Fuzzy dynamic reasoning method, device and equipment for industrial time series

Abstract

The embodiments of the present application provide a fuzzy dynamic reasoning method, apparatus, and device for industrial time series. These methods are applicable to the fields of artificial intelligence and industrial control. The method comprises: preprocessing pre-collected industrial time series data of target industrial equipment to obtain industrial time series sample data; inputting the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. The above method solves the problem of difficulty in handling uncertainty in industrial time series data and the problem of insufficient computing resources in industrial edge scenarios, thereby improving prediction accuracy.

Description

Technical Field

[0001] The present application relates to the fields of artificial intelligence and industrial control, and in particular to a fuzzy dynamic reasoning method, device and equipment for industrial time series. Background Art

[0002] In complex industrial scenarios, equipment operation is affected by both inherent long-term degradation and external fluctuations such as load, temperature, and sensor errors. Consequently, the collected time series data inevitably contains significant uncertainty. Data-driven deep learning methods, which rely on completely deterministic representations, are unable to handle this uncertainty in the data, severely hindering their deployment in complex industrial scenarios.

[0003] Industrial time series data is typically long and high-dimensional, requiring complex deep neural networks to perform significantly more computationally. However, computing resources are limited in edge industrial scenarios, making it difficult for complex neural networks to meet real-time requirements. Existing training methods for time series dynamic networks ignore the contribution of samples of varying difficulty to different output predictors. Early exit behavior during the inference phase is neglected, leading to a gap between training and testing, impacting the performance of dynamic models.

[0004] In summary, how to deal with the uncertainty of time series data in industrial scenarios is a technical problem that needs to be solved urgently in this field. Summary of the Invention

[0005] The embodiments of the present application provide a fuzzy dynamic reasoning method, apparatus, and device for industrial time series, to solve the problem of how to deal with the uncertainty of time series data in industrial scenarios.

[0006] In a first aspect, an embodiment of the present application provides a fuzzy dynamic reasoning method for industrial time series, including:

[0007] Preprocess the pre-collected industrial time series data of the target industrial equipment to obtain industrial time series sample data;

[0008] The industrial time series sample data is input into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. The fuzzy dynamic network includes a fuzzy logic representation layer, a sample information fusion layer and a time series multi-outlet dynamic network layer. The multi-outlet dynamic network layer includes multiple sub-networks.

[0009] In a possible implementation, inputting the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment includes:

[0010] Performing fuzzy logic processing on the industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information;

[0011] Splicing the fuzzy membership information and the industrial time series sample data to obtain splicing information;

[0012] Processing the spliced ​​information through the sample information fusion layer to obtain mixed features;

[0013] The hybrid features are inferred through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment.

[0014] In a possible implementation, the reasoning on the hybrid feature through the time-series multi-exit dynamic network layer to obtain the remaining life of the target industrial equipment includes:

[0015] Step a: inputting the mixed features into the first-level sub-network, obtaining the first-level deep features through feature extraction, and generating the first-level prediction value and the first-level confidence;

[0016] Step b: If the first-level confidence is less than a preset threshold, the first-level depth feature is input into the second-level subnetwork to obtain a second-level depth feature, the first-level depth feature is weighted according to a preset attenuation coefficient and then superimposed with the second-level depth feature to generate a second-level fusion feature, and a second-level prediction value and a second-level confidence are generated based on the second-level fusion feature;

[0017] Step c: repeat step b until a prediction value corresponding to a subnetwork with a confidence level greater than the preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

[0018] In a possible implementation, the processing of the splicing information by the sample information fusion layer to obtain mixed features includes:

[0019] The splicing information is sequentially subjected to mean trend extraction, peak trend extraction, and cross-channel dependency extraction in a time dimension, and the extracted features are fused to obtain the hybrid feature.

[0020] In one possible implementation, each subnetwork includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer of the first-level subnetwork is a one-dimensional convolutional neural network, and the feature extraction layers of other subnetworks except the first-level subnetwork are Transformer encoders.

[0021] In one possible implementation, the method further includes:

[0022] A preset initial fuzzy dynamic network is trained based on pre-acquired training data to obtain the fuzzy dynamic network, wherein the training data includes a plurality of industrial time series data.

[0023] In a possible implementation, the training of a preset initial fuzzy dynamic network based on pre-acquired training data to obtain the fuzzy dynamic network includes:

[0024] According to the training data, the network parameters of the initial fuzzy dynamic network are iteratively updated by weighted superposition of the outputs of the shallow sub-network and the deep sub-network based on a preset gradient redistribution rule until a preset number of iterations is reached, thereby obtaining the fuzzy dynamic network;

[0025] Among them, the gradient redistribution rule includes: blocking the gradient transfer of shallow sub-networks during back propagation, and for the shared parameters of the n-th level sub-network, the gradient of the shared parameters is jointly calculated by the prediction loss of the n-th level to the last level sub-network, and the gradient value is weighted averaged according to the number of sub-networks participating in the calculation, where n is an integer greater than or equal to 1.

