Industrial internet-based device comprehensive efficiency evaluation method and device

Abstract

The application provides a kind of equipment comprehensive efficiency evaluation method and device based on industrial internet, it is related to data processing technical field, by obtaining the equipment operating parameter of industrial equipment;By feature extraction model, target feature is extracted;Based on the data flow change of target feature, the network layer connection mode of auto-encoder is adjusted, and the key low-dimensional representation in target feature is captured.Through preliminary prediction of data by extreme learning machine, the classification threshold of extreme learning machine is adjusted based on the result of preliminary prediction, to determine the efficiency level of industrial equipment.The application can adapt to data flow change, effectively capture low-dimensional feature representation, and can quickly respond to data distribution change.In addition, the neurons of the feature extraction model are assigned with stress values based on layer depth, and the neural network parameters are optimized by simulating the internal stress of the earth's crust, which can capture complex data relationships in equipment operation and improve the flexibility and accuracy of classification.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for evaluating comprehensive equipment efficiency based on industrial interconnection. Background Art

[0002] The continuous advancement of industrial automation and intelligence is placing higher demands on the operational efficiency and maintenance management of industrial equipment. Traditional equipment efficiency assessment methods typically rely on manual monitoring or simple automated systems. These methods often fail to fully capture the complex data relationships within equipment operation and lack flexibility and accuracy in data processing and efficiency assessment. Furthermore, due to the diversity of industrial equipment types and operating environments, equipment data exhibits high dimensionality and complex distribution characteristics, making effective data analysis and feature extraction more difficult. Summary of the Invention

[0003] In view of this, the purpose of the present invention is to provide a method and device for evaluating the comprehensive efficiency of equipment based on industrial interconnection, which can comprehensively capture the complex data relationships in equipment operation and improve the flexibility and accuracy of classification.

[0004] In a first aspect, an embodiment of the present invention provides a method for evaluating the comprehensive efficiency of equipment based on industrial interconnection, wherein the method includes: obtaining equipment operating parameters of industrial equipment; extracting target features from the equipment operating parameters through a pre-built feature extraction model; wherein neurons of the feature extraction model are assigned pressure values ​​based on layer depth, and the neural network parameters are optimized by simulating the internal stress of the earth's crust; the layer depth is used to characterize the number of layers of neurons; inputting the target features into a preset autoencoder, adjusting the network layer connection method of the autoencoder based on the data flow changes of the target features, and capturing the key low-dimensional representations in the target features; inputting the key low-dimensional representations into a preset extreme learning machine for preliminary prediction, and adjusting the classification threshold of the extreme learning machine based on the results of the preliminary prediction; classifying the key low-dimensional representations based on the adjusted classification threshold to determine the efficiency level of the industrial equipment.

[0005] In combination with the first aspect, an embodiment of the present invention also provides a first implementation method of the first aspect, wherein the target feature is input into a preset autoencoder, and the step of adjusting the network layer connection mode of the autoencoder based on the data flow change of the target feature includes: inputting the target feature into the preset autoencoder, encoding it through the multi-layer neural network of the autoencoder; dynamically reconstructing the topology of each layer of the neural network of the autoencoder through an adaptive activation function, and adjusting the connection structure of the network layer of the autoencoder.

[0006] In combination with the first aspect, an embodiment of the present invention also provides a second implementation of the first aspect, wherein the above method also includes: training a preset autoencoder through a preset training sample set, and calculating the reconstruction error corresponding to the training sample set; calculating the loss function of the autoencoder based on the reconstruction error and the preset regularization term; and updating the parameters of the autoencoder using the loss function and gradient descent method.

[0007] In combination with the first aspect, an embodiment of the present invention also provides a third implementation of the first aspect, wherein the step of adjusting the classification threshold of the extreme learning machine based on the result of the preliminary prediction includes: comparing the result of the preliminary prediction with the predicted output to determine the error size; based on the error size and the variance of the error size, calculating a new classification threshold to adjust the classification threshold of the extreme learning machine.

[0008] In combination with the first aspect, an embodiment of the present invention also provides a fourth implementation of the first aspect, wherein the above method also includes: performing classification training on a preset extreme learning machine through a preset training sample set; performing a gradient accumulation operation during the training process of the extreme learning machine, and updating the parameters of the extreme learning machine based on the accumulated gradient.

[0009] In combination with the first aspect, an embodiment of the present invention also provides a fifth implementation of the first aspect, wherein a method for constructing a feature extraction model includes: obtaining a preset training sample set and a feature extraction neural network; assigning a pressure value to each layer of neurons based on the layer depth parameters of each layer of neurons in the feature extraction neural network; training the feature extraction neural network through the training sample set, and calculating an adjustment value of the pressure value based on the loss function value of the feature extraction neural network; optimizing the neural network parameters of the feature extraction neural network based on the adjustment value until the feature extraction neural network meets the preset requirements, thereby constructing a feature extraction model.

[0010] In combination with the first aspect, an embodiment of the present invention also provides a sixth implementation of the first aspect, wherein the above method also includes: calculating the pressure adjustment factor of the neuron based on the activity of the neuron; and updating the parameters of the feature extraction neural network based on the pressure adjustment factor and the adjustment value.

[0011] In combination with the first aspect, an embodiment of the present invention also provides a seventh implementation method of the first aspect, wherein the above method also includes: obtaining pre-collected equipment operating parameter samples; mapping the equipment operating parameter samples to a high-dimensional kernel space, and based on the sample category of the equipment operating parameter samples, performing sample expansion on the equipment operating parameter samples to generate expanded samples; merging the expanded samples and the equipment operating parameter samples to construct a training sample set.

[0012] In combination with the first aspect, an embodiment of the present invention also provides an eighth implementation method of the first aspect, wherein the device operating parameter samples are mapped to a high-dimensional kernel space, and based on the sample category of the device operating parameter samples, the device operating parameter samples are sample expanded to generate expanded samples, including: mapping the device operating parameter samples to the high-dimensional kernel space through a kernel function; in the high-dimensional kernel space, identifying the majority class samples among the k nearest neighbors of the minority class samples of the device operating parameters, and determining the generation sample area; interpolating the generated sample area to generate synthetic samples; and using an inverse mapping function to inversely map the synthetic samples back to the original feature space to generate expanded samples.

[0013] In a second aspect, an embodiment of the present invention provides an equipment comprehensive efficiency evaluation device based on industrial interconnection, wherein the device includes: a parameter acquisition module for acquiring equipment operating parameters of industrial equipment; a feature extraction module for extracting target features in the equipment operating parameters through a pre-built feature extraction model; wherein the neurons of the feature extraction model are assigned pressure values ​​based on the layer depth, and the neural network parameters are optimized by simulating the internal stress of the earth's crust; the layer depth is used to characterize the number of neuron layers; a data processing module for inputting the target features into a preset autoencoder, adjusting the network layer connection method of the autoencoder based on the data flow changes of the target features, and capturing the key low-dimensional representations in the target features; an execution module for inputting the key low-dimensional representations into a preset extreme learning machine for preliminary prediction, and adjusting the classification threshold of the extreme learning machine based on the results of the preliminary prediction; an output module for classifying the key low-dimensional representations based on the adjusted classification threshold to determine the efficiency level of the industrial equipment.

