Lightweight adaptive prediction method for residual service life of industrial equipment, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a lightweight self-adaptive prediction method for the residual service life of industrial equipment, electronic equipment and a storage medium, and the method comprises the steps: obtaining continuous operation data generated by equipment components of the industrial equipment in an operation process, and generating an operation time sequence corresponding to the equipment components according to the continuous operation data; performing resolution reduction processing on the operation time sequence to obtain a target sequence corresponding to the operation time sequence; identifying a resolution level corresponding to the target sequence, and determining a target life prediction model corresponding to the target sequence according to the resolution level; wherein the resolution level is used for indicating the complexity of the operation time sequence; and sending the target sequence to the target life prediction model to obtain an equipment component life prediction result of the industrial equipment, and the method is used for achieving the effects of accurately predicting the equipment component life of the industrial equipment and reducing the operation cost.

Description

Technical Field

[0001] The present application relates to the field of data prediction, and in particular to a lightweight adaptive prediction method, electronic device and storage medium for the remaining useful life of industrial equipment. Background Art

[0002] In industrial scenarios, since industrial processes are often dynamic, using a unified model structure to process all samples during industrial time series analysis will result in insufficient model representation and reduced accuracy. At the same time, industrial time series data are usually long and high-dimensional, and using complex neural networks to process these data consumes a lot of computing resources, resulting in high computing costs.

[0003] Furthermore, there is an urgent need for a technical solution that can accurately predict the life of equipment components based on the complexity of the time series and reduce computing costs. Summary of the invention

[0004] The embodiments of the present application provide a lightweight adaptive prediction method, device, electronic device, storage medium and program product for the remaining useful life of industrial equipment, so as to achieve the effect of accurately predicting the life of equipment components according to the complexity of the time series and reducing computing costs.

[0005] In a first aspect, an embodiment of the present application provides a method for predicting the life of equipment components of industrial equipment, comprising: obtaining continuous operation data generated by equipment components of industrial equipment during operation, and generating an operation time series corresponding to the equipment components based on the continuous operation data; performing resolution reduction processing on the operation time series to obtain a target sequence corresponding to the operation time series; identifying the corresponding resolution level of the target sequence, and determining a target life prediction model corresponding to the target sequence based on the resolution level; wherein the resolution level is used to indicate the complexity of the operation time series; and sending the target sequence to the target life prediction model to obtain a life prediction result of the equipment components of the industrial equipment.

[0006] In a possible implementation, the running time series is processed by a first algorithm to obtain a first time series; wherein the first algorithm is used to reduce the resolution corresponding to the running time series to a first resolution level; and the first time series is determined as a target sequence.

[0007] In a possible implementation, the operation time series is smoothed through a first sliding window to obtain a smoothed sequence corresponding to the operation time series, wherein the smoothed sequence is used to characterize the overall trend of the operation time series; the operation time series is peak processed through a second sliding window to obtain a peak sequence corresponding to the operation time series, wherein the peak sequence is used to characterize the peak trend of the operation time series; the operation time series is feature extracted through a third sliding window to obtain a frequency feature sequence corresponding to the operation time series, wherein the frequency feature sequence is used to characterize the frequency trend of the time series; a first time series is generated based on the smoothed sequence, the peak sequence, and the frequency feature sequence; wherein the window sizes of the first sliding window, the second sliding window, and the third sliding window are the same or different.

[0008] In a possible implementation, the first sliding window is moved along the running time sequence until the first sliding window traverses the running time sequence; the mean of the running time sequence located in the first sliding window each time the first sliding window slides is calculated; the smoothing sequence is generated according to the mean of the running time sequence obtained each time; the second sliding window is moved along the running time sequence until the second sliding window traverses the running time sequence; the maximum value of the running time sequence located in the second sliding window each time the second sliding window slides is calculated; the peak sequence is generated according to the maximum value of the running time sequence obtained each time; the third sliding window is moved along the running time sequence until the third sliding window traverses the running time sequence; the frequency feature of the running time sequence located in the third sliding window each time the third sliding window slides is calculated; and the frequency feature sequence is generated according to the frequency feature of the running time sequence obtained each time.

[0009] In a possible implementation, the target sequence is sent to a life prediction model corresponding to the preset resolution level for processing to obtain a preprocessing result; when the predicted value carried in the preprocessing result falls within a preset confidence interval, the preprocessing result is determined as the life prediction result of the industrial equipment; if the predicted value carried in the preprocessing result does not fall within the preset confidence interval, the operating time series is processed by a second algorithm to obtain a second time series; wherein the second algorithm is used to reduce the resolution corresponding to the operating time series to a second resolution level; the second resolution level is greater than the first resolution level; the second time series is determined as the target sequence and sent to a downstream prediction model for processing; wherein the resolution level corresponding to the downstream prediction model is greater than the resolution level currently corresponding to the target time series.

[0010] In a possible implementation, the target sequence is sent to each layer of the sub-network of the life prediction model for transfer processing to obtain a transfer prediction result; and the transfer prediction result is determined as the life prediction result.

