An on-line production monitoring method and system for a power cord

Abstract

The present application provides an on-line production monitoring method and system for power cords. By obtaining sample production parameter status path data with and without abnormal status chain tag data, and using these sample production parameter status path data to train and optimize the abnormal status chain prediction network, it can not only effectively identify the abnormal status chain in the production line, but also improve the accuracy and reliability of abnormal status prediction by introducing a guidance machine learning network to generate confidence evaluation. The optimized abnormal status chain prediction network can accurately predict the abnormal status chain of the target production parameter status path data, thus realizing the real-time monitoring and fault warning of the power cord production process, and significantly improving the production efficiency and product quality.

Description

Technical Field

[0001] This application relates to the technical field of intelligent systems for power cord production and detection. Specifically, it relates to a method and system for on-line production monitoring of power cords. Background Art

[0002] During the production process of power cords, due to the influence of various factors, such as equipment failures, raw material problems, or improper operations, etc., the produced power cords may have quality problems or abnormal states. These abnormal states not only affect the performance and service life of the power cords, but also may pose a threat to the safety of users. Therefore, it is particularly important to monitor the power cord production process in real time and identify abnormal states.

[0003] Traditional power cord production monitoring methods mainly rely on manual inspections and sampling detections. This method is not only inefficient, but also difficult to comprehensively cover all aspects of the production process. Machine learning technology has gradually been introduced into production monitoring to improve the accuracy and efficiency of monitoring. However, when existing machine learning algorithms are used to handle power cord production monitoring tasks, they often face problems such as difficult data annotation and low accuracy in identifying abnormal states.

[0004] Specifically, due to the variety of abnormal states in the power cord production process, and many abnormal states are rare or difficult to reproduce, it is difficult to obtain a large amount of training data with abnormal state labels. This results in a lack of sufficient supervision information in the training process of existing machine learning algorithms, making it difficult to learn an effective abnormal state recognition model. At the same time, due to the complexity of the power cord production process, abnormal states are often not caused by a single factor, but are the result of the interaction of multiple factors, which makes the identification of abnormal states more difficult. Summary of the Invention

[0005] In view of this, the purpose of this application is to provide a method and system for on-line production monitoring of power cords.

[0006] According to the first aspect of this application, a method for on-line production monitoring of power cords is provided. The method includes:

[0007] Obtain the first sample production parameter status path data and the second sample production parameter status path data of each reference power cord production line. The first sample production parameter status path data carries abnormal state chain label data, and the second sample production parameter status path data does not carry abnormal state chain label data;

[0008] Load the first sample production parameter status path data and the second sample production parameter status path data into an abnormal status chain prediction network serving as a guided machine learning network respectively, to generate a first abnormal status chain prediction result of the first sample production parameter status path data and a second abnormal status chain prediction result of the second sample production parameter status path data;

[0009] Load the second sample production parameter status path data into a guiding machine learning network corresponding to the guided machine learning network, to generate a third abnormal status chain prediction result of the second sample production parameter status path data, where the third abnormal status chain prediction result represents the confidence levels of an abnormal status chain and a target abnormal status label;

[0010] If the confidence level is greater than a threshold value corresponding to the target abnormal status label defined in advance, determine the third abnormal status chain prediction result as a trusted result, and use the abnormal status chain and the target abnormal status label represented in the third abnormal status chain prediction result as the abnormal status chain label data corresponding to the second sample production parameter status path data, so as to optimize the abnormal status chain prediction network based on the first abnormal status chain prediction result, the second abnormal status chain prediction result, the abnormal status chain label data corresponding to the first sample production parameter status path data, and the abnormal status chain label data corresponding to the second sample production parameter status path data;

[0011] Obtain target production parameter status path data, load the target production parameter status path data into the optimized abnormal status chain prediction network, and generate an abnormal status chain prediction result output by the abnormal status chain prediction network.

[0012] According to a second aspect of the present application, there is provided a power cord on-line production monitoring system, where the power cord on-line production monitoring system includes a processor and a readable storage medium, and the readable storage medium stores a program, and when the program is executed by the processor, the foregoing power cord on-line production monitoring method is implemented.

[0013] According to a third aspect of the present application, there is provided a computer-readable storage medium, where computer-executable instructions are stored in the computer-readable storage medium, and when it is monitored that the computer-executable instructions are executed, the foregoing power cord on-line production monitoring method is implemented.