[0026] In a second aspect, an embodiment of the present application provides a fuzzy dynamic reasoning device for industrial time series, comprising:

[0027] The preprocessing module preprocesses the industrial time series data of the target industrial equipment collected in advance to obtain industrial time series sample data;

[0028] The prediction module is used to input the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. The fuzzy dynamic network includes a fuzzy logic representation layer, a sample information fusion layer and a time series multi-outlet dynamic network layer. The multi-outlet dynamic network layer includes multiple sub-networks.

[0029] In a possible implementation, the prediction module is configured to:

[0030] Performing fuzzy logic processing on the industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information;

[0031] Splicing the fuzzy membership information and the industrial time series sample data to obtain splicing information;

[0032] Processing the spliced ​​information through the sample information fusion layer to obtain mixed features;

[0033] The hybrid features are inferred through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment.

[0034] In a possible implementation, the prediction module infers the hybrid features through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment, specifically including:

[0035] Step a: inputting the mixed features into the first-level sub-network, obtaining the first-level deep features through feature extraction, and generating the first-level prediction value and the first-level confidence;

[0036] Step b: If the first-level confidence is less than a preset threshold, the first-level depth feature is input into the second-level subnetwork to obtain a second-level depth feature, the first-level depth feature is weighted according to a preset attenuation coefficient and then superimposed with the second-level depth feature to generate a second-level fusion feature, and a second-level prediction value and a second-level confidence are generated based on the second-level fusion feature;

[0037] Step c: repeat step b until a prediction value corresponding to a subnetwork with a confidence level greater than the preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

[0038] In a possible implementation, the prediction module processes the splicing information through the sample information fusion layer to obtain mixed features, specifically including:

[0039] The splicing information is sequentially subjected to mean trend extraction, peak trend extraction, and cross-channel dependency extraction in a time dimension, and the extracted features are fused to obtain the hybrid feature.

[0040] In one possible implementation, each subnetwork includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer of the first-level subnetwork is a one-dimensional convolutional neural network, and the feature extraction layers of other subnetworks except the first-level subnetwork are Transformer encoders.

[0041] In a possible implementation, the device further includes:

[0042] The training module is used to train a preset initial fuzzy dynamic network based on pre-acquired training data to obtain the fuzzy dynamic network, wherein the training data includes a plurality of industrial time series data.

[0043] In a possible implementation, the training module is specifically configured to:

[0044] According to the training data, the network parameters of the initial fuzzy dynamic network are iteratively updated by weighted superposition of the outputs of the shallow sub-network and the deep sub-network based on a preset gradient redistribution rule until a preset number of iterations is reached, thereby obtaining the fuzzy dynamic network;

[0045] Among them, the gradient redistribution rule includes: blocking the gradient transfer of shallow sub-networks during back propagation, and for the shared parameters of the n-th level sub-network, the gradient of the shared parameters is jointly calculated by the prediction loss of the n-th level to the last level sub-network, and the gradient value is weighted averaged according to the number of sub-networks participating in the calculation, where n is an integer greater than or equal to 1.

[0046] In a third aspect, an embodiment of the present application provides an electronic device, comprising: a memory, a processor;

[0047] The memory stores computer-executable instructions;

[0048] The processor executes the computer-executable instructions stored in the memory, so that the processor executes the above first aspect and / or various possible implementations of the first aspect.

[0049] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the first aspect above and / or various possible implementation methods of the first aspect.

[0050] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program, which, when executed by a processor, implements the above first aspect and / or various possible implementation methods of the first aspect.

[0051] The embodiments of the present application provide a fuzzy dynamic reasoning method, apparatus, and device for industrial time series. Pre-collected industrial time series data for target industrial equipment is preprocessed to obtain industrial time series sample data. This industrial time series sample data is then input into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. This method addresses the difficulty in handling uncertainty in industrial time series data and the issue of insufficient computing resources in industrial edge scenarios, thereby improving prediction accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0053] Figure 1 Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 1 ;

[0054] Figure 2 This is a schematic diagram of the fuzzy dynamic network architecture;

[0055] Figure 3 Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 2 ;

[0056] Figure 4 Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 3 ;

[0057] Figure 5 Schematic diagram of back-propagation gradient calculation;

[0058] Figure 6 A schematic diagram of the structure of the fuzzy dynamic reasoning device for industrial time series provided in this application;

[0059] Figure 7 This is a schematic diagram of the structure of the electronic device provided in this application.

[0060] The above drawings illustrate specific embodiments of the present application, which will be described in more detail below. These drawings and the textual description are not intended to limit the scope of the present application in any way, but rather to illustrate the concepts of the present application to those skilled in the art by reference to specific embodiments. DETAILED DESCRIPTION

[0061] Exemplary embodiments are described in detail herein, with examples illustrated in the accompanying drawings. When the following description refers to the drawings, identical numerals in different drawings represent identical or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all embodiments consistent with this application.