[0011] In one possible implementation, the target sequence is sent to a first layer sub-network for processing to obtain a first prediction result; a downstream sub-network of the first layer sub-network is allowed to perform prediction processing based on the first prediction result to obtain a second prediction result; until the last layer sub-network of the prediction model outputs a final prediction result, the final prediction result is determined as the transfer prediction result.

[0012] In one possible implementation, the processing result of the i+1th layer subnetwork and the processing result of the i-th layer subnetwork are sent to the classifier corresponding to the i+1th layer subnetwork to obtain a feedback result of the classifier; wherein 1<i+1≤total number of layers of the prediction model subnetwork, and the feedback result is used to indicate the degree of error; the feedback result is multiplied by the weight value of the i+1th layer subnetwork to obtain a first adjustment value, and the i+1th layer subnetwork is adjusted according to the first adjustment value.

[0013] In a possible implementation, a gradient value between each layer of sub-networks is determined, and the gradient value between each layer of sub-networks is multiplied by a gradient alignment factor.

[0014] In a second aspect, an embodiment of the present application provides a lightweight adaptive prediction device for the remaining useful life of industrial equipment, comprising: an acquisition module, used to acquire continuous operating data generated by equipment components of industrial equipment during operation, and generate an operating time series corresponding to the equipment components based on the continuous operating data; a resolution module, used to perform resolution reduction processing on the operating time series to obtain a target sequence corresponding to the operating time series; an identification module, used to identify the corresponding resolution level of the target sequence, and determine a target life prediction model corresponding to the target sequence based on the resolution level; wherein the resolution level is used to indicate the complexity of the operating time series; a sending module, used to send the target sequence to the target life prediction model to obtain a life prediction result of the industrial equipment.

[0015] In a third aspect, an embodiment of the present application provides a lightweight adaptive prediction device for the remaining useful life of industrial equipment, including: a memory, a processor;

[0016] The memory stores computer-executable instructions;

[0017] 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.

[0018] 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 implementations of the first aspect.

[0019] The lightweight adaptive prediction method, device, electronic device, storage medium and program product for the remaining useful life of industrial equipment provided in the embodiments of the present application reduce the resolution of the operating time series of the equipment components, and then send the time series with reduced resolution to the life prediction model for prediction, thereby ensuring the accuracy of the life prediction while reducing the computing cost, thereby achieving the effect of accurately predicting the life of the equipment components according to the complexity of the time series and reducing the computing cost. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] 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.

[0021] Figure 1 Schematic diagram of the process of lightweight adaptive prediction of the remaining useful life of industrial equipment provided in this application Figure 1 ;

[0022] Figure 2 Schematic diagram of the process of lightweight adaptive prediction of the remaining useful life of industrial equipment provided in this application Figure 2 ;

[0023] Figure 3 Schematic diagram of gradient alignment factors for lightweight adaptive prediction of remaining useful life of industrial equipment provided in this application;

[0024] Figure 4 A schematic diagram of the structure of a lightweight adaptive prediction device for the remaining useful life of industrial equipment provided in this application;

[0025] Figure 5 A schematic diagram of the structure of a lightweight adaptive prediction device for the remaining useful life of industrial equipment provided in this application.

[0026] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0027] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0028] First, the terms involved in this application are explained:

[0029] Resolution of time series: The resolution of a time series refers to the minimum time interval between two adjacent observations in a continuous time series data. It determines the frequency and granularity of data sampling and is an important indicator for describing the precision of time series data. When the resolution is low, it means that the time interval between adjacent observation points is larger, so the interval is higher. On the contrary, if the resolution is high, the time interval between adjacent observation points is smaller, and the interval is lower.

[0030] Time series: refers to a set of data arranged at certain time intervals. These data can be continuous or discrete.

[0031] Figure 1 Schematic diagram of the process of lightweight adaptive prediction of the remaining useful life of industrial equipment provided in this application Figure 1 ,like Figure 1 As shown, the method includes:

[0032] S101, acquiring continuous operation data generated by a device component of an industrial device during operation, and generating an operation time series corresponding to the device component according to the continuous operation data;

[0033] It should be noted that, as the equipment is in operation, its components will generate operation data. Based on the analysis of the operation data, the remaining life of the equipment components can be predicted, and the remaining life of the equipment corresponding to the equipment components can be further determined by combining the remaining life of multiple equipment components.

[0034] Optionally, various sensors are deployed on the equipment components of the industrial equipment, such as temperature, pressure, vibration, current, voltage, etc. These sensors can monitor the operating status and parameters of the equipment components in real time.

[0035] S102, performing resolution reduction processing on the running time sequence to obtain a target sequence corresponding to the running time sequence;

[0036] S103, identifying the resolution level corresponding to the target sequence, and determining the target life prediction model corresponding to the target sequence according to the resolution level; wherein the resolution level is used to indicate the complexity of the running time sequence;

[0037] It should be noted that the resolution level is used to indicate the complexity of the running time series. It is usually related to the frequency of data collection, the number of data points, and the range of data changes. High-resolution time series contain more details and changes, while low-resolution time series are more simplified and summarized.

