Method and device for predicting residual life of equipment, equipment and storage medium

Abstract

The invention discloses an equipment residual life prediction method and device, equipment and a storage medium. The method comprises the following steps: acquiring equipment attribute data corresponding to target equipment; and inputting the equipment attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data. And inputting the attribute feature data into a pre-trained target feature dimension reduction model for feature dimension reduction processing to obtain attribute dimension reduction data. And inputting the attribute dimension reduction data into a pre-trained target life prediction model for residual life prediction to obtain the residual life of the predicted equipment, thereby solving gradient explosion in high-dimensional data processing and redundant information in high-dimensional data features, improving the feature expression ability, and further improving the accuracy of predicting the residual life of the equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method, apparatus, device and storage medium for predicting the remaining life of equipment. Background Art

[0002] With the rapid development of industrial automation and intelligent manufacturing, equipment health management and preventive maintenance have become increasingly important. In this context, accurately predicting the remaining life of equipment is crucial for reducing operation and maintenance costs, avoiding production accidents, and improving efficiency.

[0003] While existing technologies can achieve some degree of remaining life prediction using neural network models, they often face challenges with insufficient training data. Furthermore, existing neural networks are prone to vanishing or exploding gradients when processing high-dimensional data, impacting model training stability and the effectiveness of feature extraction. Traditional feature dimensionality reduction methods often fail to effectively remove redundant information and maintain the key structural characteristics of the reduced data, resulting in weak model generalization and low prediction accuracy. Summary of the Invention

[0004] The present invention provides a method, apparatus, device and storage medium for predicting the remaining service life of equipment to solve the problems of gradient explosion in high-dimensional data processing and redundant information in high-dimensional data features, thereby improving the expressiveness of features and thereby improving the accuracy of predicting the remaining service life of equipment.

[0005] According to one aspect of the present invention, a method for predicting the remaining life of equipment is provided. The method comprises:

[0006] Get the device attribute data corresponding to the target device;

[0007] Inputting the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data;

[0008] Inputting the attribute feature data into a pre-trained target feature dimensionality reduction model to perform feature dimensionality reduction processing to obtain attribute dimensionality reduction data;

[0009] The attribute dimension reduction data is input into a pre-trained target life prediction model to perform remaining life prediction to obtain the predicted remaining life of the equipment.

[0010] According to another aspect of the present invention, a device for predicting the remaining life of equipment is provided. The device comprises:

[0011] Device attribute data acquisition module, used to obtain device attribute data corresponding to the target device;

[0012] An attribute feature data determination module is used to input the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data;

[0013] An attribute dimension reduction data determination module is used to input the attribute feature data into a pre-trained target feature dimension reduction model to perform feature dimension reduction processing to obtain attribute dimension reduction data;

[0014] The device remaining life prediction module is used to input the attribute dimension reduction data into a pre-trained target life prediction model to perform remaining life prediction and obtain a predicted remaining life of the device. According to another aspect of the present invention, an electronic device is provided, comprising:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the device remaining life prediction method described in any embodiment of the present invention.

[0018] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the device remaining life prediction method described in any embodiment of the present invention when executed.

[0019] The technical solution of the embodiment of the present invention obtains the device attribute data corresponding to the target device. The device attribute data is input into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute feature data is input into a pre-trained target feature dimensionality reduction model for feature dimensionality reduction processing to obtain attribute dimensionality reduction data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute dimensionality reduction data is input into a pre-trained target life prediction model for remaining life prediction to obtain the predicted remaining life of the device, thereby improving the model's ability to capture complex patterns in equipment operation data, thereby improving the accuracy of remaining life prediction.

[0020] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0022] Figure 1 This is a flow chart of a method for predicting the remaining life of equipment provided in accordance with the first embodiment of the present invention;

[0023] Figure 2 This is a structural diagram of a device for predicting the remaining life of equipment provided in accordance with a second embodiment of the present invention;

[0024] Figure 3 The present invention is a schematic diagram of the structure of an electronic device for implementing the device remaining life prediction method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described are only 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 making creative efforts should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", etc. in the description and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0027] Example 1

[0028] Figure 1 A flowchart of a method for predicting the remaining service life of a device is provided for the first embodiment of the present invention. This embodiment is applicable to the case of predicting the remaining service life of a device. The method can be executed by a device remaining service life prediction device. The device remaining service life prediction device can be implemented in the form of hardware and / or software. The device remaining service life prediction device can be configured in an electronic device. Figure 1 As shown, the method includes:

[0029] S101: Obtain device attribute data corresponding to a target device.

[0030] The target device may refer to any device for which the remaining life is to be predicted. The device attribute data may refer to attribute information of the target device, such as ax (operating voltage), ay (operating current), az (ambient temperature), aw (ambient humidity), av (device vibration index), au (operating frequency), at (number of device starts), as (number of maintenance visits), ar (number of failures), and aq (device operating time).

[0031] It should be noted that this embodiment is only intended to illustrate one data format and type of the present invention. In actual applications, the attributes of a target device are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds.