[0014] According to any of the above aspects, the embodiments of the present application obtain the sample production parameter status path data with and without abnormal status chain tag data, and use these sample production parameter status path data to train and optimize the abnormal status chain prediction network. This can not only effectively identify the abnormal status chain in the production line, but also improve the accuracy and reliability of abnormal status prediction by introducing guidance for the machine learning network to generate confidence evaluation. The optimized abnormal status chain prediction network can accurately predict the abnormal status chain for the target production parameter status path data, thereby realizing real-time monitoring and fault warning of the power cord production process, and significantly improving production efficiency and product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] To more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required in the embodiments. It should be understood that the following drawings only show some embodiments of the present application, and thus should not be regarded as limiting the scope. For those of ordinary skill in the art, without creative efforts, other related drawings can be obtained based on these drawings.

[0016] Figure 1 FIG. shows a schematic flowchart of the on-line production monitoring method for power cords provided by the embodiments of the present application;

[0017] Figure 2 FIG. shows a schematic component structure diagram of the on-line production monitoring system for power cords for implementing the above on-line production monitoring method for power cords. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0018] To enable those skilled in the art to better understand the solutions of the present application, the following clearly and completely describes the technical solutions in the embodiments of the present application with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present application.

[0019] It should be noted that the terms "first", "second", etc. in the description, claims and the above-mentioned drawings of this application are used to distinguish similar objects, and do not necessarily have to be used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present application described here can be implemented in an order other than those illustrated or described here. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or are inherent to these processes, methods, products or devices.

[0020] Figure 1 The flowchart of the on-line production monitoring method for power cords provided by the embodiments of the present application is shown. It should be understood that in other embodiments, the order of some steps of the on-line production monitoring method for power cords can be interchanged according to actual needs, or some of the steps can also be omitted or deleted. The detailed steps of the on-line production monitoring method for power cords are introduced as follows.

[0021] Step S110, obtain the first sample production parameter status path data and the second sample production parameter status path data of each reference power cord production line, where the first sample production parameter status path data carries abnormal status chain label data, and the second sample production parameter status path data does not carry abnormal status chain label data.

[0022] In this embodiment, the server monitors the production conditions of multiple power cord production lines. In this large production workshop, there are many power cord production lines operating simultaneously.

[0023] Regarding the acquisition of the first sample production parameter status path data:

[0024] The server is connected to the production management systems and sensor networks of each power cord production line. For example, on production line A, sensors can detect parameters such as the feeding speed, temperature, pressure, current, etc. of raw materials. These parameters change continuously as the production process progresses, forming a time-series parameter status path. Suppose that within a certain period, product quality non-conformance occurs on production line A. After investigation by engineers, it is found that at a certain stage of the production process, the temperature is too high and the feeding speed is unstable. This series of abnormal states is marked as abnormal state chain tag data. This production parameter status path data with abnormal state chain tag data becomes the first sample production parameter status path data. Specifically, at a certain moment, the temperature sensor may detect that the temperature suddenly rises to 80 degrees Celsius, which is beyond the normal range (the normal range is 60 - 70 degrees Celsius). At the same time, the feeding speed sensor shows that the feeding speed suddenly drops from a stable 10 meters per minute to 5 meters per minute. This abnormal state chain is marked with the tag "too high temperature - unstable feeding speed", and this tag is associated with the entire production parameter status path data, such as the value change paths of temperature parameters, feeding speed parameters, and other relevant parameters such as current and pressure during this period. These data constitute the first sample production parameter status path data for production line A. The same situation also occurs on other production lines, such as production lines B, C, etc. The server collects the first sample production parameter status path data with abnormal state chain tag data from these production lines.

[0025] For obtaining the second sample production parameter status path data:

[0026] On other normally operating production lines in the same workshop, such as production line D. The server obtains production parameters from the production management system and sensor network of production line D. During the entire production process, there are no product quality problems or any known abnormal situations. The sensors continuously detect various parameters. For example, the temperature remains stable at around 65 degrees Celsius, the feeding speed remains at 10 meters per minute, and parameters such as current and pressure also fluctuate within the normal range. The status path data formed by these production parameters under normal operation over time is the second sample production parameter status path data. Since there are no abnormal situations, it does not carry abnormal state chain tag data. The server collects similar second sample production parameter status path data from multiple normally operating production lines.