[0038] After down-resolution processing is performed on the running time series, it is also necessary to determine the resolution level corresponding to the target sequence. Different resolution levels are preset to correspond to different life prediction models.

[0039] S104: Send the target sequence to a target life prediction model to obtain a lightweight adaptive prediction result of the remaining service life of the industrial equipment.

[0040] The prediction is performed based on the target sequence to obtain a life prediction result of the equipment component, wherein the life prediction result is used to indicate the remaining life of the equipment component.

[0041] It should be noted that the remaining life of the industrial equipment can also be predicted based on the remaining life of the core performance equipment components of the industrial equipment. For example, if the remaining life of the core performance equipment components is 1 month, the remaining life of the industrial equipment is also 1 month.

[0042] The lightweight adaptive prediction method for the remaining useful life of industrial equipment provided in the embodiment of the present application reduces the resolution of the operating time series of the equipment components, and then sends the time series with reduced resolution to the life prediction model for prediction, thereby ensuring the accuracy of the life prediction while reducing the computing cost, thereby achieving the effect of accurately predicting the life of equipment components according to the complexity of the time series and reducing the computing cost.

[0043] like Figure 2 As shown, Figure 2 Schematic diagram of the process of lightweight adaptive prediction of the remaining useful life of industrial equipment provided in this application Figure 2 , specifically:

[0044] Input, time series data X. Decompose the input time series data X into low resolution (fs) to obtain a series of low resolution signals Z1. Then send the low resolution signal Z1 to the small encoder m for encoding to generate feature vectors . The feature vector Send it to the prediction network to determine The prediction result outputs the classification label Clabel1. If Clabel1 is greater than T1, the remaining useful life (RUL) is output; otherwise, the processing continues. Among them, the classification label Clabel1 is the confidence label of the low-resolution sub-network, which is used to indicate the feature vector The confidence level of the result. T1 is the preset confidence threshold. If the confidence value carried by Clabel1 (confidence 1) is greater than T1, it means that the result is credible, and the predicted remaining useful life can be directly output.

[0045] If Clabel1 (confidence 1) is less than or equal to T1, the time series data X is decomposed into medium resolution (fm). If the confidence corresponding to the prediction result obtained after decomposing the time series using medium resolution is lower than the confidence corresponding to the medium resolution sub-network, it is determined that the time series data X needs to be decomposed into Z2 using high resolution (fL), and Z2 is encoded using a large encoder to obtain .Will Send it to the prediction network to determine The prediction result outputs the classification label Clabel2. If Clabel2 (confidence 2) is greater than T2, the remaining useful life (RUL) is output; otherwise, the processing continues. Use the downstream network (second layer sub-network) corresponding to the high-resolution first layer sub-network to continue Perform prediction processing and output ,Sure The prediction result outputs the classification label Clabel3 (confidence 3). If Clabel3 (confidence 3) is greater than T3, the remaining useful life (RUL) is output; otherwise, the downstream network processing (third layer sub-network) of the current sub-network layer is continued.

[0046] Optionally, continue to use the second layer subnetwork When performing prediction processing, the second layer sub-network can inherit part of the prediction results of the first layer sub-network, that is, and Input into the knowledge collaboration mechanism (KC) to obtain ,Sure The prediction result outputs the classification label Clabel3. If Clabel3 (confidence 3) is greater than T3, the remaining useful life (Remaining Useful Life, referred to as RUL) is output.

[0047] In a possible implementation, the running time series is processed by a first algorithm to obtain a first time series; wherein the first algorithm is used to reduce the resolution corresponding to the running time series to a first resolution level; and the first time series is determined as a target sequence.

[0048] Operation time series refers to the continuous data sequence generated during the operation of equipment components of industrial equipment. These data usually include parameters such as temperature, pressure, vibration, current, voltage, etc., which can reflect the operating status and health of equipment components.

[0049] The first algorithm is a specific data processing algorithm used to down-sample the running time series. The purpose of this algorithm is to convert high-resolution raw data into lower-resolution data in order to reduce the amount of data, improve computational efficiency, or meet specific analysis requirements.

[0050] The running time series after being processed by the first algorithm is called the first time series. It retains the main features of the original data, but has a lower resolution. This means that the number of data points in the first time series is smaller, and the time interval between each data point is larger.

[0051] The first resolution level refers to the resolution level of the first time series. It is used to indicate the complexity of the running time series and the level of detail of the data. A higher resolution level means more data points and more detailed information, while a lower resolution level means fewer data points and more simplified information.

[0052] The first time series is determined as the target series in order to prepare for subsequent analysis or prediction tasks. The target series is a time series that has been processed by resolution reduction and can be used for trend analysis, anomaly detection, prediction model or other statistical analysis tasks.