[0062] In complex industrial scenarios, equipment operation is affected by both inherent long-term degradation factors and external fluctuations such as load, temperature, and sensor errors. Consequently, the collected time series data inevitably contains significant uncertainty. Data-driven deep learning methods rely on completely deterministic representations and are unable to handle this uncertainty, severely hindering their deployment in complex industrial scenarios. Industrial time series data is typically long and high-dimensional, and the use of complex deep neural networks significantly increases the computational workload. However, computing resources are limited in edge industrial scenarios, making it difficult for complex neural networks to meet real-time requirements. Existing training methods for time series dynamic networks ignore the contribution of samples of varying difficulty to different output predictors. Early exit behavior during the inference phase is neglected, leading to a gap between training and testing, impacting the performance of dynamic models. In summary, handling the uncertainty of time series data in industrial scenarios is a pressing technical challenge in this field.

[0063] In response to the above problems, the present application provides a fuzzy dynamic reasoning method, device and equipment for industrial time series, which solves the problem that the uncertainty in industrial time series data is difficult to handle. Specifically, the existing collected time series inevitably have a large amount of uncertainty, and the data-driven deep learning method is a completely deterministic representation, which cannot handle the uncertainty in the data, seriously hindering its deployment in complex industrial scenarios. Industrial time series are usually long time series and high-dimensional. The use of complex deep neural networks such as Transformer will lead to a significant increase in the amount of calculation. However, the computing resources in edge industrial scenarios are limited, and complex neural networks are difficult to meet real-time requirements. In addition, the existing training methods for time series dynamic networks ignore the contribution of samples of different difficulty levels to different export predictors. The early exit behavior in the inference stage is ignored, resulting in a gap between training and testing, affecting the performance of the dynamic model. Taking these issues into consideration, we studied whether it is possible to introduce fuzzy learning methods into deep learning models to solve the problems of insufficient computing resources and uncertainty in industrial edge scenarios, and to obtain fuzzy feature information in fuzzy membership, supplement sample information, make sample information representation more comprehensive, and solve the problem of uncertainty in industrial time series data being difficult to handle. Then, by jointly training regression predictors, weighting the outputs of shallow sub-networks with the outputs of deep sub-networks, and redistributing sub-network gradients, we can bridge the gap between time series multi-export network training and reasoning. Based on this, the solution of this application is proposed.

[0064] The following specific embodiments describe in detail the technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.

[0065] Figure 1 Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0066] S101: Preprocess the pre-collected industrial time series data of target industrial equipment to obtain industrial time series sample data.

[0067] In this step, in order to accurately predict the remaining life of the target industrial equipment, the industrial time series data of the target industrial equipment can be collected in advance and the data can be preprocessed.

[0068] For example, identify the type of industrial equipment for which remaining life prediction is needed, such as mechanical parts and electronic components. Select appropriate sensors and data acquisition equipment based on the equipment type and performance indicators. Ensure the accuracy and stability of the data acquisition equipment to obtain high-quality time series data. Continuously collect time series data during normal equipment operation. For example, record equipment operating hours, load conditions, and environmental parameters for subsequent analysis.

[0069] Data preprocessing can include:

[0070] 1. Data cleaning:

[0071] Identify and remove outliers: Check the data for outliers that may significantly deviate from the normal range. These values ​​may be caused by sensor failure, data transmission errors, etc. For outliers, you can choose to delete, replace, or smooth them.

[0072] Filling missing values: For missing data points, interpolation methods (such as linear interpolation, spline interpolation, etc.) can be used to fill them, or they can be estimated based on the operating status and trends of the equipment.

[0073] 2. Data transformation:

[0074] Normalization: Convert data of different dimensions to the same scale to improve the convergence speed and prediction accuracy of the model.

[0075] Feature extraction: Extract meaningful features from raw data, such as statistics such as mean, standard deviation, kurtosis, skewness, and frequency domain features (such as spectrum analysis).

[0076] 3. Data division:

[0077] The dataset is divided into training set, validation set, and test set. The training set is used to build and train the prediction model, the validation set is used to adjust model parameters and optimize model performance, and the test set is used to evaluate the final prediction accuracy of the model.

[0078] S102: Inputting industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment.

[0079] In this step, after obtaining the industrial time series sample data, the data is input into the pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment.

[0080] Specifically, during operation, equipment is affected by both its own inherent long-term degradation factors and external fluctuations such as load, temperature, and sensor errors. Consequently, the collected time series inevitably contain a large amount of uncertainty. Therefore, the fuzzy dynamic network configuration consists of a fuzzy logic representation layer, a sample information fusion layer, and a time series multi-outlet dynamic network layer. The multi-outlet dynamic network layer includes multiple sub-networks.