[0014] The embodiments of the present invention bring the following beneficial effects: The embodiments of the present invention provide a method and apparatus for evaluating comprehensive equipment efficiency based on industrial interconnection. By integrating geomechanical principles into neural network design, the model can understand and learn data in a manner similar to the structure of the Earth's crust, thereby improving the model's adaptability and accuracy to high-dimensional and complexly distributed equipment operating parameters. The encoder's dynamic structural adjustment capability ensures that the model can more effectively extract target features based on changes in the data stream, enhancing learning efficiency and prediction accuracy in dynamic data environments. Combining the rapid prediction capabilities of the extreme learning machine (ELM) with a dynamic classification threshold adjustment strategy achieves rapid response to changes in data distribution, improves the flexibility and robustness of classification, and enables the model to maintain high performance in the face of data distribution offsets or noise.

[0015] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or understood by practicing the present invention. The purposes and other advantages of the present invention are realized and obtained by the structures particularly pointed out in the description, claims and drawings.

[0016] 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

[0017] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the specific embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0018] Figure 1 A flowchart of a method for evaluating comprehensive equipment efficiency based on industrial interconnection provided by an embodiment of the present invention;

[0019] Figure 2 A flowchart of another method for evaluating comprehensive equipment efficiency based on industrial interconnection provided by an embodiment of the present invention;

[0020] Figure 3 A flowchart of a method for constructing a training sample set provided by an embodiment of the present invention;

[0021] Figure 4 A schematic diagram of the structure of an equipment comprehensive efficiency evaluation device based on industrial interconnection provided by an embodiment of the present invention;

[0022] Figure 5 A schematic diagram of the structure of another device for evaluating comprehensive equipment efficiency based on industrial interconnection provided by an embodiment of the present invention;

[0023] Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of them. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0025] The embodiments of the present invention provide a method and device for evaluating comprehensive equipment efficiency based on industrial interconnection, which can comprehensively capture the complex data relationships in equipment operation and improve the flexibility and accuracy of classification.

[0026] To facilitate understanding of this embodiment, a method for evaluating comprehensive equipment efficiency based on industrial interconnection disclosed in an embodiment of the present invention is first described. Figure 1 FIG. 1 shows a flow chart of a method for evaluating comprehensive equipment efficiency based on industrial interconnection provided by an embodiment of the present invention. Figure 1 As shown, the method includes the following steps:

[0027] Step S102: Acquire equipment operating parameters of industrial equipment.

[0028] Step S104: extracting target features from the equipment operating parameters using a pre-built feature extraction model.

[0029] In specific implementation, in order to effectively evaluate the efficiency of the equipment, the embodiment of the present invention performs data processing on the equipment operating parameters of the industrial equipment. First, the target features in the equipment operating parameters are extracted through a pre-built feature extraction model. In order to solve the problem that the existing technology is difficult to effectively analyze high-dimensional and complexly distributed data, and the feature extraction of this type of data is difficult, the neurons of the feature extraction model of the embodiment of the present invention are assigned pressure values ​​based on the layer depth, and the neural network parameters are optimized by simulating the internal stress of the crust. At different depths of the network, neurons are given weights or activation thresholds that reflect geological pressure. Simulating the law of pressure changes with depth in nature enables the model to adjust its parameters in a manner similar to the crustal structure, thereby better understanding and learning the data. The pressure value of the neuron can be adjusted dynamically to ensure that the network can better simulate the geomechanical processes in the real world, so as to be able to process high-dimensional and complexly distributed data and achieve more efficient and more realistic feature representation.

[0030] The data in the embodiment of the present invention is collected from the operating parameters of various industrial equipment. The corresponding equipment operating parameters are monitored in real time by sensors and collected through the industrial Internet platform. The collected data is stored in a structured manner, specifically in JSON format.

[0031] In one embodiment, the attributes of the data include: a1 refers to the equipment operating time (hours); a2 refers to the energy consumption (kWh); a3 refers to the equipment load ratio (%); a4 refers to the temperature (degrees Celsius); a5 refers to the vibration frequency (Hz); a6 refers to the noise level (dB); a7 refers to the number of maintenance times (times / month); a8 refers to the time between failures (days); a9 refers to the output efficiency (unit / hour); a 10 Refers to the estimated life of the equipment (in years). It should be noted that this embodiment is only used to illustrate one data format and type of the present invention. In actual applications, the attributes of the data are usually more than 10 attributes, and the number of attributes of the data may reach dozens or even hundreds.

[0032] Step S106: input the target feature into a preset autoencoder, adjust the network layer connection mode of the autoencoder based on the data flow change of the target feature, and capture the key low-dimensional representation of the target feature.

[0033] In order to extract the most meaningful low-dimensional representation from the data after high-dimensional feature extraction, the complexity of the data in subsequent processing is effectively reduced and the processing efficiency is improved. The embodiment of the present invention uses a pre-built autoencoder to reduce the dimensionality of the target features and capture their key low-dimensional representations. The structure of the traditional autoencoder is predefined and fixed, which may appear to be relatively rigid when processing complex or highly variable data. The autoencoder of the embodiment of the present invention dynamically adjusts the network structure to better match the target features to adapt to changes in the data stream, which can effectively improve the efficiency and accuracy of the model in a changing data environment.

[0034] Step S108: input the key low-dimensional representation into a preset extreme learning machine for preliminary prediction, and adjust the classification threshold of the extreme learning machine based on the result of the preliminary prediction.

[0035] Step S110 , classifying the key low-dimensional representations based on the adjusted classification threshold to determine the efficiency level of the industrial equipment.

[0036] The autoencoder of the embodiment of the present invention has greater adaptability to different types of features, can more accurately capture the key low-dimensional representations of the target features, reduce the encoding of irrelevant information, and improve the efficiency and quality of feature representation. The embodiment of the present invention uses a preset extreme learning machine to identify the data and determine the classification result. The extreme learning machine achieves classification by labeling the efficiency level of the learning data. In one embodiment, the labeled categories include low efficiency level, medium efficiency level, and high efficiency level, for a total of three categories.

[0037] Traditional extreme learning machines (ELMs) typically use a fixed classification threshold, such as 0, as the decision boundary for binary classification problems. This limits their adaptability to complex decision boundaries. In cases where data is unevenly distributed or noisy, a fixed threshold may not be sufficient to accurately distinguish samples of different categories. In an embodiment of the present invention, an extreme learning machine is first used to make a preliminary prediction of the data and then adjusts its classification threshold based on the prediction results. This method of dynamically adjusting the threshold can better fit the actual data distribution, provide a more flexible and robust classification strategy, and better adapt to changes in data distribution, thereby potentially improving performance on unseen data and enhancing the adaptability and accuracy of the model.