[0053] In a possible implementation, the operation time series is smoothed through a first sliding window to obtain a smoothed sequence corresponding to the operation time series, wherein the smoothed sequence is used to characterize the overall trend of the operation time series; the operation time series is peak processed through a second sliding window to obtain a peak sequence corresponding to the operation time series, wherein the peak sequence is used to characterize the peak trend of the operation time series; the operation time series is feature extracted through a third sliding window to obtain a frequency feature sequence corresponding to the operation time series, wherein the frequency feature sequence is used to characterize the frequency trend of the time series; a first time series is generated based on the smoothed sequence, the peak sequence, and the frequency feature sequence; wherein the window sizes of the first sliding window, the second sliding window, and the third sliding window are the same or different.

[0054] Optionally, the first sliding window is used to smooth the running time series. The purpose of smoothing is to reduce noise and fluctuations in the data and make the data more stable. By calculating the average value of the data in the sliding window or using other smoothing algorithms (such as moving average, exponential smoothing, etc.), the corresponding smoothed sequence of the running time series can be obtained. The smoothed sequence is used to characterize the overall trend of the running time series.

[0055] The smoothing formula is as follows:

[0056]

[0057] : Represents the average trend characteristics of the kth dimension in the i-th time step.

[0058] : Indicates the index position of the i-th time step.

[0059] m: indicates the size of the first sliding window, which is used to calculate the average value within the specified range.

[0060] Optionally, the second sliding window is used to perform peak processing on the operating time series. The purpose of peak processing is to detect and extract peak points in the time series, which may correspond to important events or abnormal conditions in the operation of the equipment component. By comparing the size of the data points in the sliding window, the local maximum value can be identified, thereby obtaining the peak sequence corresponding to the operating time series. The peak sequence is used to characterize the peak trend of the operating time series.

[0061] The peak processing formula is as follows:

[0062]

[0063] : Represents the peak trend characteristics of the kth dimension at the i-th time step.

[0064] j: represents the maximum value to be found in the second sliding window.

[0065] Optionally, the third sliding window is used to perform feature extraction processing on the running time series. The purpose of feature extraction is to extract useful information from the time series, such as frequency features. By applying Fourier transform or other frequency domain analysis methods in the sliding window, a frequency feature sequence corresponding to the running time series can be obtained. The frequency feature sequence is used to characterize the frequency trend of the time series.

[0066] The feature extraction processing formula is as follows:

[0067]

[0068]

[0069] : Represents the convolution feature (frequency feature) of the i-th time step and the k-th dimension.

[0070] : Represents the convolution kernel of the kth dimension (the third sliding window), which is used to perform convolution operations on time series.

[0071] b: Bias term.

[0072] : Hyperbolic tangent function, as the activation function.

[0073] : Batch normalization operation, used to accelerate training and stabilize gradients.

[0074] The convolution kernel operation is used to extract local frequency information from time series data and capture fine-grained changes and temporal dependencies.

[0075] Optionally, a first time series is generated based on a smoothing sequence, a peak sequence and a frequency feature sequence. The first time series is a time series that integrates multiple features, and it integrates information on the overall trend, the peak trend and the frequency trend. By combining these three feature sequences according to a certain weight or method, a first time series that more comprehensively and accurately describes the characteristics of the running time series can be obtained.

[0076] Finally, the output is:

[0077] The window sizes of the first sliding window, the second sliding window, and the third sliding window can be the same or different. The choice of window size depends on the specific analysis requirements and the characteristics of the data. A larger window can provide smoother results, but may lose some details; a smaller window can retain more details, but may introduce more noise.

[0078] The feature complexity of samples is different: for simple samples, only lower resolution feature extraction is needed to meet the accuracy of prediction; while complex samples require higher resolution to extract fine-grained information. By dynamically selecting the appropriate resolution, the model can strike a balance between efficiency and accuracy. This adaptability can avoid using overly complex models to process simple samples, reduce computational overhead, and also retain important information in complex samples to ensure the prediction accuracy of the model.

[0079] In a possible implementation, the first sliding window is moved along the running time sequence until the first sliding window traverses the running time sequence; the mean of the running time sequence located in the first sliding window each time the first sliding window slides is calculated; the smoothing sequence is generated according to the mean of the running time sequence obtained each time; the second sliding window is moved along the running time sequence until the second sliding window traverses the running time sequence; the maximum value of the running time sequence located in the second sliding window each time the second sliding window slides is calculated; the peak sequence is generated according to the maximum value of the running time sequence obtained each time; the third sliding window is moved along the running time sequence until the third sliding window traverses the running time sequence; the frequency feature of the running time sequence located in the third sliding window each time the third sliding window slides is calculated; and the frequency feature sequence is generated according to the frequency feature of the running time sequence obtained each time.

