target vehicle based on the classification result.
Citation Information
Patent Citations
BIM-based crane rolling bearing fault detection method and system
CN116878880A
Energy facility fault prediction method and device based on deep learning, and electronic equipment
CN116956197A
Commercial vehicle engine lubricating oil maintenance management method and system, terminal and medium
CN117726323A
Intelligent equipment lubrication monitoring method and device based on artificial intelligence
CN118032327A
Text instruction intention recognition method and device based on natural language processing
CN118503798A
Cited By
Real-time wheel center force prediction method and device and server
CN120068282A
Vehicle sale prediction method and device
CN120822990A