[0032] Furthermore, the collected data is annotated. The annotating method of the present invention is manual annotating. In one embodiment, the annotated content is the remaining life in hours.

[0033] S102: Input the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data.

[0034] The target feature extraction model is a neural network model based on a preset wolf pack optimization algorithm. For example, the preset wolf pack optimization algorithm can be an adaptive wolf pack optimization algorithm. The target feature extraction model is trained based on device attribute samples and utilizes a three-layer fully connected neural network for feature extraction.

[0035] Specifically, device attribute data is fed into a target feature extraction model for feature extraction. Based on the output of the target feature extraction model, attribute feature data is obtained. The pre-set wolf pack optimization algorithm enhances the efficiency and accuracy of the neural network's search in multidimensional data space, effectively improving the stability and effectiveness of feature extraction.

[0036] Exemplarily, the training process of the target feature extraction model includes:

[0037] Obtain device attribute samples and feature extraction samples corresponding to the device attribute samples; input the device attribute samples into a pre-acquired preset feature extraction model for feature extraction to obtain an input extraction result; determine the loss function gradient based on the input extraction result and the feature extraction samples; update the network parameters corresponding to the preset feature extraction model based on the loss function gradient and historical neural network parameters; and determine the preset feature extraction model as the target feature extraction model when a preset stop iteration condition is met.

[0038] The device attribute samples are obtained by expanding historical attribute data using a target data expansion model. The device attribute samples can be used to train a preset feature extraction model to obtain a target feature extraction model. The feature extraction samples can be used to verify the training effect of the preset feature extraction model.

[0039] It should be noted that the preset feature extraction model includes a preset wolf pack optimization algorithm. The preset wolf pack optimization algorithm can be an adaptive wolf pack optimization algorithm. The adaptive wolf pack optimization algorithm adjusts the parameters of the neural network by simulating the social behavior of the wolf pack. Each wolf represents a set of potential solutions for the neural network, and the social behavior of the wolf pack helps explore the solution space and find the optimal solution. The wolf pack position update method is expressed as:

[0040]

[0041] Where, and are the positions of the wolf at iterations t and t+1 respectively; x cp is the optimal position of an individual in the wolf pack; x cg is the global optimal position; α c and β c is a parameter that balances exploration and development. c and β c Set them to 0.3 and 0.5 respectively.

[0042] 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:

[0043]

[0044] Where, is the updated α c ;λ c is the adjustment factor used to adjust α c ; ∥∥ indicates the Euclidean distance.

[0045] In one embodiment, the adjustment factor λ cAdjustments are made based on feedback from historical optimization performance, and the calculation method is expressed as:

[0046]

[0047] Where, T ve is the number of historical iterations considered, indicating how many past generations’ performance influences the current adjustment; is the attenuation factor, γ c 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 optimal position x of the individual in the wolf pack of the t-th iteration cp The corresponding neural network parameters. Preferably, γ c Set to 3.

[0048] Specifically, device attribute samples, feature extraction samples, and a preset feature extraction model are obtained. The preset feature extraction model includes weights and biases, and the network parameters of the preset feature extraction model are initialized. Exemplarily, the initialization method is expressed as:

[0049]

[0050] Where, represents the initial value of the neural network parameters; σ cs is the initialization standard deviation; G t (0,I) represents a Gaussian distribution with mean 0 and covariance as the identity matrix. Preferably, σ cs Set to 0.01.

[0051] The device attribute samples are input into a pre-acquired preset feature extraction model for feature extraction, and an input extraction result is obtained based on the output of the preset feature extraction model. The gradient of the loss function is determined based on the input extraction result and the feature extraction samples.

[0052] For example, the loss function gradient The calculation method is expressed as:

[0053]

[0054] Where M is the number of samples in each batch of training data; x cm is the feature extraction sample of the mth sample; y cm is the true label of the mth sample; l() is the cross entropy loss function; f(x cm θ c ) is the input extraction result.

[0055] Furthermore, based on the loss function gradient and the historical neural network parameters, the network parameters corresponding to the preset feature extraction model are updated. Exemplarily, the updating method is expressed as:

[0056]

[0057] Where, and are the network parameters of the preset feature extraction model for the tth iteration and the t+1th iteration respectively; η c is the learning rate; In the parameter The gradient of the loss function at . Preferably, η c Set to 0.01.

[0058] 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.

[0059] The present invention adopts a neural network based on an adaptive wolf pack optimization algorithm. On the basis of the traditional wolf pack optimization algorithm, an adaptive mechanism is adopted to adjust the search strategy according to the previous optimization results, thereby improving the search efficiency and accuracy of the algorithm in high-dimensional data space.