[0080] In a possible implementation, the target sequence is sent to a life prediction model corresponding to the preset resolution level for processing to obtain a preprocessing result; when the predicted value carried in the preprocessing result falls within a preset confidence interval, the preprocessing result is determined as the life prediction result of the industrial equipment; if the predicted value carried in the preprocessing result does not fall within the preset confidence interval, the operating time series is processed by a second algorithm to obtain a second time series; wherein the second algorithm is used to reduce the resolution corresponding to the operating time series to a second resolution level; the second resolution level is greater than the first resolution level; the second time series is determined as the target sequence and sent to a downstream prediction model for processing; wherein the resolution level corresponding to the downstream prediction model is greater than the resolution level currently corresponding to the target time series.

[0081] By comparing the predicted value in the preprocessing result with the preset confidence interval, it can be determined whether the prediction is credible. If it is credible, the preprocessing result can be directly used as the life prediction result; if it is not credible, it is necessary to further process the running time series through the second algorithm to obtain the second time series, and determine it as the target series, and then send it to the downstream prediction model for processing. This can ensure the accuracy and reliability of the prediction results.

[0082] In a possible implementation, the target sequence is sent to each layer of the sub-network of the life prediction model for transfer processing to obtain a transfer prediction result; and the transfer prediction result is determined as the life prediction result.

[0083] In one possible implementation, the target sequence is sent to a first layer sub-network for processing to obtain a first prediction result; a downstream sub-network of the first layer sub-network is allowed to perform prediction processing based on the first prediction result to obtain a second prediction result; until the last layer sub-network of the prediction model outputs a final prediction result, the final prediction result is determined as the transfer prediction result.

[0084] The first layer of the sub-network is the first layer of the life prediction model, which is responsible for receiving the target sequence and performing preliminary processing on it. After processing by the first layer of the sub-network, an intermediate prediction result, called the first prediction result, is obtained. This result is obtained based on the preliminary analysis of the target sequence. The downstream sub-networks of the first layer of the sub-network refer to other layers in the model, which perform further prediction processing based on the first prediction result. These sub-networks can gradually extract deeper features and perform more complex analysis and prediction. After processing by the downstream sub-networks, another intermediate prediction result, called the second prediction result, is obtained. This result is obtained based on the comprehensive analysis of the first prediction result and deeper features, providing more information about the life of the equipment component. Until the last layer of the sub-network of the prediction model outputs the final prediction result. This result is the comprehensive output of the entire model, based on the processing and analysis of all levels, and is the most accurate prediction of the life of the equipment component. The final prediction result is determined as the transfer prediction result. This result will be used to guide the maintenance and replacement decisions of the equipment.

[0085] In one possible implementation, the processing result of the i+1th layer subnetwork and the processing result of the i-th layer subnetwork are sent to the classifier corresponding to the i+1th layer subnetwork to obtain a feedback result of the classifier; wherein 1<i+1≤total number of layers of the prediction model subnetwork, and the feedback result is used to indicate the degree of error; the feedback result is multiplied by the weight value of the i+1th layer subnetwork to obtain a first adjustment value, and the i+1th layer subnetwork is adjusted according to the first adjustment value.

[0086] The i+1th layer subnetwork is a layer in the life prediction model, which is responsible for receiving the processing results from the previous layer (i-th layer) subnetwork and performing further processing. A classifier is a machine learning algorithm that is used to assign input data to predefined categories. The classifier is used to evaluate the processing results of the i+1th layer subnetwork and provide feedback. The feedback result is the classifier's evaluation of the processing results of the i+1th layer subnetwork. This result is used to indicate the degree of error between the processing result and the actual label. The weight value is a measure of the importance of each feature in the model. The weight value is used to adjust the parameters of the i+1th layer subnetwork to reduce errors and improve prediction accuracy. The first adjustment value is calculated based on the feedback result and the weight value. This value is used to adjust the i+1th layer subnetwork to improve its performance. The total number of layers of the prediction model subnetwork refers to the number of all subnetworks in the life prediction model. This number determines the complexity and capability of the model.

[0087] In a possible implementation, a gradient value between each layer of sub-networks is determined, and the gradient value between each layer of sub-networks is multiplied by a gradient alignment factor.

[0088] The gradient alignment strategy optimizes the training of different subnetworks by assigning different training gradients to upstream and downstream subnetworks. The knowledge collaboration mechanism allows the downstream network to use the knowledge of the upstream network for collaborative reasoning and learn more accurate multi-scale features. Its purpose is to improve the reasoning ability and performance of the model.

[0089] like Figure 3 As shown, Figure 3 Schematic diagram of the gradient alignment factor for lightweight adaptive prediction of the remaining useful life of industrial equipment provided in this application. Specifically: a gradient alignment factor is introduced to directly act on the gradient of the model. During the back propagation process, the gradient of each parameter is multiplied by this alignment factor to appropriately adjust the size of the gradient. The calculation formula of the gradient alignment factor is as follows:

[0090]

[0091] Formula composition:

[0092] 1. Gradient symbol: Represents the loss function for the bth task About the parameters of the ath task gradient.