[0081] For example, Figure 2 This is a schematic diagram of the fuzzy dynamic network architecture, as shown in Figure 2 As shown, the fuzzy logic representation layer is composed of different fuzzy membership functions. The time-series multi-output dynamic neural network layer consists of four subnetworks, each of which includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer used by the first-level subnetwork is a one-dimensional convolutional neural network, while the remaining subnetworks use Transformer encoders. First, sample data enters the fuzzy logic representation layer to obtain fuzzy membership information. Then, the fuzzy membership information is concatenated with the original sample and enters the sample information fusion layer. Through adaptive information fusion, hybrid feature information containing fuzzy features is obtained, enhancing the deep neural network's ability to handle uncertainty. Finally, the hybrid feature information enters the time-series multi-output dynamic network for inference. The network depth is determined based on the sample difficulty, and the predicted value is obtained.

[0082] The fuzzy dynamic reasoning method for industrial time series provided in this application's embodiment preprocesses pre-collected industrial time series data for target industrial equipment to obtain industrial time series sample data. This industrial time series sample data is then fed into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. This method addresses the difficulty in handling uncertainty in industrial time series data and the lack of computing resources in industrial edge scenarios, thereby improving prediction accuracy.

[0083] Figure 3 Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 2 ,like Figure 3 As shown, based on the above embodiment, step S102 specifically includes:

[0084] S301: Perform fuzzy logic processing on industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information.

[0085] In this step, fuzzy feature extraction technology is used to obtain fuzzy feature information from fuzzy membership, and the fuzzy feature representation is combined with the original sample data to make the information representation more comprehensive and solve the problem of uncertainty in industrial time series data that is difficult to handle.

[0086] For example, Figure 2As shown, for each sensor data in the industrial time series sample data, the fuzzy logic representation layer processes the sensor data through the pre-set fuzzy membership function corresponding to the sensor to obtain the result corresponding to the sensor, and combines the results corresponding to each sensor to obtain the fuzzy membership information corresponding to the industrial time series sample data.

[0087] First, input the industrial time series sample data Dimension-by-dimensional decomposition, where L represents the time length of the sequence and K represents the dimension of K sensor data. Each sensor feature dimension For each sensor dimension , and set up independent fuzzy membership functions for them. Each dimension fuzzy membership function calculates the degree to which the input data belongs to the fuzzy set of the current dimension. The input data of that dimension is mapped to the fuzzy membership interval [0,1].

[0088] Fuzzy logic representation layer is based on fuzzy mean and fuzzy variance The fuzzy mean fuzzifies the core trend of the sensor dimension data, reflecting the sensor's typical or expected value. The fuzzy variance quantifies the dispersion of the data around the fuzzy mean, thus reflecting the uncertainty of the data. The fuzzy mean and fuzzy variance can effectively capture the uncertainty characteristics of sensor measurements, thereby more accurately reflecting the operating status of industrial equipment. The Gaussian membership function is as follows:

[0089]

[0090] The results of the membership functions of each sensor dimension are combined into fuzzy features , which can be described as:

[0091]

[0092] Fuzzy features , that is, the final expression of fuzzy membership information is:

[0093]

[0094] The fuzzy membership degree obtained by the Gaussian membership function can effectively capture the uncertainty in the input data and provide effective fuzzy feature information for subsequent sample information fusion.

[0095] S302: Splicing the fuzzy membership information and the industrial time series sample data to obtain splicing information.

[0096] In this step, in order to obtain enhanced features, the fuzzy membership information and the industrial time series sample data can be spliced.

[0097] Specifically, the splicing can be performed along the feature dimension (i.e., the column direction) so that the feature vector at each time point contains both the original sensor data and the fuzzy membership information.

[0098] S303: Processing the spliced ​​information through the sample information fusion layer to obtain mixed features.

[0099] In this step, the splicing information input value sample information obtained in the above step is processed by the fusion layer to obtain a mixed feature containing fuzzy feature information.

[0100] Specifically, the splicing information is sequentially subjected to mean trend extraction, peak trend extraction, and cross-channel dependency extraction in the time dimension, and the extracted features are fused to obtain hybrid features.

[0101] For example, combined Figure 2 , the sample information fusion layer captures the splicing information The mean trend and peak trend of the time dimension are combined with the correlation information between channels to realize the splicing information. Feature extraction and information fusion. The sample information fusion layer uses average pooling to extract the average trend feature, which can be expressed as:

[0102]

[0103] Where K represents the window size.

[0104] The sample information fusion layer uses maximum pooling to capture significant local changes and emphasize local extreme values ​​in time series data. Its representation method is as follows:

[0105]

[0106] Where K represents the window size.

[0107] Based on the extracted temporal features, the sample information fusion layer further captures the correlation between sensor channels using one-dimensional convolution. The convolution kernel propagates along the time dimension and simultaneously fuses the data of all sensor channels to extract cross-channel dependencies. This can be expressed as:

[0108]

[0109] in, It is splicing information. and represents the weight and bias of the convolution kernel, and express dimension.

[0110] Finally, all feature information is concatenated along the feature dimension and further fused through the fully connected layer to obtain the mixed feature The final output can be described as:

[0111]

[0112] in Indicates a connection operation. represents a linear layer.

[0113] By further fusing sample and fuzzy membership information, it can provide strong support for handling uncertainty factors in subsequent industrial time series forecasting.

[0114] S304: Inferring the mixed features through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment.