[0060] Exemplarily, the training process of the target data expansion model includes:

[0061] Acquiring historical attribute data and a preset data expansion model, and determining an initial quantum state corresponding to the preset data expansion model;

[0062] Based on a random noise vector, the initial quantum state is transformed according to a predefined quantum gate operation to determine a generator quantum state corresponding to the generator, wherein the quantum gate operation is a single-qubit rotation gate;

[0063] Performing a measurement operation on the generator quantum state to obtain generated extended data;

[0064] Determining, by the discriminator, a loss function corresponding to the generator based on the generated augmented data and the historical attribute data, and adjusting internal parameters of the generator based on the loss function;

[0065] When a preset condition for stopping iteration is met, the preset data expansion model is determined as a target data expansion model.

[0066] It should be noted that the preset data augmentation model includes a generator and a discriminator. The goal of the generator is to generate new data that is as close to the real device data as possible, and the task of the discriminator is to distinguish whether the input data is real or generated by the generator. 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 and affect the accuracy of the model. The present invention uses a generative adversarial network algorithm based on adaptive quantum random walks for sample generation, thereby achieving data augmentation.

[0067] Specifically, the preset data augmentation model is initialized, the initial quantum state is constructed, the initial state is set for the generator and the discriminator, and the initial quantum state is obtained. For example, the initial quantum state can be expressed as:

[0068]

[0069] In the formula, |ψ init > represents the initial quantum state, H i represents the Hadamard gate acting on the i-th quantum bit, which is used to generate a uniform quantum superposition state; ny represents the number of quantum bits.

[0070] The generator receives a random noise vector and transforms the initial quantum state according to a predefined quantum gate operation, where the quantum gate operation is a single-qubit rotation gate, thereby obtaining the generator quantum state corresponding to the generator. For example, the quantum gate operation is used to transform it into the generator quantum state, which is expressed as:

[0071] |φ g >=U(θ g ,φ g ,λ g )|ψ init >

[0072] Where, |φ g > is the generator quantum state of the generator, U() represents the quantum state of the generator with parameters θ g 、φ g ,λ g The quantum gate operation defined, θ g is the parameter of the generator, φ g is the parameter of the discriminator, λ g is the quantum gate coefficient. Preferably, λ g Set to 2π.

[0073] In one embodiment, the quantum gate operation U(θ g ,φ g ,λ g ) is defined as a single-qubit rotation gate, and the calculation method is expressed as:

[0074] U(θ g ,φ g ,λ g )=R z (φ g )R y (θ g )R z (λ g )

[0075] Where, is the rotation about the Z axis, σ z is the Pauli-Z matrix; is the rotation about the Y axis, σ y is the Pauli-Y matrix.

[0076] The generator measures the quantum state and collapses it to a specific data instance, simulating the state collapse process in quantum computing to generate new generative expansion data. The method of generating new generative expansion data is expressed as:

[0077] D g =Measure(|φ g >)

[0078] Where D g It represents the generated extended data generated by the generator, and Measure() represents the measurement operation on the quantum state.

[0079] The discriminator receives the generated augmented data and real historical attribute data from the generator and calculates the loss function. The calculation method is expressed as:

[0080]

[0081] Where, L D represents the loss function of the discriminator, D(x) is the output of the discriminator for the historical attribute data x, G(z) is the data generated by the generator based on the random noise vector z, and p data and p z Represent the distribution of historical attribute data and noise data respectively, represents expectation, and ~ represents compliance with a specific distribution.

[0082] Furthermore, the discriminator loss function L D The calculation method of the expectation term is expressed as:

[0083]

[0084]

[0085] Where p data(x) is the probability density function of historical attribute data, p z (z) is the probability density function of the noise data.

[0086] According to the loss function output by the discriminator, the parameters of the generator are adjusted. The generator is expressed as follows according to the loss function of the discriminator:

[0087]

[0088] Where, and Represent the parameters of the generator at iteration t+1 and t, respectively, vg is the learning rate, is the loss function L of the generator G About parameter θ g Preferably, η vg Set to 0.01.

[0089] For example, the loss function The calculation method is expressed as:

[0090]

[0091] Where, represents the sensitivity of the discriminator output to the generated data, is the output of the generator G to its parameter θ g gradient.

[0092] 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.

[0093] The present invention adopts a loss function based on quantum information theory, which can more accurately reflect the differences between quantum states and the authenticity of generated data, thereby improving the adaptability and diversity of the model to new data generation.

[0094] S103: Input the attribute feature data into a pre-trained target feature dimensionality reduction model to perform feature dimensionality reduction processing to obtain attribute dimensionality reduction data.

[0095] The target feature dimensionality reduction model uses an autoencoder neural network model based on mask enhancement. The target feature dimensionality reduction model can include an encoder and a decoder, which achieve data compression and reconstruction by learning an effective representation of the input data.

[0096] Specifically, attribute feature data is fed into a target feature dimensionality reduction model for feature dimensionality reduction. Based on the output of the target feature dimensionality reduction model, attribute dimensionality reduction data can be obtained. Using a target feature dimensionality reduction model based on mask enhancement solves the problem of redundant information in high-dimensional data features and improves the expressiveness of features.