[0093] 2. Gradient alignment factor: This part consists of two main components:

[0094] Product term: is a multiplication expression that spans all intermediate tasks d from a to b, assigning a scaling factor to each task. The purpose of this factor is to adjust the gradients from subsequent tasks so that they do not dominate when updating the current task parameters.

[0095] Division term: is a direct scaling of the gradient of the bth task. It ensures that even if b is a later task in the sequence, its gradient will not have too much impact on the overall learning process.

[0096] Gradient alignment logic:

[0097] The formula performs a normalization of the gradient, making the impact of all tasks on parameter updates more balanced during training. The effect of the multiplication term is to reduce the gradient contribution of all tasks from a to b, preventing later tasks from having too much impact on the parameter updates of earlier tasks. The division term directly reduces the size of the gradient of the bth task, ensuring that the gradient size between different tasks is relatively balanced. Finally, after combining all these adjustment factors, the result is shown as This shows that the gradient of each task is appropriately scaled by its position in the task sequence.

[0098] Specifically, a knowledge transfer path is added between every two adjacent classifiers, and the basic temporal feature representation Zi∈ R Li∗dmodel (dmodel represents the dimension of the feature representation) of the i-th stage is superimposed with the basic temporal feature representation Zi+1 of the i+1 stage, thereby obtaining Zconi+1 and using it in the i+1 classifier to obtain more accurate prediction results. In this case, each classifier (except the first classifier) ​​can be regarded as performing residual learning, applying different weights to networks in different dynamic layers, and solving the problem of inconsistent gradients in dynamic network training inference.

[0099] The lightweight adaptive prediction method for the remaining useful life of industrial equipment provided in the embodiment of the present application obtains continuous operating data generated by equipment components of industrial equipment during operation, and generates an operating time series corresponding to the equipment components based on the continuous operating data; performs resolution reduction processing on the operating time series to obtain a target sequence corresponding to the operating time series; identifies the corresponding resolution level of the target sequence, and determines a target life prediction model corresponding to the target sequence based on the resolution level; wherein the resolution level is used to indicate the complexity of the operating time series; sends the target sequence to the target life prediction model to obtain a lightweight adaptive prediction result means for the remaining useful life of the industrial equipment, thereby achieving accurate prediction of the life of the equipment components based on the complexity of the time series and reducing the computing cost.

[0100] Figure 4A schematic diagram of the structure of a lightweight adaptive prediction device for the remaining useful life of industrial equipment provided in this application, such as Figure 4 As shown, the lightweight adaptive prediction device 40 for the remaining useful life of industrial equipment provided in this embodiment includes:

[0101] The acquisition module 401 is used to acquire continuous operation data generated by a device component of the industrial device during operation, and generate an operation time series corresponding to the device component according to the continuous operation data;

[0102] A resolution module 402 is used to perform resolution reduction processing on the running time sequence to obtain a target sequence corresponding to the running time sequence;

[0103] An identification module 403 is used to identify the resolution level corresponding to the target sequence, and determine the target life prediction model corresponding to the target sequence according to the resolution level; wherein the resolution level is used to indicate the complexity of the running time series;

[0104] The sending module 404 is used to send the target sequence to the target life prediction model to obtain the life prediction result of the industrial equipment.

[0105] In a possible implementation, the resolution module 402 is also used to process the running time series through a first algorithm to obtain a first time series; wherein the first algorithm is used to reduce the resolution corresponding to the running time series to a first resolution level; and determine the first time series as a target sequence.

[0106] In a possible implementation, the resolution module 402 is also used to smooth the operation time series through a first sliding window to obtain a smoothed sequence corresponding to the operation time series, wherein the smoothed sequence is used to characterize the overall trend of the operation time series; perform peak processing on the operation time series through a second sliding window to obtain a peak sequence corresponding to the operation time series, wherein the peak sequence is used to characterize the peak trend of the operation time series; perform feature extraction processing on the operation time series through a third sliding window to obtain a frequency feature sequence corresponding to the operation time series, wherein the frequency feature sequence is used to characterize the frequency trend of the time series; generate a first time series based on the smooth sequence, the peak sequence, and the frequency feature sequence; wherein the window sizes of the first sliding window, the second sliding window, and the third sliding window are the same or different.

[0107] In a possible implementation, the resolution module 402 is also used to move the first sliding window along the running time sequence until the first sliding window traverses the running time sequence; calculate the mean of the running time sequence located in the first sliding window each time the first sliding window slides; generate the smoothing sequence according to the mean of the running time series obtained each time; move the second sliding window along the running time sequence until the second sliding window traverses the running time sequence; calculate the maximum value of the running time series located in the second sliding window each time the second sliding window slides; generate the peak sequence according to the maximum value of the running time series obtained each time; move the third sliding window along the running time sequence until the third sliding window traverses the running time sequence; calculate the frequency characteristics of the running time series located in the third sliding window each time the third sliding window slides; generate the frequency characteristic sequence according to the frequency characteristics of the running time series obtained each time sliding.