[0027] Step S120, load the first sample production parameter status path data and the second sample production parameter status path data into an abnormal state chain prediction network serving as a supervised machine learning network respectively, to generate a first abnormal state chain prediction result for the first sample production parameter status path data and a second abnormal state chain prediction result for the second sample production parameter status path data.

[0028] In this embodiment, the server transmits the previously collected first sample production parameter status path data and the second sample production parameter status path data to the abnormal status chain prediction network. This abnormal status chain prediction network is a pre-designed machine learning model, possibly constructed based on a neural network architecture.

[0029] Taking the first sample production parameter status path data of production line A as an example, when it is loaded into the abnormal status chain prediction network:

[0030] The abnormal status chain prediction network first parses the first sample production parameter status path data to identify the values of each production parameter at different time points. Suppose this abnormal status chain prediction network has multiple hidden layers. The data first enters the input layer and then passes through the first hidden layer. In this layer, the neurons perform a weighted sum of the input data and perform a non-linear transformation through an activation function. For example, for the temperature parameter, the neurons may weight the temperature value according to the previously learned weights, and then combine it with the results of other parameters such as the feeding speed after being processed by an activation function (such as the ReLU function), and continue to pass to deeper hidden layers. As the data propagates in the network, the network analyzes this production parameter status path data with abnormal status chain labels according to the patterns learned during previous training. Since similar abnormal situations have been encountered during the previous training process, the network will predict results related to the previously labeled abnormal status chain label data based on the current input data features. For example, the network may predict which links in this production process are most likely to have abnormalities and give the probability of each abnormal link. This result is the first abnormal status chain prediction result.

[0031] For the second sample production parameter status path data of production line D:

[0032] Similarly, the data is loaded into the abnormal status chain prediction network. The network analyzes the data according to the same processing flow. Since this data is the parameter status path data under normal production conditions, during the analysis process, the network will judge whether the values of each production parameter are within the normal range and whether the entire production process conforms to the normal pattern. For example, the network will detect the stable states of parameters such as temperature, feeding speed, and current, and based on these normal features, predict that this production process is normal and there is no abnormal status chain. This result is the second abnormal status chain prediction result. At the same time, the network may also give some indicators of production stability, such as whether the fluctuation ranges of each parameter are within a very small normal interval, etc.

[0033] Step S130: Load the second sample production parameter status path data into the guiding machine learning network corresponding to the guided machine learning network, and generate a third abnormal status chain prediction result of the second sample production parameter status path data, where the third abnormal status chain prediction result represents the confidence levels of the abnormal status chain and the target abnormal status label.

[0034] In this embodiment, the server transmits the second sample production parameter status path data obtained from production line D to the guiding machine learning network corresponding to the abnormal status chain prediction network. This guiding machine learning network is also a complex model, and its structure and parameter settings are related to those of the abnormal status chain prediction network.

[0035] When the data enters the guiding machine learning network: Assume that this guiding machine learning network is a network constructed based on a probability model. The network first extracts features from the input production parameter status path data. For example, for the temperature parameter, it analyzes the fluctuation characteristics of the temperature during the entire production process, whether it has been stable around a certain value or has small periodic fluctuations. For the feed rate parameter, it pays attention to whether the feed rate remains constant or changes according to a certain pattern. Based on these feature extraction results, the network performs calculations in its internal probability model. Assume that there is a probability distribution model for the abnormal status chain in this network. For different target abnormal status labels, such as "abnormal temperature increase" and "abnormal decrease in feed rate", the network calculates the probability of each target abnormal status label appearing according to the characteristics of the current data. This probability is the confidence level representing the abnormal status chain and the target abnormal status label. For example, for the target abnormal status label "abnormal temperature increase", based on the production parameter status path data of production line D, the network finds that the temperature has been stable within the normal range, and after calculation, the probability of this target abnormal status label appearing is very low, such as only 0.01, which means that the network has a very low confidence in the abnormal status chain of "abnormal temperature increase" occurring in production line D. For the default target abnormal status label of "normal production", the probability calculated by the network may be very high, such as 0.99, indicating that the network is very confident that production line D is in a normal production state and there is no abnormal status chain. This result containing the confidence levels of each target abnormal status label is the third abnormal status chain prediction result.