[0097] Exemplarily, the training process of the target feature dimensionality reduction model includes:

[0098] Acquire feature extraction data samples and a preset feature dimensionality reduction model, and initialize the preset feature dimensionality reduction model; mask the feature extraction data samples based on a random mask matrix to obtain masked feature data samples; optimize the layer model according to the masked feature data samples, the encoder and the decoder to determine the model parameters of each layer in the preset feature dimensionality reduction model; integrate the layer model parameters to obtain an integrated feature dimensionality reduction model, and iteratively train the integrated feature dimensionality reduction model; when the preset stop iteration condition is met, determine the integrated feature dimensionality reduction model as the target feature dimensionality reduction model.

[0099] It should be noted that the preset feature dimensionality reduction model includes an encoder and a decoder. Specifically, the preset feature dimensionality reduction model is initialized, including the weights and bias parameters of the autoencoder network. For example, let Θ (l) is the parameter set of the lth layer, including weights and biases, that is, Θ (l) ={W tu (l) ,b tu (l)}, and W tu (l) is the weight matrix of the lth layer, b tu (l) is the bias vector of layer l. In one embodiment, the initialization method is random initialization.

[0100] For each feature extraction data sample, the feature extraction data sample is masked based on the random mask matrix to obtain a masked feature data sample. Exemplarily, the method of generating the masked feature data sample is expressed as:

[0101] m j ~Ber(p)

[0102] Where m j is the jth element of the mask vector m, p is the mask ratio; Ber() is the Bernoulli distribution discrete function, which is used to generate the mask matrix, and each element m jIndependently drawn from the Ber(p) distribution, the probability that the element has p is set to 1 (i.e., selected or activated), and the probability of 1-p is set to 0 (i.e., covered or inactivated).

[0103] Furthermore, the calculation method of the mask ratio p is expressed as:

[0104]

[0105] In the formula, Var(X TH ) represents the input data X TH The variance of θ tu It is the preset variance threshold, and min() represents the minimum value function.

[0106] The masked feature data samples after mask processing are input into the encoder and decoder respectively for layer model optimization, so as to obtain the model parameters of each layer in the preset feature dimensionality reduction model.

[0107] Exemplarily, before determining the model parameters of each layer in the preset feature dimensionality reduction model, it also includes: setting a training cycle and a learning rate for each layer in the preset feature dimensionality reduction model; and determining the layer model parameters in the preset feature dimensionality reduction model based on the training cycle and the learning rate.

[0108] Specifically, according to the framework of the greedy algorithm, an independent training cycle and learning rate are set for each layer of the preset feature dimensionality reduction model to ensure that each layer can reach the local optimum before being added to the overall network. For example, the number of independent training cycle iterations T for layer l is l Set to:

[0109]

[0110] Where Ep is the total number of iterations, Ler is the total number of layers, δ l,Ler is the Kronecker function, which ensures that the last layer can have more training cycles.

[0111] After all layers are trained, the models of each layer are integrated to form a complete autoencoder model, and the overall fine-tuning is performed to optimize the global performance of the model. The fine-tuning method is expressed as follows:

[0112]

[0113] Where, Θ fine-tune is the fine-tuned parameter, N tu is the number of training samples, is the i-th training sample, Indicates that the parameter is and Θ are loss functions.

[0114] 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.

[0115] Exemplarily, the performing layer model optimization according to the mask feature data sample, the encoder, and the decoder to determine the model parameters of each layer in the preset feature dimensionality reduction model includes:

[0116] For each layer in the preset feature dimensionality reduction model, the mask feature data sample is input into the encoder for data compression processing to obtain compressed low-dimensional data; the compressed low-dimensional data is input into the decoder for data reconstruction processing to obtain reconstructed original data; based on the reconstructed original data and the feature extraction data sample, the training error corresponding to the layer model is determined; according to the training error, the network parameters corresponding to the layer model are optimized to determine the layer model parameters in the preset feature dimensionality reduction model.

[0117] Specifically, the masked feature data sample after mask processing is input to the encoder, and the encoder compresses the data into compressed low-dimensional data after processing by the activation function. The forward propagation method of the encoder is expressed as:

[0118] z tu =Sig(W tu (enc) ⊙X th +b tu (enc) )

[0119] Where z tu is the compressed low-dimensional data after encoding, Sig() is the Sigmoid activation function, W tu (enc) and b tu (enc) are the weights and biases of the encoder, and ⊙ represents the Hadamard product.

[0120] The compressed low-dimensional data is fed into the decoder, which attempts to reconstruct the data to obtain the reconstructed original data. By calculating the loss function, the backpropagation algorithm updates the network parameters corresponding to the layer model, thereby determining the layer model parameters in the preset feature dimensionality reduction model.

[0121] Exemplarily, the output of the decoder is represented as:

[0122] X th ^=Sig(W tu (dec) ·z tu +b tu (dec) )

[0123] Where, X th ^ is the reconstructed original data output by the decoder.