[0108] In a possible implementation, the sending module 404 is also used to send the target sequence to a life prediction model corresponding to the preset resolution level for processing to obtain a preprocessing result; when the predicted value carried in the preprocessing result falls within a preset confidence interval, the preprocessing result is determined as the life prediction result of the industrial equipment; if the predicted value carried in the preprocessing result does not fall within the preset confidence interval, the running time series is processed by a second algorithm to obtain a second time series; wherein the second algorithm is used to reduce the resolution corresponding to the running time series to a second resolution level; the second resolution level is greater than the first resolution level; the second time series is determined as the target sequence and sent to a downstream prediction model for processing; wherein the resolution level corresponding to the downstream prediction model is greater than the resolution level currently corresponding to the target time series.

[0109] In a possible implementation, the sending module 404 is further used to send the target sequence to each layer of the sub-network of the life prediction model for transfer processing to obtain a transfer prediction result; and determine the transfer prediction result as the life prediction result.

[0110] In a possible implementation, the sending module 404 is also used to send the target sequence to the first layer sub-network for processing to obtain a first prediction result; allow the downstream sub-network of the first layer sub-network to perform prediction processing based on the first prediction result to obtain a second prediction result; until the last layer sub-network of the prediction model outputs the final prediction result, the final prediction result is determined as the transfer prediction result.

[0111] In one possible implementation, the sending module 404 is also used to send the processing result of the i+1th layer sub-network and the processing result of the i-th layer sub-network to the classifier corresponding to the i+1th layer sub-network to obtain a feedback result of the classifier; wherein 1<i+1≤total number of layers of the prediction model sub-network, and the feedback result is used to indicate the degree of error; the feedback result is multiplied by the weight value of the i+1th layer sub-network to obtain a first adjustment value, and the i+1th layer sub-network is adjusted according to the first adjustment value.

[0112] In a possible implementation, a gradient value between each layer of sub-networks is determined, and the gradient value between each layer of sub-networks is multiplied by a gradient alignment factor.

[0113] The lightweight adaptive prediction device for the remaining useful life of industrial equipment provided in this embodiment can execute the method provided in the above method embodiment. Its implementation principle and technical effect are similar, and this embodiment will not be described in detail here.

[0114] Figure 5 This is a schematic diagram of the structure of a lightweight adaptive prediction device for the remaining useful life of industrial equipment provided in this application. Figure 5 As shown, the electronic device 50 provided in this embodiment includes: at least one processor 501 and a memory 502. Optionally, the device 50 also includes a communication component 503. The processor 501, the memory 502 and the communication component 503 are connected via a bus 504.

[0115] In a specific implementation process, at least one processor 501 executes the computer-executable instructions stored in the memory 502, so that at least one processor 501 executes the above method.

[0116] The specific implementation process of the processor 501 can be found in the above method embodiment, and its implementation principle and technical effect are similar, so this embodiment will not be repeated here.

[0117] In the above embodiments, it should be understood that the processor can be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the invention can be directly implemented as a hardware processor, or can be implemented by a combination of hardware and software modules in the processor.

[0118] The memory may include a high-speed memory (Random Access Memory, RAM), and may also include a non-volatile memory (NVM), such as at least one disk storage.

[0119] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of this application is not limited to only one bus or one type of bus.

[0120] The present application also provides a lightweight adaptive prediction device for the remaining useful life of industrial equipment, comprising: a memory, a processor; the memory stores computer-executable instructions; the processor executes the computer-executable instructions stored in the memory, so that the processor executes the above method.

[0121] The present application also provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0122] 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.

[0123] The above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage 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 memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.

[0124] 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 a component of the processor. The processor and the readable storage medium can be located in an application specific integrated circuit (Application Specific Integrated Circuits, referred to as: ASIC). Of course, the processor and the readable storage medium can also exist in the device as discrete components.

[0125] The division of units is only a logical function division, and there may be other divisions in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0126] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0127] 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.

[0128] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0129] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.

[0130] Finally, it should be noted that those skilled in the art will readily conceive of 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, which follow the general principles of the present invention and include common knowledge or customary technical means in the art not disclosed by the present invention, are not limited to the precise structure described above and shown in the drawings, and may be modified and changed in various ways without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims

1. A lightweight adaptive prediction method for the remaining useful life of industrial equipment, characterized in that: include: Acquire continuous operation data generated by a device component of the industrial equipment during operation, and generate an operation time series corresponding to the device component according to the continuous operation data; Performing resolution reduction processing on the running time sequence to obtain a target sequence corresponding to the running time sequence; Identifying a resolution level corresponding to the target sequence, and determining a target life prediction model corresponding to the target sequence according to the resolution level; wherein the resolution level is used to indicate the complexity of the running time sequence; The target sequence is sent to a target life prediction model to obtain a life prediction result of the equipment component of the industrial equipment.