[0115] In this step, in order to achieve accurate prediction of the remaining life of industrial equipment, balance computing efficiency and model performance, and adapt to the resource constraints of edge industrial scenarios, the hybrid features can be adaptively inferred through a time-series multi-export dynamic network layer.

[0116] Among them, each subnetwork of the time-series multi-output dynamic network layer includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer of the first-level subnetwork is a one-dimensional convolutional neural network, and the feature extraction layers of other subnetworks except the first-level subnetwork are Transformer encoders.

[0117] Specifically, the mixed features are input into the first-level sub-network, the first-level deep features are obtained through feature extraction, and the first-level prediction value and the first-level confidence are generated. If the first-level confidence is less than the preset threshold, the first-level deep features are input into the second-level sub-network to obtain the second-level deep features. The first-level deep features are weighted according to the preset attenuation coefficient and superimposed with the second-level deep features to generate the second-level fusion features. The second-level prediction value and the second-level confidence are generated based on the second-level fusion features. The above operation is repeated until the prediction value corresponding to the sub-network with a confidence greater than the preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

[0118] The fuzzy dynamic reasoning method for industrial time series provided in the embodiments of this application performs fuzzy logic processing on industrial time series sample data through a fuzzy logic representation layer to obtain fuzzy membership information. This fuzzy membership information is then concatenated with the industrial time series sample data to obtain concatenated information. This concatenated information is then processed through a sample information fusion layer to obtain hybrid features. This hybrid feature is then inferred through a time series multi-export dynamic network layer to determine the remaining life of the target industrial equipment. This method improves the uncertainty data modeling capability, enables lightweight dynamic reasoning, and enhances multi-scale feature representation capabilities.

[0119] Figure 4Schematic diagram of the process of the fuzzy dynamic reasoning method for industrial time series provided in this application Figure 3 ,like Figure 4 As shown, based on the above embodiment, step S304 specifically includes:

[0120] S401: Input the mixed features into the first-level sub-network, obtain the first-level deep features through feature extraction, and generate the first-level prediction value and the first-level confidence.

[0121] S402: If the first-level confidence is less than the preset threshold, the first-level depth feature is input into the second-level sub-network to obtain the second-level depth feature, the first-level depth feature is weighted according to the preset attenuation coefficient and superimposed with the second-level depth feature to generate the second-level fusion feature, and the second-level prediction value and the second-level confidence are generated based on the second-level fusion feature.

[0122] S403: Repeat the above step S402 until a prediction value corresponding to a sub-network with a confidence level greater than a preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

[0123] After the mixed features are input into the time-series multi-output dynamic network layer, they are first processed by the first-level sub-network. The one-dimensional convolutional neural network of the first-level sub-network extracts the mixed features and outputs the first-level deep features. Specifically, the design of the one-dimensional convolutional neural network is that the convolution kernel slides along the time dimension to capture local time series patterns (such as short-term fluctuations in vibration signals).

[0124] The regression predictor 1 performs a fully connected mapping on the first-level deep features and outputs the first-level prediction value. The confidence predictor is used to calculate the confidence. Optionally, the confidence can be estimated based on the prediction residual or variance.

[0125] Then, the confidence of the first-level subnetwork is compared with the preset threshold. If the confidence of the first-level subnetwork is less than the preset threshold, it triggers the entry into the second subnetwork, inputs the first-level deep features into the second subnetwork for feature extraction to obtain the second-level deep features, and then weights the first-level deep features by the attenuation coefficient, superimposes the weighted first-level deep features and the second-level deep features, and obtains the second-level prediction value after prediction by the regression predictor 2. The confidence corresponding to the second-level subnetwork is calculated by logic, and the confidence of the second-level subnetwork is compared with the preset threshold. If the confidence of the second-level subnetwork is less than the preset threshold, it triggers the entry into the third-level subnetwork, and repeats the above steps until the obtained confidence is greater than the preset threshold. The predicted value of the subnetwork corresponding to the confidence is determined as the remaining life of the target industrial equipment.

[0126] The fuzzy dynamic reasoning method for industrial time series provided in the embodiment of the present application inputs mixed features into the first-level sub-network, obtains the first-level deep features through feature extraction, and generates the first-level prediction value and the first-level confidence. If the first-level confidence is less than the preset threshold, the first-level deep features are input into the second-level sub-network to obtain the second-level deep features. The first-level deep features are weighted according to the preset attenuation coefficient and superimposed with the second-level deep features to generate the second-level fusion features. The second-level prediction value and the second-level confidence are generated based on the second-level fusion features. The above operation is repeated until the prediction value corresponding to the sub-network with a confidence greater than the preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment. The above method achieves an adaptive precision-efficiency balance for the remaining life prediction of industrial equipment through the dynamic inference mechanism of multi-level sub-networks, combined with confidence threshold judgment and hierarchical feature fusion: the lightweight shallow sub-network (1D-CNN) quickly processes most simple samples and generates initial predictions. When the confidence is insufficient, the deep sub-network (Transformer) is triggered to perform weighted residual correction on the features, and local and global timing patterns are integrated step by step. Ultimately, while ensuring prediction accuracy, the computational complexity is significantly reduced. At the same time, fuzzy features and confidence quantification are used to effectively deal with the uncertainty of industrial data, providing a highly robust and low-latency life prediction solution for edge devices.