[0124] Furthermore, the calculation method of the loss function is expressed as:

[0125]

[0126] Where, L au is the total loss, is the reconstruction error, λ ru is the weight to control the isolated loss; ∥z tu ∥0 is the isolated loss, indicating that z tu The number of non-zero elements of is used to measure the sparsity of the code. ru Set to 0.3.

[0127] Furthermore, the isolated loss ∥z tu The calculation method of ∥0 is expressed as:

[0128]

[0129] Where n z is the vector z tu Dimensions; is the encoded feature vector of the i-th dimension; 1() is the indicator function, when The value is 1 when , otherwise it is 0.

[0130] For example, a greedy algorithm may be used to optimize the parameters of each layer in the preset feature dimensionality reduction model. In one embodiment, the parameter update method uses the Adam optimizer, and the update method is expressed as follows:

[0131]

[0132] Where, and Represent the parameters of the tth iteration and t+1th iteration of the lth layer respectively, Adam() is the Adam optimizer function, η vu is the learning rate, L au is the calculated loss, is the gradient of the parameters of the t-th iteration of the l-th layer with respect to the loss function. Preferably, η vu Set the dynamic adjustment strategy, and its initial value is set to 0.01.

[0133] In one embodiment, let the learning rate of the tth iteration be Its dynamic adjustment method is expressed as:

[0134]

[0135] Where, is the initial learning rate, is the initial reconstruction weight, is the reconstruction weight of the tth iteration, δ vu is the adjustment influence coefficient. Preferably, δ vu Set to 3, Set to 5.

[0136] Furthermore, the reconstruction weight of the tth iteration The adjustment of is based on the current effect of the reconstruction error and the sparsity weight, and is calculated as:

[0137]

[0138] Where, γ yc is the adjustment rate parameter, is the current reconstruction error, β yc is the sparsity weight, X th is the input data, X th ^ t is the reconstructed output of the tth iteration. Preferably, γ yc Set to 0.3, β yc Set to 5.

[0139] Furthermore, the gradient adjustment method in the Adam optimizer is expressed as:

[0140]

[0141] Where m^ t and v^ t are the bias-corrected first-order and second-order moment estimates of the tth iteration, and ∈ is a small constant. Preferably, ∈ is set to 0.0001.

[0142] Furthermore, the bias-corrected first-order and second-order moment estimates m^ t and v^ t The calculation method is:

[0143]

[0144] Where m t and v t is the exponential moving average of the gradient and the exponential moving average of the square of the gradient at the t-th iteration, and is the decay rate of the tth iteration. Preferably, and Set to 0.9 and 0.999 respectively.

[0145] Based on the traditional autoencoder, the present invention uses random masks in the input layer to selectively hide some features of the input data, forcing the network to reconstruct the complete output from the partially visible input data, thereby enhancing the model's ability to learn the implicit relationships in the data, and ensuring that the data after dimensionality reduction can still maintain the original data structure characteristics.

[0146] S104: Input the attribute dimension reduction data into a pre-trained target life prediction model to perform remaining life prediction to obtain the predicted remaining life of the equipment.

[0147] Among them, the target life prediction model adopts a regression model based on the Transformer algorithm with multi-attention weight coupling synchronization.

[0148] Specifically, attribute dimension reduction data is input into a pre-trained target life prediction model for remaining life prediction. Based on the output of the target life prediction model, the predicted remaining life of the equipment can be obtained. This embodiment of the present invention builds on the traditional Transformer model by integrating the outputs of the attention mechanism from different data processing layers. This improves the model's ability to capture complex patterns in equipment operating data, thereby increasing the accuracy of equipment remaining life prediction.

[0149] Exemplarily, the training process of the target life prediction model includes:

[0150] Acquire feature dimensionality reduction data samples and a preset life prediction model; input the feature dimensionality reduction data samples into the multi-head attention mechanism for parallel data processing to obtain the remaining life of the output device; compare the remaining life of the output device with the actual remaining life of the device obtained in advance to determine the training error, and adjust the network parameters of the preset remaining life prediction model according to the training error; when the preset stop iteration condition is met, determine the preset life prediction model as the target life prediction model.

[0151] The preset remaining life prediction model includes a multi-head attention mechanism. The present invention adopts a Transformer algorithm based on multi-attention weighted coupling synchronization as the preset remaining life prediction model.

[0152] Specifically, the weights and biases of the Transformer algorithm in the preset life prediction model are initialized. In one embodiment, the initialization method is expressed as:

[0153]

[0154] Where w init is the initialized model weight; σ cs is the standard deviation of initialization; represents a normal distribution with a mean of 0 and a covariance matrix of the identity matrix. Preferably, σ cs Set to 0.1.

[0155] The feature-reduced data samples are input into the preset life prediction model. The preset life prediction model processes the data in parallel through a multi-head attention mechanism. Each attention head processes different aspects of the data, enhancing the model's understanding of the data. For example, the multi-head attention mechanism processes the data in the following way:

[0156]

[0157] Where Q, K, and V are query, key, and value matrices respectively; d k is the dimension of the key vector; softmax() is the Softmax activation function; Attention() is the attention mechanism function.