[0038] In summary, the embodiment of the present invention provides an industrial interconnected equipment comprehensive efficiency assessment method. By incorporating geomechanical principles into neural network design, the model can understand and learn data in a manner similar to the structure of the Earth's crust, thereby improving the model's adaptability and accuracy to high-dimensional and complexly distributed industrial equipment parameters. The encoder's dynamic structural adjustment capability ensures that the model can more effectively extract target features based on changes in the data stream, enhancing learning efficiency and prediction accuracy in dynamic data environments. Combining the rapid prediction capabilities of the extreme learning machine (ELM) with a dynamic classification threshold adjustment strategy achieves a rapid response to changes in data distribution, improves the flexibility and robustness of classification, and enables the model to maintain high performance in the face of data distribution offsets or noise.

[0039] Furthermore, based on the above embodiment, the embodiment of the present invention also provides another method for evaluating the comprehensive efficiency of equipment based on industrial interconnection. Figure 2 FIG. 1 is a flow chart showing another method for evaluating comprehensive equipment efficiency based on industrial interconnection provided by an embodiment of the present invention. Figure 2 As shown, the method includes the following steps:

[0040] Step S202: Acquire equipment operating parameters of industrial equipment.

[0041] Step S204: extract target features from the equipment operating parameters using a pre-built feature extraction model.

[0042] In combination with the above embodiments, the embodiments of the present invention illustrate the optimization process of the neural network parameters of the feature extraction model. Among them, existing feature extraction technologies such as conventional neural networks may also encounter the problem of gradient disappearance or explosion, and are prone to falling into local optimal solutions, which limits the performance and stability of the model on complex data sets. In this regard, the present invention is inspired by the pressure distribution in geology and the deformation process of rock layers, and simulates the dynamic adjustment process of internal stress of the crust to optimize the weights and bias parameters of the neural network. The present invention adopts a geological pressure simulation optimization algorithm to adjust the parameters during the neural network training process, and automatically adjusts the intensity and direction of the pressure simulation according to the feedback of the loss function to achieve more accurate parameter optimization, thereby improving training efficiency and model performance. Specifically, the method for constructing the feature extraction model of the embodiment of the present invention is as follows:

[0043] 1) Obtain the preset training sample set and feature extraction neural network.

[0044] In one embodiment, the training sample set of the embodiment of the present invention corresponds to the equipment operating parameters of the industrial equipment. The feature extraction neural network used is a 3-layer fully connected neural network for feature extraction. First, the parameters of the neural network are initialized, including weights and bias In one embodiment, the initialization method is random initialization.

[0045] 2) Based on the layer depth parameters of each layer of neurons in the feature extraction neural network, a pressure value is assigned to each layer of neurons.

[0046] According to the optimization principle of geological pressure simulation, the first The initial pressure value of each neuron The initial pressure value is based on the preset layer depth parameter Distribution, the deeper the layer, the greater the initial pressure.

[0047] Among them, the geological pressure simulation optimization algorithm is used to simulate the pressure distribution in the underground rock layer and find the most suitable "mining" plan through optimization technology. This type of algorithm usually combines the principles of geomechanics, fluid mechanics and numerical simulation technology to maximize resource extraction efficiency or minimize risks under given constraints (such as cost and safety factors). The initial pressure of the neuron in this embodiment of the present invention is defined as:

[0048]

[0049] Where, is a hyperparameter that adjusts the pressure sensing depth. Preferably, Set to 5. In this embodiment, the neural network used is a 3-layer fully connected neural network, so the value of d is 3, and d1 is 1, d2 is 2, and d3 is 3.

[0050] 3) Train the feature extraction neural network using the training sample set, and calculate the adjustment value of the pressure value based on the loss function value of the feature extraction neural network.

[0051] The input data x (i.e., the training sample set) passes through the neural network and calculates the output result y of the neural network. The output result y of the neural network is calculated by the preset Softmax function. The output of the lth layer of the neural network is set to y l , calculated as:

[0052]

[0053] Where, The neural network The input of the layer, is the Sigmoid activation function, and The neural network is The weight matrix and bias vector of the layer.

[0054] The loss function of the neural network is calculated as:

[0055]

[0056] Where, is the target output, that is, the true label of the sample; is the loss function of the neural network.

[0057] Furthermore, according to the value of the loss function of the current neural network, the pressure adjustment value of each layer is calculated, thereby updating the neural network parameters.

[0058] Among them, according to the gradient of the loss function , calculate the pressure adjustment value, adjust the pressure adjustment increment of each layer, the calculation method is expressed as:

[0059]

[0060] Where, is the learning rate of the neural network, For the The pressure adjustment increment of the layer neural network. Preferably, Set to 0.3.

[0061] Furthermore, the gradient of the loss function of the neural network The calculation method is expressed as:

[0062]

[0063] Substituting the above formula into the calculation formula of pressure adjustment increment, it is expressed as:

[0064]

[0065] 4) Optimizing the neural network parameters of the feature extraction neural network based on the adjustment values ​​until the feature extraction neural network meets the preset requirements and constructing a feature extraction model.

[0066] In order to make the model pay more attention to the part that contributes most to the final output, the embodiment of the present invention also calculates the pressure adjustment factor of the neuron based on the activity of the neuron; based on the pressure adjustment factor and the adjustment value, the parameters of the feature extraction neural network are updated. Among them, by adopting a pressure-sensitive feedback mechanism, the influence of each neuron is dynamically adjusted based on the activity of the neuron, so that the network pays more attention to the part that contributes most to the final output. Specifically, define Tier The activity of neurons for:

[0067]

[0068] Where, It is Tier The net input to a neuron.

[0069] Furthermore, the pressure adjustment factor of each neuron takes into account the activity of the neuron and is calculated as:

[0070]

[0071] Where, It is Tier The stress adjustment factor of each neuron, is the regularization coefficient of the neuron, It is The total number of neurons in the layer, No. Tier The activity of neurons. Set to 0.3.

[0072] Furthermore, according to the pressure response results of each layer, the network parameters of the corresponding layer are adjusted. The adjustment method is expressed as:

[0073]

[0074]

[0075] Where, Update the symbols for the parameters, is the proportional factor adjusted according to pressure, represents element-wise product, For the The weights of the neurons in the layer, For the The bias of the neurons in the layer, For the The gradient of the weights of the neurons in the layer, For the The gradient of the bias of the layer neurons. Preferably, Set to 0.5.

[0076] The embodiment of the present invention also performs collaborative optimization on all network parameters, and transmits the pressure change effect to all relevant parameters through the back propagation mechanism to ensure the collaborative optimization of all network parameters. The calculation method of the weight gradient of the layer neurons is expressed as:

[0077]

[0078] In one embodiment, let the activation function The derivative of , then the weight gradient The calculation method is expressed as:

[0079]

[0080] Where, The neural network The transpose of the layer's input, To find the sign of the partial derivative.

[0081] 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.

[0082] In summary, the embodiment of the present invention adopts a three-layer fully connected neural network, which automatically adjusts the intensity and direction of the pressure simulation based on the feedback of the loss function and combines the geological pressure simulation optimization algorithm to update the neural network parameters, thereby solving the problems of gradient disappearance or explosion and falling into local optimal solutions during the feature extraction process.