[0036] In step S140, if the confidence level is greater than the threshold value predefined for the corresponding target abnormal state label, determine that the prediction result of the third abnormal state chain is a trusted result, and use the abnormal state chain and the target abnormal state label represented in the prediction result of the third abnormal state chain as the abnormal state chain label data corresponding to the second sample production parameter status path data, so as to optimize the abnormal state chain prediction network based on the prediction result of the first abnormal state chain, the prediction result of the second abnormal state chain, the abnormal state chain label data corresponding to the first sample production parameter status path data, and the abnormal state chain label data corresponding to the second sample production parameter status path data.

[0037] In this embodiment, a threshold value is predefined for each target abnormal state label in the server. For example, for the target abnormal state label of "abnormal temperature increase", the threshold value is set to 0.1.

[0038] For the case of production line D:

[0039] Previously in step S130, the guidance machine learning network calculated that the confidence level of the target abnormal state label of "abnormal temperature increase" was 0.01, and this confidence level was less than the threshold value of 0.1. Therefore, this result is not regarded as a trusted result, and no abnormal state chain and target abnormal state label will be used as the abnormal state chain label data corresponding to the second sample production parameter status path data of production line D.

[0040] However, assume that the situation of the second sample production parameter status path data of another production line E is different. After loading the data of production line E into the guidance machine learning network, for the target abnormal state label of "abnormal decrease in feeding speed", the confidence level calculated by the network is 0.2, and the threshold value corresponding to this target abnormal state label is set to 0.15. Since 0.2 is greater than 0.15, it is determined that the prediction result of this third abnormal state chain is a trusted result. At this time, the server will use the abnormal state chain of "abnormal decrease in feeding speed" and the target abnormal state label of "abnormal decrease in feeding speed" represented in the prediction result of this third abnormal state chain as the abnormal state chain label data corresponding to the second sample production parameter status path data of production line E.

[0041] Then, based on the abnormal state chain label data corresponding to the first sample production parameter status path data (such as the data of production line A with real abnormal state chain label data), the newly determined abnormal state chain label data of the second sample production parameter status path data of production line E, and the previously obtained first abnormal state chain prediction result (the prediction result for production line A, etc.) and the second abnormal state chain prediction result (the prediction result for production line D, etc.), optimize the abnormal state chain prediction network. For example, the server will use an algorithm based on error backpropagation to adjust the neuron weights in the abnormal state chain prediction network. If in the first sample production parameter status path data, the real abnormal state chain label data indicates that the feeding speed should abnormally decrease at a certain time point, and the previous first abnormal state chain prediction result fails to accurately predict this situation, then during the optimization process, the neuron weights related to the feeding speed parameter processing in the network will be adjusted according to this error, so that the network can predict more accurately when encountering a similar situation next time. The same applies to the second sample production parameter status path data. If there is a difference between the newly determined abnormal state chain label data and the previous second abnormal state chain prediction result, the network weights will also be adjusted according to this difference to improve the prediction accuracy of the network.

[0042] Step S150, obtain the target production parameter status path data, load the target production parameter status path data into the optimized abnormal state chain prediction network, and generate the abnormal state chain prediction result output by the abnormal state chain prediction network.

[0043] In this embodiment, after the optimization in the previous steps, continue to monitor the entire power cord production workshop. Now there is a new production line F. The server obtains the target production parameter status path data from production line F. This target production parameter status path data is similar to the first sample and second sample production parameter status path data obtained previously, and includes the numerical change paths of various production parameters (such as temperature, feeding speed, current, pressure, etc.) during the production process.

[0044] The server loads the target production parameter status path data into the anomaly status chain prediction network that has been optimized. The anomaly status chain prediction network processes the input target production parameter status path data according to the previously optimized neuron weights and structure. For example, when the target production parameter status path data enters the input layer of the anomaly status chain prediction network, the neurons in the network perform weighted summation on the target production parameter status path data according to their respective weights, and process it through an activation function. Then, the target production parameter status path data is passed between the hidden layers of the anomaly status chain prediction network. The anomaly status chain prediction network analyzes the production parameter status path data of production line F according to the patterns learned previously. Suppose that during the production process of production line F, the feeding speed fluctuates within a certain period, and the temperature also has a slightly rising trend. After analysis by the anomaly status chain prediction network, the anomaly status chain prediction network may predict an anomaly status chain such as "unstable feeding speed - rising temperature trend", and give the probability of each anomaly link. For example, the probability of unstable feeding speed is 0.6, and the probability of rising temperature trend is 0.4. The result including the predicted anomaly status chain and the probability of each link is the anomaly status chain prediction result output by the anomaly status chain prediction network. The server can take timely measures to avoid possible product quality problems or production line failures based on this result, such as adjusting the feeding equipment or checking the temperature control system, etc.