[0158] For example, the calculation of key (K), query (Q) and value (V) is obtained by linear transformation of the input feature X after dimensionality reduction, and the calculation method is expressed as:

[0159] Q=XW Q

[0160] K=XW K

[0161] V=XW V

[0162] Where W Q 、W K 、W V are weight matrices for query, key, and value respectively, and the weight matrices are learned through training.

[0163] Furthermore, the outputs of each attention mechanism are integrated through a fusion layer to obtain the remaining life of the output device. Specifically, the fusion and gating adjustment of the multi-head attention output is expressed as:

[0164]

[0165] G=Sig(W G H+b G )

[0166]

[0167] Where H is the weighted sum of the attention outputs of all heads, i.e., the remaining life of the output device; n h is the number of attention heads; W i O A is the output linear transformation matrix of the i-th head;i is the output of the i-th head; G is the gating coefficient, which is generated by the linear layer with Sigmoid activation; Sig() is the Sigmoid function, W G and b G are the gated weight matrix and bias respectively; is the output after gate adjustment, and ⊙ represents element-wise multiplication.

[0168] The output device remaining life is compared with the actual device remaining life obtained in advance to determine a training error, and the network parameters of the preset remaining life prediction model are adjusted according to the training error.

[0169] The training error is calculated as:

[0170]

[0171] Where L is the training error; N is the number of samples; y i is the true label of the i-th sample; is the model prediction value; λ is the regularization coefficient; R(w) is the L2 regularization term of the weight.

[0172] In one embodiment, the regularization term R(w) uses the L2 norm of the weights and is calculated as:

[0173]

[0174] Where w k Represents the kth element in the model weight, and the L2 norm helps control the model complexity.

[0175] Exemplarily, the method of adjusting the network parameters of the preset remaining life prediction model according to the training error is expressed as follows:

[0176]

[0177] w t+1 =w t -v t+1

[0178] Where, v t is the momentum of the tth iteration; v t+1 is the momentum of the t+1th iteration; γ is the momentum coefficient; η is the learning rate; is the gradient of the loss function L with respect to the weight w; w t+1 is the weight of the t+1th iteration; w t is the weight of the tth iteration. Preferably, γ is set to 0.8 and η is set to 0.01.

[0179] Furthermore, the gradient of the loss function L with respect to the weight w is The calculation method is expressed as:

[0180]

[0181] Where, is the model's predicted output for the i-th sample.

[0182] 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.

[0183] Based on the traditional Transformer model, by fusing the attention mechanism outputs from different data processing layers, the model's ability to capture complex patterns in equipment operation data is improved, thereby improving the accuracy of equipment remaining life prediction.

[0184] The technical solution of the embodiment of the present invention obtains the device attribute data corresponding to the target device. The device attribute data is input into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute feature data is input into a pre-trained target feature dimensionality reduction model for feature dimensionality reduction processing to obtain attribute dimensionality reduction data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute dimensionality reduction data is input into a pre-trained target life prediction model for remaining life prediction to obtain the predicted remaining life of the device, thereby improving the model's ability to capture complex patterns in equipment operation data, thereby improving the accuracy of remaining life prediction.

[0185] Example 2

[0186] Figure 2 This is a schematic diagram of the structure of a device for predicting the remaining life of equipment provided by the third embodiment of the present invention. Figure 2 As shown, the device includes:

[0187] The device attribute data acquisition module 301 is used to acquire the device attribute data corresponding to the target device;

[0188] The attribute feature data determination module 302 is used to input the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data;

[0189] The attribute dimension reduction data determination module 303 is used to input the attribute feature data into a pre-trained target feature dimension reduction model to perform feature dimension reduction processing to obtain attribute dimension reduction data;

[0190] The equipment remaining life prediction module 304 is used to input the attribute dimension reduction data into a pre-trained target life prediction model to perform remaining life prediction and obtain the predicted equipment remaining life.

[0191] The technical solution of the embodiment of the present invention obtains the device attribute data corresponding to the target device. The device attribute data is input into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute feature data is input into a pre-trained target feature dimensionality reduction model for feature dimensionality reduction processing to obtain attribute dimensionality reduction data, which solves the gradient vanishing or explosion problem that traditional neural networks may encounter in high-dimensional data processing. The attribute dimensionality reduction data is input into a pre-trained target life prediction model for remaining life prediction to obtain the predicted remaining life of the device, thereby improving the model's ability to capture complex patterns in equipment operation data, thereby improving the accuracy of remaining life prediction.

[0192] Optionally, the device includes a feature extraction model training module.

[0193] The feature extraction model training module includes:

[0194] Acquire device attribute samples and feature extraction samples corresponding to the device attribute samples, wherein the device attribute samples are obtained by expanding historical attribute data using a target data expansion model;

[0195] Inputting the device attribute sample into a pre-acquired preset feature extraction model for feature extraction to obtain an input extraction result, wherein the preset feature extraction model includes a preset wolf pack optimization algorithm;

[0196] Determining a loss function gradient based on the input extraction result and the feature extraction sample;

[0197] Based on the loss function gradient and historical neural network parameters, updating the network parameters corresponding to the preset feature extraction model;

[0198] When a preset condition for stopping iteration is met, the preset feature extraction model is determined as a target feature extraction model.