[0083] Step S206: input the target feature into a preset autoencoder, adjust the network layer connection mode of the autoencoder based on the data flow change of the target feature, and capture the key low-dimensional representation of the target feature.

[0084] Existing autoencoder technology is commonly used as an unsupervised learning method, primarily for data dimensionality reduction, feature learning, and generative modeling. Its basic architecture consists of two parts: an encoder and a decoder. The encoder compresses the input data into a latent representation, often of lower dimensionality; the decoder attempts to reconstruct the original input data based on this latent representation. This process can be thought of as data being "encoded" by the encoder and then "decoded" by the decoder. The goal is to make the reconstructed data as close to the original input as possible, thereby forcing the model to learn an effective representation of the data. In traditional autoencoders, the network topology is fixed: the number of encoder and decoder layers, the number of neurons in each layer, and the connections between them are defined during model initialization. This static autoencoder relies primarily on training algorithms (such as gradient descent) and regularization techniques (such as dropout and L1 / L2 regularization) to optimize network weights and improve the quality of data representation.

[0085] In order to better capture the key features of industrial equipment parameters that exhibit high dimensionality and complex distribution characteristics, an embodiment of the present invention dynamically optimizes the topology of the autoencoder and automatically adjusts the connection mode of the network layer according to changes in the data stream. This allows for more flexible adaptation to the inherent structure of different data sets, dynamically optimizes the network, and more efficiently captures the key features of the data, especially when processing non-static data sets that change over time or conditions. Specifically, the target features after feature extraction are input into the improved autoencoder of the embodiment of the present invention and encoded through the multi-layer neural network of the autoencoder; the topology of each layer of the neural network of the autoencoder is dynamically reconstructed through an adaptive activation function, and the connection structure of the network layer of the autoencoder is adjusted.

[0086] The steps to capture the key low-dimensional representation are as follows:

[0087] 1) Initialize the autoencoder network parameters, including weights and biases. In one embodiment, the parameters are initialized according to a normal distribution, which can be expressed as:

[0088]

[0089]

[0090] Where, For the autoencoder Neuron to The initial weights of neurons; For the autoencoder The initial bias of each neuron; Indicates that it obeys a specific distribution. represents a normal distribution; is the variance of the autoencoder initialization. Preferably, Set to 0.01.

[0091] 2) In the encoding phase, the input data is encoded by a multi-layer neural network. Each layer of the network adaptively adjusts its structure through dynamic topology reconstruction, which can be expressed as:

[0092]

[0093]

[0094] Where, The autoencoder The weighted input sum of the layer; For the autoencoder Tier Neuron to Tier The weights of neurons; For the autoencoder Tier neurons to the Tier The bias of each neuron; The autoencoder The output of the layer; is the activation function of the autoencoder, whose input is divided by In addition, it also includes dynamically adjusted adaptive activation function parameters Preferably, The ReLU activation function calculation method is used.

[0095] In one embodiment, the adaptive activation function parameters The calculation method is:

[0096]

[0097] Where, Represents the Sigmoid activation function; and They represent the first Layer weights and biases; Represents the first The output of the layer.

[0098] 3) At the deepest level of the encoding process, the data is compressed into a low-dimensional intermediate representation. The intermediate representation is the feature vector after dimensionality reduction, and the calculation method is expressed as:

[0099]

[0100] Where, is the encoded intermediate representation of the autoencoder; is the final weighted sum of the encoding layer of the autoencoder; are the specific activation parameters of the intermediate layers of the autoencoder.

[0101] 4) In the decoding phase, starting from the intermediate representation, the data is reconstructed through the decoding network layers. Each layer of the network attempts to reconstruct the original data while retaining as much key information as possible. The decoding process is expressed as:

[0102]

[0103]

[0104] Where, Represents the decoding layer of the autoencoder, which is the opposite of the encoding layer; and They are the decoding layer of the autoencoder The input and output of the layer; is the decoding layer of the autoencoder the output of the layer; and respectively represent the weight and bias of the first decoding layer of the autoencoder; the weight and bias of the first decoding layer of the autoencoder; is a specific activation parameter of the first decoding layer of the autoencoder, and the calculation manner is the same as that of the encoding layer.

[0105] In summary, the embodiment of the present application uses an autoencoder to perform feature dimension reduction, and simultaneously performs dynamic optimization on the network topology structure, automatically adjusts the connection mode of the network layer according to the change of the data stream, so that the network structure is more suitable for the data features, and the problem that the fixed network structure cannot effectively adapt to the change of the data is solved.

[0106] Further, the autoencoder of the embodiment of the present application is trained through a preset training sample set, and the data obtained after the training sample set is subjected to feature extraction by the above-mentioned feature extraction model can be used as the training sample set. The embodiment of the present application also calculates the reconstruction error corresponding to the training sample set, calculates the loss function of the autoencoder based on the reconstruction error and a preset regularization term, and updates the parameters of the autoencoder by using the loss function and the gradient descent method.

[0107] In the training process, the loss function is calculated by using the reconstruction error and the regularization term in the encoding and decoding process, the network parameters are optimized to minimize the difference between the original input and the reconstruction output, and the calculation manner is represented as:

[0108]

[0109] In the formula, is the loss function of the autoencoder; is the input of the autoencoder; is the reconstruction output of the autoencoder; is the regularization parameter of the autoencoder; is the L2 norm. Preferably, is set to 0.3.

[0110] Further, the gradient descent method is used to update the parameters, and the loss function The calculation manner of the partial derivative of the weight is represented as:

[0111]

[0112] In the formula, and respectively represent the actual input and the network reconstruction output of the autoencoder for the i-th sample; is the total number of samples input into the autoencoder in the current batch; ​​is the partial derivative of the autoencoder's output with respect to the weights, computed via the chain rule in backpropagation.

[0113] Furthermore, during the decoding phase, the partial derivative of the output with respect to the activation input is calculated as:

[0114]

[0115] Where, Output partial derivatives of the decoding layer with respect to the activation input; is the activation function The derivative of a function with respect to its input.

[0116] Through multiple iterative training, the network parameters are continuously adjusted. The adjustment method is to update the weights and biases through the gradient descent method. The update method is:

[0117]

[0118]

[0119] Where, is the learning rate of the autoencoder; and is the gradient of the autoencoder's loss function with respect to weights and biases. Preferably, Set to 0.03.

[0120] 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.

[0121] Step S208: input the key low-dimensional representation into a preset extreme learning machine for preliminary prediction.

[0122] In the forward propagation calculation process of the extreme learning machine of the embodiment of the present invention, the input dimensionality reduction features are processed by weighted sum and activation function to generate the output of the hidden layer, and then the output layer generates the prediction result. Specifically, let The input features of the extreme learning machine, that is, the features after dimensionality reduction of the autoencoder, are used to calculate the output h of the hidden layer of the extreme learning machine and the prediction results of the output layer. as follows:

[0123]

[0124]

[0125] Where, and are the weight matrices of the hidden layer and output layer of the current iteration of the extreme learning machine; and are the bias vectors of the hidden layer and output layer of the current iteration of the extreme learning machine; is the output vector of the hidden layer of the extreme learning machine; is the prediction result of the output layer of the extreme learning machine.