[0045] Based on the above steps, in the embodiment of the present application, by obtaining sample production parameter status path data with and without anomaly status chain label data, and using these sample production parameter status path data to train and optimize the anomaly status chain prediction network, it can not only effectively identify the anomaly status chain in the production line, but also improve the accuracy and reliability of anomaly status prediction by introducing a guiding machine learning network to generate confidence evaluation. The optimized anomaly status chain prediction network can accurately predict the anomaly status chain for the target production parameter status path data, thereby realizing real-time monitoring and fault warning of the power cord production process, and significantly improving production efficiency and product quality.

[0046] In a possible implementation manner, the second sample production parameter status path data is any one in the training data sequence composed of multiple sample production parameter status path data that do not carry anomaly status chain label data.

[0047] The method further includes:

[0048] Step A110, using the guiding machine learning network to obtain multiple anomaly status chain prediction results corresponding to the multiple sample production parameter status path data.

[0049] Step A120: Determine the confidence levels corresponding to the target abnormal state labels in the multiple abnormal state chain prediction results, and the number of times the target abnormal state labels are triggered in the multiple abnormal state chain prediction results.

[0050] Step A130: Based on the confidence levels corresponding to the target abnormal state labels in the multiple abnormal state chain prediction results, and the number of times the target abnormal state labels are triggered in the multiple abnormal state chain prediction results, determine the threshold value corresponding to the target abnormal state label.

[0051] In this embodiment, among the numerous normal operation data of the production line, the second sample production parameter state path data is selected from the training data sequence composed of multiple sample production parameter state path data that do not carry abnormal state chain labels. For example, in the production lines G, H, I, etc. that are operating normally, the server obtains their respective production parameter state path data, and these data form the training data sequence. Any one of them, such as the data of production line G, is used as the second sample production parameter state path data.

[0052] The server uses the guiding machine learning network to process these multiple sample production parameter state path data. Taking the data of production lines G, H, and I as an example, these data are respectively loaded into the guiding machine learning network. The network starts to analyze each data, just like processing a single second sample production parameter state path data, extracts the features of production parameters such as temperature, feed rate, current, pressure, etc., and then calculates according to the internal probability model or algorithm rules to obtain the abnormal state chain prediction results corresponding to each data. For example, for the data of production line G, the network may predict that there is a very small probability of an abnormal state chain of "slight fluctuation in feed rate" at a certain moment, but overall considers the production line normal; for the data of production line H, it may predict an abnormal state chain of "instantaneous slight increase in temperature" with a very low probability, etc.

[0053] After obtaining the multiple abnormal state chain prediction results, the server starts to process the target abnormal state label. For example, if the target abnormal state label is "abnormal fluctuation in feed rate", the server searches for the confidence level corresponding to this label in each prediction result. In the prediction result of production line G, the confidence level of "abnormal fluctuation in feed rate" may be 0.05, indicating that the network does not quite believe that production line G will have this abnormal state; in the prediction result of production line H, this confidence level may be 0.03, etc. At the same time, the server counts the number of times the target abnormal state label of "abnormal fluctuation in feed rate" is triggered in these prediction results. Suppose after analyzing the data of 10 production lines, only the prediction results of production line G and another production line J trigger this label, then the number of trigger times is 2 times.

[0054] Determine the threshold value based on the confidence levels and trigger counts corresponding to the obtained target abnormal status labels respectively. For the target abnormal status label of "abnormal fluctuation of feeding speed", if in the prediction results of all production lines, the confidence levels are generally low and the trigger counts are few, for example, the confidence levels are mostly between 0.01 and 0.05 and the trigger counts do not exceed 3 times, the server will set a relatively low threshold value, such as 0.08. This is because from the existing data, this abnormal status rarely appears and the network's confidence in judging it is not high, so when judging whether this abnormal status appears in new data subsequently, the threshold value is also relatively low. Conversely, if there are higher confidence values and more trigger counts, a higher threshold value will be set.

[0055] In a possible implementation manner, the first sample production parameter status path data is any one of the training data sequences composed of multiple sample production parameter status path data all carrying abnormal status chain label data, and the abnormal status chain label data includes an abnormal status label and an abnormal status chain.