[0199] Optionally, the device includes a data augmentation model training module, configured to:

[0200] Acquiring historical attribute data and a preset data expansion model, and determining an initial quantum state corresponding to the preset data expansion model, wherein the preset data expansion model includes a generator and a discriminator;

[0201] Based on a random noise vector, the initial quantum state is transformed according to a predefined quantum gate operation to determine a generator quantum state corresponding to the generator, wherein the quantum gate operation is a single-qubit rotation gate;

[0202] Performing a measurement operation on the generator quantum state to obtain generated extended data;

[0203] Determining, by the discriminator, a loss function corresponding to the generator based on the generated augmented data and the historical attribute data, and adjusting internal parameters of the generator based on the loss function;

[0204] When a preset condition for stopping iteration is met, the preset data expansion model is determined as a target data expansion model.

[0205] Optionally, the device includes a feature dimensionality reduction model training module.

[0206] Among them, the feature dimensionality reduction model training module includes:

[0207] An extraction data sample acquisition unit is used to acquire feature extraction data samples and a preset feature dimensionality reduction model, and initialize the preset feature dimensionality reduction model, wherein the preset feature dimensionality reduction model includes an encoder and a decoder;

[0208] a mask feature data sample determining unit, configured to perform mask processing on the feature extraction data sample based on a random mask matrix to obtain a mask feature data sample;

[0209] a layer model parameter determination unit, configured to perform layer model optimization based on the mask feature data sample, the encoder, and the decoder, and determine the model parameters of each layer in the preset feature dimensionality reduction model;

[0210] a layer model parameter integration unit, configured to integrate the layer model parameters to obtain an integrated feature dimensionality reduction model, and iteratively train the integrated feature dimensionality reduction model;

[0211] The target feature dimensionality reduction model determining unit is used to determine the integrated feature dimensionality reduction model as the target feature dimensionality reduction model when a preset stopping iteration condition is met.

[0212] Optionally, the layer model parameter determination unit is specifically configured to:

[0213] For each layer in the preset feature dimensionality reduction model, inputting the mask feature data sample into the encoder for data compression processing to obtain compressed low-dimensional data;

[0214] Inputting the compressed low-dimensional data into the decoder for data reconstruction processing to obtain reconstructed original data;

[0215] Determining a training error corresponding to a layer model based on the reconstructed original data and the feature extracted data samples;

[0216] The network parameters corresponding to the layer model are optimized according to the training error, and the layer model parameters in the preset feature dimensionality reduction model are determined.

[0217] Optionally, the feature dimensionality reduction model training module further includes:

[0218] The training parameter determination unit is used to:

[0219] Before determining the model parameters of each layer in the preset feature dimensionality reduction model, setting a training cycle and a learning rate for each layer in the preset feature dimensionality reduction model;

[0220] Determine layer model parameters in the preset feature dimensionality reduction model according to the training cycle and the learning rate.

[0221] Optionally, the device further includes a lifespan prediction model training module, configured to:

[0222] Obtaining a feature dimensionality reduction data sample and a preset remaining life prediction model, wherein the preset remaining life prediction model includes a multi-head attention mechanism;

[0223] Inputting the feature dimension reduction data sample into the multi-head attention mechanism for parallel data processing to obtain the remaining life of the output device;

[0224] Comparing the outputted remaining life of the device with the actual remaining life of the device obtained in advance, determining a training error, and adjusting the network parameters of the preset remaining life prediction model according to the training error;

[0225] When a preset condition for stopping iteration is met, the preset life prediction model is determined as the target life prediction model.

[0226] The device for predicting the remaining service life of equipment provided by the embodiment of the present invention can execute the method for predicting the remaining service life of equipment provided by any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0227] Example 3

[0228] Figure 3A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or claimed herein.

[0229] like Figure 3 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor. The processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the RAM 13. The processor 11, ROM 12, and RAM 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0230] Multiple components in the electronic device 10 are connected to the I / O interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0231] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors that run machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the method for predicting the remaining life of a device.

[0232] In some embodiments, the method for predicting the remaining life of a device may be implemented as a computer program tangibly embodied in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the method for predicting the remaining life of a device described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the method for predicting the remaining life of a device in any other suitable manner (e.g., via firmware).

[0233] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0234] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the computer program is executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer program may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0235] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0236] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0237] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0238] A computing system may include clients and servers. The clients and servers are typically remote from each other and typically interact via a communication network. This client-server relationship arises through computer programs running on the respective computers, creating a client-server relationship. The server may be a cloud server, also known as a cloud computing server or cloud host. This server is a hosting product within the cloud computing service ecosystem that addresses the management difficulties and limited scalability of traditional physical hosting and VPS services.