[0126] Step S210 , comparing the preliminary prediction result with the prediction output to determine the error size; and calculating a new classification threshold based on the error size and the variance of the error size.

[0127] Traditional extreme learning machines (ELMs) typically use a fixed classification threshold, such as 0, as the decision boundary for binary classification problems. These methods often focus on achieving fast and efficient learning by utilizing randomly weighted hidden layers and minimizing output layer weights. Classification results are directly determined by the positive or negative value generated by the output layer's activation function (such as the sigmoid function).

[0128] The embodiment of the present invention compares the actual output with the predicted output and dynamically adjusts the threshold according to the size of the error. The embodiment of the present invention dynamically adjusts the threshold so that the model can better adapt to the distribution of categories in the data set, especially in the case of category imbalance, and can effectively improve the overall classification accuracy of the model. Moreover, in the face of noisy data or data distribution that changes over time, the dynamic threshold strategy helps the model maintain stable performance and reduce misclassification. In addition, by adjusting the threshold based on the error and its variability, the model can learn from the training data more flexibly and improve its performance on unseen data, that is, enhance the generalization ability. The dynamic threshold strategy helps prevent the model from overfitting the noise in the training data, especially when the amount of data is limited.

[0129] If the prediction is wrong, the threshold is increased to enhance the conservatism of the model; otherwise, the threshold is decreased to improve the sensitivity of the model. is the actual output, error and the adjusted threshold The calculation method is:

[0130]

[0131]

[0132] In the formula is a sign function; It is the maximum value symbol; is the prediction error of the extreme learning machine; and are the thresholds for the current and next iterations of the extreme learning machine, respectively; is the learning rate of the extreme learning machine threshold adjustment; and It is the extreme learning machine’s adjustment of the threshold sensitivity and the threshold offset.

[0133] Furthermore, define and The calculation method is:

[0134]

[0135]

[0136] Where, is the extreme learning machine scaling factor, represents the variance of the extreme learning machine error. Preferably, Set to 0.3.

[0137] Furthermore, embodiments of the present invention perform classification training on a preset extreme learning machine using a preset training sample set to obtain a trained extreme learning machine. During the training process of the extreme learning machine, embodiments of the present invention also perform a gradient accumulation operation to update the parameters of the extreme learning machine based on the accumulated gradients.

[0138] Specifically, the extreme learning machine of the present invention is based on a learnable threshold, which functions as a classifier model to identify data. Furthermore, embodiments of the present invention employ an adaptive gradient accumulation mechanism within the traditional extreme learning machine framework to fine-tune hidden layer weights. This improves the model's ability to identify abnormal data and generalize, helping to stabilize the training process and avoid overfitting.

[0139] Specifically, the training process of the extreme learning machine algorithm based on the learnable threshold is as follows:

[0140] a- Initialize the weight, bias and threshold parameters of the extreme learning machine. represents the initial value of the weight, represents the initial value of the bias, represents the initial value of the classification threshold. In one embodiment, and The initialization method is random initialization. The initial value of is set to 1.0.

[0141] b-In each iteration, not only the gradient of the current batch is calculated, but also the previous gradient information is accumulated, and the accumulated gradient is used to update the weights to achieve a smoother learning process. The extreme learning machine accumulates gradients. and update the weights using the accumulated gradients The calculation method is expressed as:

[0142]

[0143]

[0144] Where, and are the cumulative gradients of the current and previous iterations of the extreme learning machine, respectively; is the momentum factor of the extreme learning machine gradient accumulation; is the loss function of the extreme learning machine under the current weight about gradient; is the learning rate of the extreme learning machine; and are the weight matrices of the current and next iterations of the extreme learning machine, respectively. Preferably, Set to 0.3, Set to 0.01.

[0145] Further, The calculation method is expressed as:

[0146]

[0147] Where, is the loss function of the extreme learning machine For model output The partial derivative of is the output of the extreme learning machine For weight In one embodiment, the loss function of the extreme learning machine adopts cross entropy loss.

[0148] The accuracy of the classification results is fed back into the model training process, and the learning rate of the next round of extreme learning machine is adjusted according to the classification accuracy. The way to dynamically adjust the learning rate is expressed as:

[0149]

[0150] Where, and are the learning rates of the current and next iterations of the extreme learning machine; is the learning rate adjustment factor of the extreme learning machine; It's about error Preferably, Set to 0.4, is the mean error of the last five iterations.

[0151] c- 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.

[0152] In summary, the extreme learning machine of the embodiment of the present invention adopts a dynamically adjusted classification threshold, allowing the model to automatically adjust the classification decision boundary based on the training data, which is crucial for distinguishing normal data from abnormal data. Compared with a fixed threshold, a learnable threshold can more accurately reflect the distribution characteristics of the data and improve the accuracy of identifying outliers. Moreover, the embodiment of the present invention combines an adaptive gradient accumulation mechanism to train the extreme learning machine. By accumulating historical gradient information, each weight update can incorporate feedback from more training instances, which helps the model learn the global characteristics of the data rather than just fitting a single batch of data, thereby improving the model's performance on unseen data, that is, its generalization ability.

[0153] Compared to using only instantaneous gradients, cumulative gradients provide a more stable gradient signal, reducing fluctuations during training and helping to avoid drastic weight fluctuations caused by individual anomalous batches of data, thereby making the training process smoother and more stable. Furthermore, by more broadly considering data distribution information in the cumulative gradient, the embodiments of the present invention help the model learn more general patterns rather than the details of specific samples, helping to mitigate overfitting, especially when the amount of data is limited.

[0154] Step S212 : classifying the key low-dimensional representations based on the adjusted classification threshold to determine the efficiency level of the industrial equipment.

[0155] In one embodiment, the classification categories of the embodiment of the present invention include three categories: low efficiency level, medium efficiency level, and high efficiency level.

[0156] Furthermore, based on the above embodiments, the embodiments of the present invention also design a training sample set. Specifically, the embodiments of the present invention also provide another method for evaluating the comprehensive efficiency of equipment based on industrial interconnection. This embodiment mainly describes the method for constructing a training sample set. Figure 3 A flowchart of a method for constructing a training sample set is shown in FIG. Figure 3 As shown, the method includes the following steps:

[0157] Step S10: obtaining pre-collected equipment operation parameter samples.

[0158] The equipment operating parameter samples can refer to the equipment operating parameters in the above embodiment. When constructing the training sample set, the embodiment of the present invention also annotates the collected data. The annotation method of the present invention is manual annotation. The annotated labels refer to the above embodiment, such as low efficiency level, medium efficiency level, and high efficiency level, for a total of three categories.

[0159] Step S11 : Mapping the device operation parameter samples to a high-dimensional kernel space, and performing sample expansion on the device operation parameter samples based on the sample categories of the device operation parameter samples to generate expanded samples.