[0127] Based on the above embodiments, the method further includes training a preset initial fuzzy dynamic network based on pre-acquired training data to obtain a fuzzy dynamic network. The specific training process includes: according to the training data, by weighted superposition of outputs of the shallow sub-network and the deep sub-network, and iteratively updating the network parameters of the initial fuzzy dynamic network based on a preset gradient redistribution rule until a preset number of iterations is reached, thereby obtaining a fuzzy dynamic network;

[0128] Among them, the gradient redistribution rules include: blocking the gradient transfer of shallow sub-networks during backpropagation, and for the shared parameters of the n-th level sub-network, the gradient of the shared parameters is jointly calculated by the prediction loss of the n-th to the last level sub-network, and the gradient value is weighted averaged according to the number of sub-networks participating in the calculation, where n is an integer greater than or equal to 1.

[0129] Specifically, in traditional training methods for temporal dynamic networks, the model optimizes all outputs simultaneously during training. That is, given a batch of training samples, all regression predictors at all levels calculate losses and update parameters on the same batch of data. This can easily lead to a training-inference mismatch: during inference, if the input sample is "easy," the network often exits at the early shallow layers, while the deep classifiers almost only encounter "difficult" samples. However, during training, all regression predictors see all samples, causing the parameter distribution of the deep predictors to be inconsistent with that during actual inference.

[0130] Unlike traditional gradient boosting, which first trains the first predictor to complete convergence and then trains the next predictor after fixing the previous predictor, the fuzzy dynamic network provided by the embodiment of the present application does not adopt a completely sequential training method. Most of the feature extraction layers are shared between all predictors. If the shallow layers are fixed first and then the deep layers are trained, the parameter coordination will be poor; and separate stages will lead to a large number of repeated iterations, which is very expensive. Therefore, the outputs of all predictors are calculated uniformly, and then the losses of each predictor are summed up before the parameters are updated. The loss function L is:

[0131]

[0132] Where x is the input sample, RMSE is the root mean square error, Fn is the output of the n-th predictor, and y is the true value label.

[0133] During the training process, all predictors will process the complete training samples, which means that the deep predictors will be affected by a large number of "simple" training samples and cannot obtain effective gradients for updating, resulting in poor deep learning results. To solve this problem, output weighted paths are added between different sub-networks. Figure 2 As shown, the output of the second-level sub-network Transformer encoder 2 in the network The output of the first-level sub-network one-dimensional convolutional neural network Weighted . Obtained by weighting The predicted value is obtained by regression predictor 2. This can be understood as the second-level sub-network compensating for the prediction of the first-level sub-network based on the first-level sub-network to obtain a more accurate prediction output. Similarly, this model regards the output of the i-th sub-network of the network as: the sum of the outputs of all previous sub-networks plus the output of the current sub-network i:

[0134]

[0135] in It is a weighting coefficient used to weaken the previous accumulated output to prevent the shallow sub-network from causing excessive interference to the deep sub-network. Set to a constant less than 1, weakening the effective output of the front predictor and increasing the prediction difficulty of the deep predictor, thereby enhancing the training of the deep predictor parameters. It should be noted that during the training process The back-propagation link is disabled.

[0136] In addition, during training, if the losses of all classifiers are directly back-propagated to the shared sub-network, it may cause excessive accumulation of gradients in the shallow layers, which is not conducive to the stable update of the early sub-network. Figure 5 This is a schematic diagram of back propagation gradient calculation, as shown Figure 5 As shown in Figure 2, during back propagation, the gradient of each exit is weighted, and the specific formula is expressed as follows:

[0137]

[0138] in is the total gradient of the nth sub-network, represents the RMSE loss of the i-th export, It represents the gradient contribution of the i-th exit loss to the n-th sub-network, and N-n+1 is the number of exits from the n-th exit to the N-th exit (a total of N exits), which is used to weight the gradient of each exit.

[0139] For the nth sub-network, since it participates in all exports from n to N, its accumulated gradient is If this gradient accumulation is used directly, when N is large, the shallow sub-network may produce gradient explosion. Weighting factor The role of is to average all the exports that contribute to the sub-network so that its gradient magnitude remains within a controllable range.

[0140] The fuzzy dynamic reasoning method for industrial time series provided in the embodiments of the present application bridges the gap between time series multi-export network training and reasoning by jointly training regression predictors, weighting the outputs of shallow sub-networks with the outputs of deep sub-networks, and redistributing sub-network gradients.

[0141] Figure 6 The structural diagram of the fuzzy dynamic reasoning device for industrial time series provided by this application is as follows: Figure 6 As shown, the fuzzy dynamic reasoning device 600 for industrial time series provided in this embodiment includes:

[0142] The preprocessing module 601 is used to preprocess the pre-collected industrial time series data of the target industrial equipment to obtain industrial time series sample data.