2. The method according to claim 1, characterized in that Performing resolution reduction processing on the running time sequence to obtain a target sequence corresponding to the running time sequence includes: After processing the running time series by a first algorithm, a first time series is obtained; wherein the first algorithm is used to reduce the resolution corresponding to the running time series to a first resolution level; The first time series is determined as a target series.

3. The method according to claim 2, characterized in that After the running time series is processed by a first algorithm, a first time series is obtained, including: Smoothing the running time series through a first sliding window to obtain a smoothed sequence corresponding to the running time series, wherein the smoothed sequence is used to characterize the overall trend of the running time series; Performing peak processing on the running time series through a second sliding window to obtain a peak sequence corresponding to the running time series, wherein the peak sequence is used to characterize a peak trend of the running time series; Performing feature extraction processing on the running time series through a third sliding window to obtain a frequency feature sequence corresponding to the running time series, wherein the frequency feature sequence is used to characterize the frequency trend of the time series; Generate a first time series based on the smoothing sequence, the peak sequence, and the frequency feature sequence; The window sizes of the first sliding window, the second sliding window, and the third sliding window are the same or different.

4. The method according to claim 3, characterized in that include: Moving the first sliding window along the running time sequence until the first sliding window traverses the running time sequence; Calculating the mean of the running time series within the first sliding window each time the first sliding window slides; Generate the smoothed sequence according to the mean of the running time series obtained by each sliding; Moving the second sliding window along the running time sequence until the second sliding window traverses the running time sequence; Calculating the maximum value of the running time series within the second sliding window each time the second sliding window slides; Generate the peak sequence according to the maximum value of the running time series obtained by each sliding; Moving the third sliding window along the running time sequence until the third sliding window traverses the running time sequence; Calculating the frequency characteristics of the running time series within the third sliding window each time the third sliding window slides; The frequency feature sequence is generated according to the frequency feature of the running time series obtained by each sliding.

5. The method according to claim 2, characterized in that: Sending the target sequence to a target life prediction model to obtain a life prediction result of a device component of the industrial equipment includes: Sending the target sequence to a life prediction model corresponding to the preset resolution level for processing to obtain a preprocessing result; In the case where the predicted value carried in the preprocessing result falls within a preset confidence interval, determining the preprocessing result as the life prediction result of the industrial equipment; If the predicted value carried in the preprocessing result does not fall within the preset confidence interval, the running time series is processed by a second algorithm to obtain a second time series; wherein the second algorithm is used to reduce the resolution corresponding to the running time series to a second resolution level; the second resolution level is greater than the first resolution level; The second time series is determined as a target sequence and sent to a downstream prediction model for processing; wherein the resolution level corresponding to the downstream prediction model is greater than the resolution level currently corresponding to the target sequence.

6. The method according to claim 1, characterized in that Sending the target sequence to a target life prediction model to obtain a life prediction result of a device component of the industrial equipment includes: Sending the target sequence to each layer of the life prediction model sub-network for transfer processing to obtain a transfer prediction result; The transfer prediction result is determined as the life prediction result.

7. The method according to claim 6, characterized in that The target sequence is sent to each layer of the sub-network of the life prediction model for transfer processing, including: Sending the target sequence to the first layer sub-network for processing to obtain a first prediction result; Allowing a downstream sub-network of the first layer of sub-network to perform prediction processing based on the first prediction result to obtain a second prediction result; Until the last layer of sub-network of the prediction model outputs the final prediction result, the final prediction result is determined as the transferred prediction result.

8. The method according to claim 6, characterized in that Before determining the transfer prediction result as the target processing result, the method further includes: Send the processing result of the i+1th layer subnetwork and the processing result of the ith layer subnetwork to the classifier corresponding to the i+1th layer subnetwork to obtain the feedback result of the classifier; wherein 1<i+1≤total number of layers of the prediction model subnetwork, and the feedback result is used to indicate the degree of error; The feedback result is multiplied by the weight value of the i+1th layer sub-network to obtain a first adjustment value, and the i+1th layer sub-network is adjusted according to the first adjustment value.

9. The method according to claim 5, characterized in that Before determining the transfer prediction processing result as the target processing result, the method further includes: Determine the gradient value between each layer of sub-networks, and multiply the gradient value between each layer of sub-networks by the gradient alignment factor.

10. A lightweight adaptive prediction device for the remaining useful life of industrial equipment, characterized in that: include: An acquisition module, used to acquire continuous operation data generated by a device component of the industrial equipment during operation, and generate an operation time series corresponding to the device component according to the continuous operation data; A resolution module, used for performing resolution reduction processing on the running time sequence to obtain a target sequence corresponding to the running time sequence; an identification module, used to identify the resolution level corresponding to the target sequence, and determine the target life prediction model corresponding to the target sequence according to the resolution level; wherein the resolution level is used to indicate the complexity of the running time sequence; The sending module is used to send the target sequence to the target life prediction model to obtain the life prediction result of the industrial equipment.

11. A lightweight adaptive prediction device for the remaining useful life of industrial equipment, 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 method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 9 when executed by a processor.

Images