[0160] 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 model generalization ability and affect the accuracy of the model. To this end, the embodiment of the present invention also performs data expansion on the samples. In the specific implementation, the present invention maps the device operating parameter samples to a high-dimensional kernel space through a kernel function;

[0161] In this embodiment of the present invention, in a high-dimensional kernel space, the majority class samples are identified from the k-nearest neighbors of the minority class samples of the device operating parameters to determine a sample generation region; the generated sample region is interpolated to generate synthetic samples; and the synthetic samples are inversely mapped back to the original feature space using an inverse mapping function to generate augmented samples. In particular, this embodiment of the present invention performs data augmentation on the minority class samples.

[0162] In a specific implementation, by using a kernel function-based method to calculate the similarity between samples, it is possible to capture complex nonlinear relationships during data expansion, more accurately simulate the real data distribution, and thus determine the samples to be expanded. The kernel function type and its parameters are set. In one embodiment, the kernel function is set to a Gaussian kernel function, and the calculation method is expressed as:

[0163]

[0164] Where, and Respectively data samples and the data samples; is the width parameter of the Gaussian kernel function; Represents the distance metric function in the kernel space after Gaussian kernel transformation.

[0165] In this embodiment, the calculation method of the width parameter of the kernel function is expressed as:

[0166]

[0167] Where, The number of samples to interpolate for the current batch.

[0168] The original data is mapped to the high-dimensional kernel space through the kernel function. The mapping method is expressed as:

[0169]

[0170] Where, is the mapping function, is the original data point; is the center point of the kernel function; is the mapping ratio.

[0171] In one embodiment, the mapping ratio is calculated as follows:

[0172]

[0173] Furthermore, the embodiment of the present invention also determines the area where synthetic samples need to be generated based on the number of sample categories of the target category sample, such as the majority class samples in the k nearest neighbors, so as to determine the samples in the area as the samples to be expanded. In the kernel space, for each minority class sample, the majority class samples in its k nearest neighbors are identified to determine the area where synthetic samples need to be generated. samples , the way to calculate the boundary distance is expressed as:

[0174]

[0175] Where, and They are samples x in the kernel space i and x j The value of the kth feature; w k is the feature weight; is a weighted distance metric used to identify boundary samples.

[0176] Furthermore, the feature weights The calculation method is expressed as:

[0177]

[0178] in, It is The standard deviation of the features, The calculation method is:

[0179]

[0180] Where, is the mean of the kth feature.

[0181] Traditional data augmentation methods, such as the SMOTE algorithm, usually generate new synthetic samples by relying on simple interpolation between minority class samples. They cannot effectively handle complex or nonlinear data distributions, limiting the model's ability to handle complex data scenarios. In an embodiment of the present invention, a kernel function-based method is used to calculate the similarity between samples. On the premise of capturing complex nonlinear relationships, the SMOTE algorithm based on inversion-guided learning is also used for sample generation. By determining the samples to be augmented in the kernel space and inversely mapping the data from the high-dimensional kernel space to the feature space of the samples after original feature extraction, the useful information of the generated samples in the high-dimensional space can be retained, while ensuring that these samples are still meaningful in the original feature space, which helps the model learn a more natural data distribution.

[0182] Among them, the traditional SMOTE algorithm operates directly in the original feature space and obtains synthetic samples through linear interpolation between minority class samples. In the kernel method, the embodiment of the present invention effectively avoids the complexity of high-dimensional calculations through kernel function calculation. The kernel technique can linearize nonlinearly separable data in high-dimensional space, so that after the samples generated in the high-dimensional space are returned to the original space, different categories can be more clearly divided, thereby improving the performance of the classifier. The embodiment of the present invention also performs inverse mapping on the generated data back to the original feature space to obtain expanded samples. It can retain the useful information of the samples generated in the high-dimensional space while ensuring that these samples are still meaningful in the original feature space, which helps the model learn a more natural data distribution. In some complex data structures, kernel function mapping can reveal the potential structure of the data. The new samples after inverse mapping can carry this structural information, which promotes the model to learn more complex patterns.

[0183] In the embodiment of the present invention, the samples to be expanded are interpolated to generate synthetic samples. The specific steps are as follows:

[0184] For each minority class sample point in the kernel space, a new sample is generated by interpolation based on its k nearest minority class neighbors in the kernel space. The generation method is expressed as:

[0185]

[0186] Where, and is the minority class sample point in the selected kernel space; are randomly generated interpolation coefficients; is the new sample point generated.

[0187] Furthermore, the calculation method of the interpolation coefficient is expressed as:

[0188]

[0189] Where, is the interpolation ratio calculated based on the sample distance, Randomly generated The value between .

[0190] In one embodiment, the interpolation ratio is calculated as follows:

[0191]

[0192] Where, is the L2 norm.

[0193] Furthermore, an inverse mapping function is used to inversely map the synthesized samples back to the original feature space to generate expanded samples. The synthesized samples generated in the kernel space are inversely mapped back to the original feature space. The inverse mapping to the original space is expressed as:

[0194]

[0195] Where, It is the data point that is inversely mapped back to the original feature space; Represents the inverse mapping function from kernel space to original space.

[0196] Furthermore, the inverse mapping function The calculation method is expressed as:

[0197]

[0198] Where, The input sample is When , the mapping function The inverse transformation function of is the weight of the inversion map; is the number of features of the sample.

[0199] Furthermore, the weight of the inversion map is calculated as follows:

[0200]

[0201] Step S12: Merge the expanded samples and the equipment operation parameter samples to construct a training sample set.

[0202] The embodiment of the present invention uses the SMOTE algorithm based on inversion guided learning and uses the Gaussian kernel function to calculate the similarity between samples and generate new samples. This allows the capture of complex nonlinear relationships during the data expansion process and more accurately simulates the real data distribution, solving the problem that traditional data expansion technology cannot effectively handle complex data distribution.

[0203] Furthermore, based on the above method embodiment, the embodiment of the present invention also provides an equipment comprehensive efficiency evaluation device based on industrial interconnection, Figure 4 FIG. 1 shows a schematic diagram of a device for evaluating comprehensive efficiency of equipment based on industrial interconnection according to an embodiment of the present invention. Figure 4 As shown, the device includes: a parameter acquisition module 100, which is used to obtain equipment operating parameters of industrial equipment; a feature extraction module 200, which is used to extract target features in the equipment operating parameters through a pre-built feature extraction model; wherein, the neurons of the feature extraction model are assigned pressure values ​​based on the layer depth, and the neural network parameters are optimized by simulating the internal stress of the earth's crust; the layer depth is used to characterize the number of layers of neurons; a data processing module 300, which is used to input the target features into a preset autoencoder, adjust the network layer connection mode of the autoencoder based on the data flow change of the target features, and capture the key low-dimensional representation of the target features; an execution module 400, which is used to input the key low-dimensional representation into a preset extreme learning machine for preliminary prediction, and adjust the classification threshold of the extreme learning machine based on the result of the preliminary prediction; an output module 500, which is used to classify the key low-dimensional representation based on the adjusted classification threshold to determine the efficiency level of the industrial equipment.