[0239] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solution of the present invention can be achieved. This is not limited herein.

[0240] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for predicting the remaining life of equipment, characterized in that: include: Get the device attribute data corresponding to the target device; Inputting the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data; Inputting the attribute feature data into a pre-trained target feature dimensionality reduction model to perform feature dimensionality reduction processing to obtain attribute dimensionality reduction data; The attribute dimension reduction data is input into a pre-trained target life prediction model to perform remaining life prediction to obtain the predicted remaining life of the equipment.

2. The method according to claim 1, characterized in that The training process of the target feature extraction model includes: Acquire device attribute samples and feature extraction samples corresponding to the device attribute samples, wherein the device attribute samples are obtained by expanding historical attribute data by a target data expansion model; Inputting the device attribute sample into a pre-acquired preset feature extraction model to perform feature extraction, and obtaining an input extraction result, wherein the preset feature extraction model includes a preset wolf pack optimization algorithm; Determining a loss function gradient based on the input extraction result and the feature extraction sample; Based on the loss function gradient and the historical neural network parameters, the network parameters corresponding to the preset feature extraction model are updated; When a preset stop iteration condition is met, the preset feature extraction model is determined as a target feature extraction model.

3. The method according to claim 2, characterized in that The training process of the target data expansion model includes: Acquire historical attribute data and a preset data expansion model, and determine an initial quantum state corresponding to the preset data expansion model, wherein the preset data expansion model includes a generator and a discriminator; Based on the random noise vector, according to a predefined quantum gate operation, the initial quantum state is transformed to determine a generator quantum state corresponding to the generator, wherein the quantum gate operation is a single quantum bit rotation gate; Performing a measurement operation on the generator quantum state to obtain generated extended data; Determining, by the discriminator, a loss function corresponding to the generator according to the generated extended data and the historical attribute data, and adjusting internal parameters of the generator according to the loss function; When a preset condition for stopping iteration is met, the preset data expansion model is determined as a target data expansion model.

4. The method according to claim 1, characterized in that: The training process of the target feature dimensionality reduction model includes: Acquire a feature extraction data sample and a preset feature dimensionality reduction model, and initialize the preset feature dimensionality reduction model, wherein the preset feature dimensionality reduction model includes an encoder and a decoder; Based on a random mask matrix, masking the feature extraction data sample to obtain a mask feature data sample; Optimizing the layer model according to the mask feature data sample, the encoder and the decoder to determine the model parameters of each layer in the preset feature dimensionality reduction model; Integrate the model parameters of each layer to obtain an integrated feature dimensionality reduction model, and iteratively train the integrated feature dimensionality reduction model; When a preset stop iteration condition is met, the integrated feature dimensionality reduction model is determined as the target feature dimensionality reduction model.

5. The method according to claim 4, characterized in that The step of performing layer model optimization according to the mask feature data sample, the encoder, and the decoder to determine the model parameters of each layer in the preset feature dimensionality reduction model includes: For each layer in the preset feature dimensionality reduction model, inputting the mask feature data sample into the encoder for data compression processing to obtain compressed low-dimensional data; Inputting the compressed low-dimensional data into the decoder for data reconstruction processing to obtain reconstructed original data; Determining a training error corresponding to a layer model based on the reconstructed original data and the feature extracted data samples; The network parameters corresponding to the layer model are optimized according to the training error, and the layer model parameters in the preset feature dimensionality reduction model are determined.

6. The method according to claim 4, characterized in that Before determining the model parameters of each layer in the preset feature dimensionality reduction model, the method further includes: For each layer in the preset feature dimensionality reduction model, setting a training cycle and a learning rate; According to the training cycle and the learning rate, layer model parameters in the preset feature dimensionality reduction model are determined.

7. The method according to claim 1, characterized in that The training process of the target life prediction model includes: Acquire feature dimension reduction data samples and a preset life prediction model, wherein the preset remaining life prediction model includes a multi-head attention mechanism; Inputting the feature dimension reduction data sample into the multi-head attention mechanism for parallel data processing to obtain the remaining life of the output device; Compare the output device remaining life with the actual device remaining life obtained in advance, determine the training error, and adjust the network parameters of the preset remaining life prediction model according to the training error; When a preset stop iteration condition is met, the preset life prediction model is determined as the target life prediction model.

8. A device for predicting the remaining life of equipment, characterized in that: include: A device attribute data acquisition module is used to acquire device attribute data corresponding to the target device; An attribute feature data determination module is used to input the device attribute data into a pre-trained target feature extraction model for feature extraction processing to obtain attribute feature data; An attribute dimension reduction data determination module is used to input the attribute feature data into a pre-trained target feature dimension reduction model for feature dimension reduction processing to obtain attribute dimension reduction data; The equipment remaining life prediction module is used to input the attribute dimension reduction data into a pre-trained target life prediction model to perform remaining life prediction and obtain the predicted equipment remaining life.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute the equipment remaining life prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the device remaining life prediction method according to any one of claims 1 to 7 when executed.

Images