[0143] The prediction module 602 is used to input industrial time series sample data into the pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment. The fuzzy dynamic network includes a fuzzy logic representation layer, a sample information fusion layer and a time series multi-outlet dynamic network layer. The multi-outlet dynamic network layer includes multiple sub-networks.

[0144] In one possible implementation, the prediction module 602 is specifically configured to:

[0145] Perform fuzzy logic processing on industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information;

[0146] The fuzzy membership information and the industrial time series sample data are spliced ​​to obtain spliced ​​information;

[0147] The spliced ​​information is processed through the sample information fusion layer to obtain mixed features;

[0148] The remaining life of the target industrial equipment is obtained by reasoning on the mixed features through a time-series multi-export dynamic network layer.

[0149] In one possible implementation, the prediction module 602 infers the mixed features through a time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment, specifically including:

[0150] Step a: Input the mixed features into the first-level sub-network, obtain the first-level deep features through feature extraction, and generate the first-level prediction value and the first-level confidence;

[0151] Step b: If the first-level confidence is less than a preset threshold, the first-level depth feature is input into the second-level sub-network to obtain a second-level depth feature, the first-level depth feature is weighted according to a preset attenuation coefficient and superimposed with the second-level depth feature to generate a second-level fusion feature, and a second-level prediction value and a second-level confidence are generated based on the second-level fusion feature;

[0152] Step c: Repeat step b until a prediction value corresponding to a subnetwork with a confidence level greater than a preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

[0153] In one possible implementation, the prediction module 602 processes the splicing information through a sample information fusion layer to obtain a mixed feature, specifically including:

[0154] The splicing information is sequentially subjected to mean trend extraction, peak trend extraction and cross-channel dependency extraction in the time dimension, and the extracted features are fused to obtain hybrid features.

[0155] In one possible implementation, each subnetwork includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer of the first-level subnetwork is a one-dimensional convolutional neural network, and the feature extraction layers of subnetworks other than the first-level subnetwork are Transformer encoders.

[0156] In a possible implementation, the fuzzy dynamic reasoning device 600 for industrial time series further includes:

[0157] The training module 603 is used to train the preset initial fuzzy dynamic network based on pre-acquired training data to obtain a fuzzy dynamic network, where the training data includes a plurality of industrial time series data.

[0158] In a possible implementation, the training module 603 is specifically configured to:

[0159] According to the training data, the outputs of the shallow sub-network and the deep sub-network are weighted and superimposed, and the network parameters of the initial fuzzy dynamic network are iteratively updated based on the preset gradient redistribution rule until the preset number of iterations is reached to obtain the fuzzy dynamic network;

[0160] Among them, the gradient redistribution rules include: blocking the gradient transfer of shallow sub-networks during backpropagation, and for the shared parameters of the n-th level sub-network, the gradient of the shared parameters is jointly calculated by the prediction loss of the n-th to the last level sub-network, and the gradient value is weighted averaged according to the number of sub-networks participating in the calculation, where n is an integer greater than or equal to 1.

[0161] The fuzzy dynamic reasoning device for industrial time series provided in this embodiment can execute the fuzzy dynamic reasoning method for industrial time series provided in the above method embodiment. Its implementation principle and technical effects are similar, and will not be described in detail in this embodiment.

[0162] Figure 7 This is a schematic diagram of the structure of the electronic device provided in this application. Figure 7 As shown, the electronic device 700 provided in this embodiment includes: at least one processor 701 and a memory 702. Optionally, the electronic device 700 further includes a communication component 703. The processor 701, the memory 702 and the communication component 703 are connected via a bus 704.

[0163] During the specific implementation process, at least one processor 701 executes the computer-executable instructions stored in the memory 702, so that the at least one processor 701 performs the above method.

[0164] The specific implementation process of the processor 701 can be found in the above method embodiment. Its implementation principle and technical effects are similar and will not be repeated here in this embodiment.

[0165] In the above embodiments, it should be understood that the processor may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASICs), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the present invention may be directly executed by a hardware processor or by a combination of hardware and software modules within the processor.

[0166] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage.

[0167] A bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus. Buses can be categorized as address buses, data buses, and control buses. For ease of illustration, the buses in the drawings of this application are not limited to just one bus or just one type of bus.

[0168] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0169] The present application also provides a computer-readable storage medium, in which computer-executable instructions are stored. When a processor executes the computer-executable instructions, the above method is implemented.

[0170] The readable storage medium may be implemented by any type of volatile or non-volatile memory device, or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0171] An exemplary readable storage medium is coupled to a processor so that the processor can read information from the readable storage medium and write information to the readable storage medium. Of course, the readable storage medium can also be an integral part of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0172] The division of units is merely a logical functional division; actual implementations may employ alternative divisions, such as combining or integrating multiple units or components into another system, or omitting or disabling certain features. Furthermore, any direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection between devices or units, either through an interface, electrical, mechanical, or other means.