[0204] An embodiment of the present invention provides an equipment comprehensive efficiency evaluation device based on industrial interconnection, 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 can be made to the corresponding content in the aforementioned method embodiment.

[0205] Furthermore, based on the above device embodiment, the embodiment of the present invention also provides another device for evaluating comprehensive efficiency of equipment based on industrial interconnection. Figure 5 FIG. 1 shows a schematic diagram of another device for evaluating comprehensive efficiency of equipment based on industrial interconnection according to an embodiment of the present invention. Figure 5 As shown, the above-mentioned data processing module 300 is also used to input the target features into a preset autoencoder and encode them through the multi-layer neural network of the autoencoder; dynamically reconstruct the topology of each layer of the neural network of the autoencoder through an adaptive activation function, and adjust the connection structure of the network layer of the autoencoder.

[0206] The above-mentioned data processing module 300 is also used to train a preset autoencoder using a preset training sample set, and calculate the reconstruction error corresponding to the training sample set; calculate the loss function of the autoencoder based on the reconstruction error and a preset regularization term; and update the parameters of the autoencoder using the loss function and the gradient descent method.

[0207] The execution module 400 is further configured to compare the preliminary prediction result with the prediction output to determine the error magnitude; and calculate a new classification threshold based on the error magnitude and the variance of the error magnitude to adjust the classification threshold of the extreme learning machine.

[0208] The execution module 400 is further configured to perform classification training on a preset extreme learning machine using a preset training sample set; perform a gradient accumulation operation during the training of the extreme learning machine, and update the parameters of the extreme learning machine based on the accumulated gradients.

[0209] The above-mentioned feature extraction module 200 is also used to obtain a preset training sample set and a feature extraction neural network; assign a pressure value to each layer of neurons based on the layer depth parameters of each layer of neurons in the feature extraction neural network; train the feature extraction neural network through the training sample set, and calculate the adjustment value of the pressure value based on the loss function value of the feature extraction neural network; optimize the neural network parameters of the feature extraction neural network based on the adjustment value until the feature extraction neural network meets the preset requirements, and construct a feature extraction model.

[0210] The feature extraction module 200 is further configured to calculate a pressure adjustment factor of a neuron based on the activity of the neuron; and update parameters of the feature extraction neural network based on the pressure adjustment factor and the adjustment value.

[0211] The device also includes a construction module 600 for obtaining pre-collected equipment operating parameter samples; mapping the equipment operating parameter samples to a high-dimensional kernel space, and performing sample expansion on the equipment operating parameter samples based on the sample categories of the equipment operating parameter samples to generate expanded samples; and merging the expanded samples and the equipment operating parameter samples to construct a training sample set.

[0212] The above-mentioned construction module 600 is also used to map the device operating parameter samples to a high-dimensional kernel space through a kernel function; in the high-dimensional kernel space, identify the majority class samples among the k nearest neighbors of the minority class samples of the device operating parameters to determine the generated sample area; interpolate the generated sample area to generate synthetic samples; and use an inverse mapping function to inversely map the synthetic samples back to the original feature space to generate expanded samples.

[0213] 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-mentioned Figures 1 to 3 The embodiment of the present invention also provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to execute the above Figures 1 to 3 Any of the steps of the method shown.

[0214] The embodiment of the present invention also provides a structural diagram of an electronic device, such as Figure 6As shown, it is a structural schematic diagram of the electronic device, wherein the electronic device comprises a processor 61 and a memory 60, the memory 60 stores computer executable instructions capable of being executed by the processor 61, and the processor 61 executes the computer executable instructions to implement the above-mentioned Figures 1 to 3 any of the methods shown.

[0215] In Figure 6 In the embodiment shown, the electronic device further comprises a bus 62 and a communication interface 63, wherein the processor 61, the communication interface 63 and the memory 60 are connected through the bus 62. The memory 60 can contain a high-speed random access memory (RAM) and can 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 can be wired or wireless), and the Internet, a wide area network, a local area network, a metropolitan area network, etc. can be used. The bus 62 can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc., and can also be an AMBA (Advanced Microcontroller Bus Architecture) bus, wherein AMBA defines three buses including an APB (Advanced Peripheral Bus) bus, an AHB (Advanced High-performance Bus) bus and an AXI (Advanced eXtensible Interface) bus. The bus 62 can 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 figure, but it does not mean that there is only one bus or one type of bus.

[0216] The processor 61 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by the hardware integrated logic circuit in the processor 61 or by instructions in the form of software. The above processor 61 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the embodiment of the present invention can be directly embodied as being executed by a hardware decoding processor, or can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory, and the processor 61 reads the information in the memory and completes the above-mentioned steps in combination with its hardware. Figures 1 to 3 Any of the methods shown.

[0217] The computer program product of a method and apparatus for evaluating comprehensive equipment efficiency based on industrial interconnection provided by embodiments of the present invention includes a computer-readable storage medium storing program code. The program code includes instructions for executing the methods described in the preceding method embodiments. For specific implementations, reference can be made to the method embodiments and are not described in detail here. Those skilled in the art will clearly understand that, for ease and brevity of description, the specific operating procedures of the system described above can refer to the corresponding procedures in the preceding method embodiments and are not described in detail here.

[0218] In addition, in the description of the embodiments of the present invention, unless otherwise explicitly specified or limited, the terms "installed," "connected," and "connected" should be understood broadly. For example, they can refer to fixed, detachable, or integral connections; mechanical or electrical connections; direct connections, indirect connections through an intermediate medium, or internal connections between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on the specific circumstances. If the functions described are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product stored in a storage medium and including instructions for enabling a computer device (such as a personal computer, server, or network device) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage media include various media capable of storing program code, such as USB flash drives, removable hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0219] In the description of the present invention, it should be noted that terms such as center, up, down, left, right, vertical, horizontal, inside, and outside, indicating positions or relationships, are based on the positions or relationships shown in the accompanying drawings and are intended solely to facilitate the description of the present invention and simplify the description. They are not intended to indicate or imply that the devices or components referred to must have a specific orientation, be constructed, or operate in a specific manner. Therefore, they should not be construed as limitations on the present invention. Furthermore, the terms "first," "second," and "third" are used for descriptive purposes only and should not be construed as indicating or implying relative importance.

[0220] Finally, it should be noted that the above embodiments are only specific implementation methods of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that any person skilled in the art can modify or easily conceive of changes to the technical solutions described in the above embodiments within the technical scope disclosed by the present invention, or replace some of the technical features therein with equivalents. Such modifications, changes or replacements do not deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection of the claims.