[0056] Step S140 includes:

[0057] Step S141, if the confidence level is greater than the threshold value corresponding to the predefined target abnormal status label, based on the abnormal status chain corresponding to the target abnormal status label in the second sample production parameter status path data, extract the corresponding first abnormal status chain path data from the second sample production parameter status path data.

[0058] Step S142, load the first abnormal status chain path data into a set graph autoencoder to generate a first graph autoencoding vector.

[0059] Step S143, obtain one or more sample production parameter status path data corresponding to the target abnormal status label in the training data sequence.

[0060] Step S144, based on the abnormal status chains respectively corresponding to the target abnormal status label in the one or more sample production parameter status path data, respectively extract the corresponding one or more second abnormal status chain path data from the one or more sample production parameter status path data.

[0061] Step S145, load the one or more second abnormal status chain path data into the graph autoencoder respectively to generate one or more second graph autoencoding vectors.

[0062] Step S146, determine the target second graph autoencoding vector based on the one or more second graph autoencoding vectors.

[0063] In step S147, if the feature distance between the first graph auto-encoding vector and the target second graph auto-encoding vector is less than the set distance, then determine that the prediction result of the third abnormal state chain is a trusted result.

[0064] In this embodiment, when the server manages the data of the power cord production line, the first sample production parameter status path data is selected from a training data sequence composed of multiple sample production parameter status path data all carrying abnormal state chain label data (including abnormal state labels and abnormal state chains). For example, the data of production line K is selected as the first sample production parameter status path data from the production line data of production lines K, L, M, etc. with abnormal situation records.

[0065] When it is found that the confidence level of the target abnormal state label is greater than the predefined threshold, subsequent operations are started. Assume that the target abnormal state label is "too high temperature and abnormally reduced feeding speed", and the confidence level exceeds the corresponding threshold. The server will extract the corresponding first abnormal state chain path data from the data of production line N based on the abnormal state chain corresponding to this target abnormal state label in the second sample production parameter status path data (such as the data from production line N). This process is like screening from the complete production parameter status path data of production line N according to the part related to "too high temperature and abnormally reduced feeding speed" and extracting this part of the data. For example, the numerical change path data of the temperature, feeding speed, and related current, pressure, etc. parameters corresponding to the suspected abnormal state are extracted.

[0066] Then, the extracted first abnormal state chain path data is loaded into the set graph auto-encoder. The graph auto-encoder processes this data and converts it into a specific representation form, that is, generates the first graph auto-encoding vector. This vector contains the feature information of the first abnormal state chain path data under the encoding of the graph auto-encoder.

[0067] Next, the server obtains one or more sample production parameter status path data corresponding to the target abnormal state label of "too high temperature and abnormally reduced feeding speed" in the training data sequence. For example, from the previously collected data of production lines K, L, M, etc. with abnormal situations, find the data related to this target abnormal state label. Based on the abnormal state chain corresponding to this target abnormal state label in these data (such as the data of production line K), extract the corresponding second abnormal state chain path data from the data of production line K. This data is also the part of the data related to "too high temperature and abnormally reduced feeding speed", such as the parameter change path of temperature, feeding speed, etc. within a specific time period. Similarly, the data of other relevant production lines (such as production lines L, M) are also operated in this way to obtain one or more second abnormal state chain path data.

[0068] Load these one or more second abnormal state chain path data into the graph autoencoder respectively. The graph autoencoder will generate a corresponding second graph autoencoded vector for each data. These second graph autoencoded vectors all contain the feature information of their respective second abnormal state chain path data.

[0069] After that, determine the target second graph autoencoded vector based on these one or more second graph autoencoded vectors. For example, by comparing a certain similarity between these vectors or according to a preset rule, select the most representative one from these vectors as the target second graph autoencoded vector.

[0070] Finally, if the feature distance between the first graph autoencoded vector and the target second graph autoencoded vector is less than the set distance, determine that the prediction result of the third abnormal state chain is a trusted result. Here, the feature distance is calculated by a certain distance metric method (such as Euclidean distance, etc.). If this distance is less than the set distance, it means that the first abnormal state chain path data extracted from production line N is very similar in features to the representative second abnormal state chain path data extracted from the training data sequence (such as production lines K, L, M, etc.). Then, the prediction result of the third abnormal state chain can be trusted, that is, it is considered that there may indeed be an abnormality related to the target abnormal state label in production line N.