[0173] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0174] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0175] If a function is implemented as a software functional unit and sold or used as an independent product, it 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. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the method of the present invention. The aforementioned storage medium includes various media that can store program code, such as USB flash drives, mobile hard drives, read-only memories (ROMs), random access memories (RAMs), magnetic disks, or optical disks.

[0176] Those skilled in the art will appreciate that all or part of the steps in the above-described method embodiments can be implemented using hardware associated with program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0177] Finally, it should be noted that those skilled in the art will readily identify other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. The present invention is intended to cover any variations, uses, or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art not disclosed herein. The present invention is not limited to the precise structure described above and illustrated in the accompanying drawings, and various modifications and variations may be made without departing from the scope thereof. The scope of the present invention is limited solely by the appended claims.

Claims

1. A fuzzy dynamic reasoning method for industrial time series, characterized by: include: Preprocess the pre-collected industrial time series data of the target industrial equipment to obtain industrial time series sample data; Inputting the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment, the fuzzy dynamic network includes a fuzzy logic representation layer, a sample information fusion layer and a time series multi-export dynamic network layer, the multi-export dynamic network layer includes multiple sub-networks; Inputting the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment includes: Performing fuzzy logic processing on the industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information; Splicing the fuzzy membership information and the industrial time series sample data to obtain splicing information; Processing the spliced ​​information through the sample information fusion layer to obtain mixed features; The hybrid features are inferred through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment.

2. The method according to claim 1, characterized in that The reasoning on the hybrid feature through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment includes: Step a: inputting the mixed features into the first-level sub-network, obtaining the first-level deep features through feature extraction, and generating the first-level prediction value and the first-level confidence; Step b: If the first-level confidence is less than a preset threshold, the first-level depth feature is input into the second-level subnetwork to obtain a second-level depth feature, the first-level depth feature is weighted according to a preset attenuation coefficient and then superimposed with the second-level depth feature to generate a second-level fusion feature, and a second-level prediction value and a second-level confidence are generated based on the second-level fusion feature; Step c: repeat step b until a prediction value corresponding to a subnetwork with a confidence level greater than the preset threshold is obtained, and the prediction value is determined as the remaining life of the target industrial equipment.

3. The method according to claim 1, characterized in that The processing of the splicing information by the sample information fusion layer to obtain mixed features includes: The splicing information is sequentially subjected to mean trend extraction, peak trend extraction, and cross-channel dependency extraction in a time dimension, and the extracted features are fused to obtain the hybrid feature.

4. The method according to claim 1, wherein Each subnetwork includes a feature extraction layer, a confidence predictor, and a regression predictor. The feature extraction layer of the first-level subnetwork is a one-dimensional convolutional neural network, and the feature extraction layers of other subnetworks except the first-level subnetwork are Transformer encoders.

5. The method according to claim 1, wherein The method further comprises: A preset initial fuzzy dynamic network is trained based on pre-acquired training data to obtain the fuzzy dynamic network, wherein the training data includes a plurality of industrial time series data.

6. The method according to claim 5, characterized in that The step of training a preset initial fuzzy dynamic network based on pre-acquired training data to obtain the fuzzy dynamic network includes: According to the training data, the network parameters of the initial fuzzy dynamic network are iteratively updated by weighted superposition of the outputs of the shallow sub-network and the deep sub-network based on a preset gradient redistribution rule until a preset number of iterations is reached, thereby obtaining the fuzzy dynamic network; Among them, the gradient redistribution rule includes: blocking the gradient transfer of shallow sub-networks during back propagation, and for the shared parameters of the n-th level sub-network, the gradient of the shared parameters is jointly calculated by the prediction loss of the n-th level to the last level sub-network, and the gradient value is weighted averaged according to the number of sub-networks participating in the calculation, where n is an integer greater than or equal to 1.

7. A fuzzy dynamic reasoning device for industrial time series, characterized in that: include: The preprocessing module preprocesses the industrial time series data of the target industrial equipment collected in advance to obtain industrial time series sample data; A prediction module is configured to input the industrial time series sample data into a pre-trained fuzzy dynamic network to predict the remaining life of the target industrial equipment, wherein the fuzzy dynamic network includes a fuzzy logic representation layer, a sample information fusion layer, and a time series multi-export dynamic network layer, wherein the multi-export dynamic network layer includes multiple sub-networks; The prediction module is specifically configured to perform fuzzy logic processing on the industrial time series sample data through the fuzzy logic representation layer to obtain fuzzy membership information; Splicing the fuzzy membership information and the industrial time series sample data to obtain splicing information; Processing the spliced ​​information through the sample information fusion layer to obtain mixed features; The hybrid features are inferred through the time-series multi-export dynamic network layer to obtain the remaining life of the target industrial equipment.

8. An electronic device, characterized in that: include: Memory, processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory, so that the processor performs the fuzzy dynamic reasoning method for industrial time series according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the fuzzy dynamic reasoning method for industrial time series according to any one of claims 1 to 6.

Images