Claims

1. A method for evaluating comprehensive equipment efficiency based on industrial interconnection, characterized in that: The method comprises: Obtain equipment operating parameters of industrial equipment; Extracting target features from the device operating parameters using a pre-built feature extraction model; wherein neurons in the feature extraction model are assigned pressure values ​​based on layer depth, and neural network parameters are optimized by simulating stress within the earth's crust; wherein the layer depth is used to represent the number of layers of neurons; Inputting the target feature into a preset autoencoder, adjusting the network layer connection mode of the autoencoder based on the data flow change of the target feature, and capturing the key low-dimensional representation of the target feature; Inputting the key low-dimensional representation into a preset extreme learning machine for preliminary prediction, and adjusting the classification threshold of the extreme learning machine based on the result of the preliminary prediction; classifying the key low-dimensional representation based on an adjusted classification threshold to determine an efficiency level of the industrial equipment; The method for constructing the feature extraction model includes: Obtain a preset training sample set and feature extraction neural network; Based on the layer depth parameter of each layer of neurons in the feature extraction neural network, a pressure value is assigned to each layer of neurons; Training the feature extraction neural network using the training sample set, and calculating an adjustment value of the pressure value based on a loss function value of the feature extraction neural network; Optimizing the neural network parameters of the feature extraction neural network based on the adjustment value until the feature extraction neural network meets the preset requirements, thereby constructing a feature extraction model; The step of calculating the adjustment value of the pressure value based on the loss function value of the feature extraction neural network includes: in, is the learning rate of the neural network, For the The adjustment value of the pressure value of the layer neural network represents the pressure adjustment increment. is the gradient of the loss function; The step of optimizing the neural network parameters of the feature extraction neural network based on the adjustment value comprises: in, Update the symbols for the parameters, is the proportional factor adjusted according to pressure, represents element-wise product, For the The weights of the neurons in the layer, For the The bias of the neurons in the layer, For the The gradient of the weights of the neurons in the layer, For the The gradient of the bias of the layer neurons; It is Tier Stress adjustment factor for each neuron; The method further comprises: Calculating a stress adjustment factor for the neuron based on the activity of the neuron; updating parameters of the feature extraction neural network based on the pressure adjustment factor and the adjustment value; The pressure adjustment factor is determined by the following formula: Where, is the regularization coefficient of the neuron, It is The total number of neurons in the layer, No. Tier The activity of neurons; Where, It is Tier The net input to each neuron; The step of adjusting the classification threshold of the extreme learning machine based on the result of the preliminary prediction includes: Comparing the preliminary prediction result with the predicted output to determine the error size; Calculating a new classification threshold based on the error magnitude and the variance of the error magnitude to adjust the classification threshold of the extreme learning machine; The step of inputting the target feature into a preset autoencoder and adjusting the network layer connection mode of the autoencoder based on the data flow change of the target feature includes: Inputting the target feature into a preset autoencoder and encoding it through a multi-layer neural network of the autoencoder; The topology of each layer of the neural network of the autoencoder is dynamically reconstructed through an adaptive activation function to adjust the connection structure of the network layer of the autoencoder.

2. The method according to claim 1, characterized in that The method further comprises: Training a preset autoencoder using a preset training sample set, and calculating a reconstruction error corresponding to the training sample set; Calculating a loss function of the autoencoder based on the reconstruction error and a preset regularization term; The parameters of the autoencoder are updated using the loss function and the gradient descent method.

3. The method according to claim 1, characterized in that The method further comprises: Perform classification training on the preset extreme learning machine through the preset training sample set; A gradient accumulation operation is performed during the training process of the extreme learning machine, and the parameters of the extreme learning machine are updated based on the accumulated gradients.

4. The method according to claim 1, wherein The method further comprises: Obtain pre-collected equipment operating parameter samples; Mapping the device operation parameter samples to a high-dimensional kernel space, and performing sample expansion on the device operation parameter samples based on sample categories of the device operation parameter samples to generate expanded samples; The expanded samples and the equipment operation parameter samples are combined to construct a training sample set.

5. The method according to claim 4, characterized in that The steps of mapping the device operation parameter samples to a high-dimensional kernel space and performing sample expansion on the device operation parameter samples based on sample categories of the device operation parameter samples to generate expanded samples include: Mapping the device operating parameter samples to a high-dimensional kernel space through a kernel function; In the high-dimensional kernel space, identifying majority class samples among the k nearest neighbors of the minority class samples of the device operating parameters, and determining a sample generation region; interpolating the generated sample region to generate a synthetic sample; An inverse mapping function is used to inversely map the synthesized samples back to the original feature space to generate expanded samples.

6. A comprehensive equipment efficiency evaluation device based on industrial interconnection, characterized in that: The device comprises: Parameter acquisition module, used to obtain equipment operating parameters of industrial equipment; a feature extraction module, configured to extract target features from the device operating parameters using a pre-built feature extraction model; wherein neurons in the feature extraction model are assigned pressure values ​​based on layer depth, and the neural network parameters are optimized by simulating stress within the earth's crust; wherein the layer depth is used to represent the number of layers of neurons; a data processing module, configured to input the target feature into a preset autoencoder, adjust the network layer connection mode of the autoencoder based on the data flow change of the target feature, and capture the key low-dimensional representation of the target feature; an execution module, configured to input the key low-dimensional representation into a preset extreme learning machine for preliminary prediction, and adjust a classification threshold of the extreme learning machine based on a result of the preliminary prediction; an output module, configured to classify the key low-dimensional representation based on an adjusted classification threshold to determine an efficiency level of the industrial equipment; The feature extraction module is further configured to obtain a preset training sample set and a feature extraction neural network; assign a pressure value to each layer of neurons in the feature extraction neural network based on a layer depth parameter of each layer of neurons; train the feature extraction neural network using the training sample set, and calculate an adjustment value of the pressure value based on a loss function value of the feature extraction neural network; optimize the neural network parameters of the feature extraction neural network based on the adjustment value until the feature extraction neural network meets preset requirements, thereby constructing a feature extraction model; The step of calculating the adjustment value of the pressure value based on the loss function value of the feature extraction neural network includes: in, is the learning rate of the neural network, For the The adjustment value of the pressure value of the layer neural network represents the pressure adjustment increment. is the gradient of the loss function; The step of optimizing the neural network parameters of the feature extraction neural network based on the adjustment value comprises: in, Update the symbols for the parameters, is the proportional factor adjusted according to pressure, represents element-wise product, For the The weights of the neurons in the layer, For the The bias of the neurons in the layer, For the The gradient of the weights of the neurons in the layer, For the The gradient of the bias of the layer neurons; It is Tier Stress adjustment factor for each neuron; The feature extraction module is further configured to: calculate a pressure adjustment factor of the neuron based on the activity of the neuron; and update parameters of the feature extraction neural network based on the pressure adjustment factor and the adjustment value; The pressure adjustment factor is determined by the following formula: Where, is the regularization coefficient of the neuron, It is The total number of neurons in the layer, No. Tier The activity of neurons; Where, It is Tier The net input to each neuron; The execution module is further configured to: compare the preliminary prediction result with the prediction output to determine an error magnitude; and calculate a new classification threshold based on the error magnitude and the variance of the error magnitude to adjust the classification threshold of the extreme learning machine; The data processing module is also used to: input the target feature into a preset autoencoder and encode it through the multi-layer neural network of the autoencoder; dynamically reconstruct the topology of each layer of the neural network of the autoencoder through an adaptive activation function, and adjust the connection structure of the network layer of the autoencoder.

Images