[0071] In a possible implementation manner, step S140 further includes:

[0072] Determine the first training error parameter based on the prediction result of the first abnormal state chain and the abnormal state chain label data corresponding to the first sample production parameter state path data.

[0073] Determine the second training error parameter based on the prediction result of the second abnormal state chain and the abnormal state chain label data corresponding to the second sample production parameter state path data.

[0074] Determine the global training error parameter based on the first training error parameter and the second training error parameter, so as to optimize the neuron weight information of the abnormal state chain prediction network based on the global training error parameter. Among them, based on the neuron weight information of the abnormal state chain prediction network, optimize the neuron weight information of the guiding machine learning network.

[0075] In this embodiment, when processing the power line production line data, in order to optimize the abnormal state chain prediction network, operations need to be performed based on the relevant results obtained previously.

[0076] First, determine the first training error parameter based on the first abnormal state chain prediction result and the abnormal state chain label data corresponding to the first sample production parameter state path data. For example, in the case of production line K, the abnormal state chain label data marked in the first sample production parameter state path data indicates that there is a situation of "too high temperature and abnormally reduced feeding speed" within a certain time period, and details such as the specific numerical range or time node corresponding to each link in this abnormal state chain are clearly defined. The first abnormal state chain prediction result may have predicted this abnormal state chain. In the prediction result, the probability of predicting the "too high temperature" link may be 0.8, and the probability of predicting the "abnormally reduced feeding speed" link may be 0.7. However, in the actual abnormal state chain label data, these two links are bound to occur (probability is 1). Determine the first training error parameter by calculating the difference between the prediction result and the actual label data. For example, the mean square error (MSE) calculation method can be used. Square the difference between the predicted probability and the actual probability (1) of the "too high temperature" link, add the square of the difference between the predicted probability and the actual probability (1) of the "abnormally reduced feeding speed" link to obtain a part of the first training error parameter, and then comprehensively calculate the errors of other relevant parameters to finally obtain the first training error parameter.

[0077] Next, determine the second training error parameter based on the second abnormal state chain prediction result and the abnormal state chain label data corresponding to the second sample production parameter state path data. Taking the data of production line N as an example, the second sample production parameter state path data has been assigned abnormal state chain label data in the previous operation (assuming that it is determined to be abnormal and marked after the previous steps), such as being marked as an abnormal state chain that may have "slight temperature fluctuations and slightly unstable feeding speed". The prediction of this abnormal state chain by the second abnormal state chain prediction result may be that the probability of "slight temperature fluctuations" is 0.1, and the probability of "slightly unstable feeding speed" is 0.2. However, in the actual marked abnormal state chain label data, the actual probability of "slight temperature fluctuations" may be 0.3, and the actual probability of "slightly unstable feeding speed" is 0.4. Similarly, use a similar mean square error calculation method to calculate the sum of the squared differences between the predicted probability and the actual probability, thereby obtaining the second training error parameter.

[0078] Finally, determine the global training error parameter based on the first training error parameter and the second training error parameter. For example, the first training error parameter and the second training error parameter can be simply added to obtain the global training error parameter, or different weights can be assigned according to the importance of these two error parameters during the entire training process, and then the weighted sum is used to obtain the global training error parameter. After obtaining the global training error parameter, the neuron weight information of the abnormal state chain prediction network can be optimized based on this parameter. The server can adopt an optimization algorithm based on gradient descent, calculate the gradient of each neuron weight according to the global training error parameter, and then update the weight at a certain learning rate, so that the network can more accurately predict the abnormal state chain during the next prediction. For example, if a certain neuron is related to the processing of the "temperature" parameter and a large prediction error related to temperature is found when calculating the global training error parameter, then the weight of this neuron will be adjusted in the direction of reducing this error when updating the weight, thereby improving the accuracy of the entire abnormal state chain prediction network for predicting the abnormal state chain.

[0079] In a possible implementation manner, the step of loading the first sample production parameter state path data and the second sample production parameter state path data into the abnormal state chain prediction network serving as the machine learning network to be guided includes:

[0080] Perform shallow path derivative reinforcement processing on the first sample production parameter state path data and the second sample production parameter state path data respectively, and load the first sample production parameter state path data after path derivative reinforcement processing and the second sample production parameter state path data after path derivative reinforcement processing into the abnormal state chain prediction network serving as the machine learning network to be guided respectively.

[0081] In this embodiment, when processing the data of the power line production line, for the operation of loading the first sample production parameter state path data and the second sample production parameter state path data into the abnormal state chain prediction network, shallow path derivative reinforcement processing needs to be performed first.

[0082] Taking the first sample production parameter status path data of production line K as an example, this data contains the numerical change paths of various production parameters such as temperature, feeding speed, current, and pressure during the production process, and is accompanied by abnormal status chain label data. When performing shallow path derivative enhancement processing, the server will perform some simple feature derivative operations on these production parameters. For the temperature parameter, the server may calculate the change rate of temperature within a certain time window. For example, the average change rate of temperature is calculated every 5 minutes to obtain a new sequence of temperature change rates, which is the shallow derivative feature of the temperature parameter. For the feeding speed, the server can calculate the difference in feeding speed between adjacent time periods to form a sequence of feeding speed differences. Similarly, similar operations are performed on the current and pressure parameters, such as calculating the fluctuation amplitude of the current and the relatively stable value of the pressure and other derivative features. Combining these derivative features with the original production parameter values completes the shallow path derivative enhancement processing of the first sample production parameter status path data of production line K.

[0083] For the second sample production parameter status path data of production line N, which is data under normal operating conditions and does not have abnormal status chain label data. When performing shallow path derivative enhancement processing, similar operations are also adopted. For the temperature parameter, calculate the average temperature change rate every 10 minutes to obtain a sequence of temperature change rates; for the feeding speed, calculate the difference between the maximum and minimum values of the feeding speed within a certain time interval to form a feeding speed difference feature. Relevant feature derivatives are also performed on the current and pressure parameters, such as calculating the average fluctuation period of the current and the maximum stable duration of the pressure. In this way, the second sample production parameter status path data after shallow path derivative enhancement processing is obtained.

[0084] Then, the server loads the first sample production parameter status path data (such as the data of production line K) after shallow path derivative enhancement processing and the second sample production parameter status path data (such as the data of production line N) after path derivative enhancement processing into the abnormal status chain prediction network that serves as the guided machine learning network respectively. After receiving this data, the network can use the original production parameter values and the newly derived features contained therein to perform prediction analysis of the abnormal status chain. For example, the neurons in the network can more comprehensively analyze the data based on features such as the temperature change rate and the feeding speed difference, combined with the original temperature and feeding speed values, to determine whether there is an abnormal status chain and its possible situation.

[0085] In a possible implementation manner, loading the first sample production parameter status path data and the second sample production parameter status path data into an abnormal status chain prediction network serving as a supervised machine learning network respectively to generate a first abnormal status chain prediction result of the first sample production parameter status path data and a second abnormal status chain prediction result of the second sample production parameter status path data includes:

[0086] Performing shallow path derivative reinforcement processing on the first sample production parameter status path data to generate third sample production parameter status path data, and the third sample production parameter status path data carries abnormal status chain label data corresponding to the first sample production parameter status path data.

[0087] Performing shallow path derivative reinforcement processing on the second sample production parameter status path data to generate fourth sample production parameter status path data.

[0088] Performing deep path derivative reinforcement processing on the second sample production parameter status path data to generate fifth sample production parameter status path data.

[0089] Performing self-attention processing on the fifth sample production parameter status path data to generate sixth sample production parameter status path data.

[0090] Loading the third sample production parameter status path data, the fourth sample production parameter status path data, the fifth sample production parameter status path data, and the sixth sample production parameter status path data into an abnormal status chain prediction network serving as a supervised machine learning network respectively to generate a first abnormal status chain prediction result of the third sample production parameter status path data, a second abnormal status chain prediction result of the fourth sample production parameter status path data, status path segments of multiple status spans of the fourth sample production parameter status path data, status path segments of multiple status spans of the fifth sample production parameter status path data, and status path segments of multiple status spans of the sixth sample production parameter status path data.

[0091] In this embodiment, when processing the power line production line data, the first sample production parameter status path data and the second sample production parameter status path data need to be loaded into the abnormal status chain prediction network and relevant results are generated.

[0092] For the first sample production parameter state path data, take the data of production line K as an example. First, perform shallow path derivation and enhancement processing on it to generate the third sample production parameter state path data and carry the original abnormal state chain label data. In the shallow path derivation and enhancement process, for the temperature parameters of production line K, in addition to the original temperature value, the server calculates the slope of the temperature in each production stage, such as the slope of the temperature rise in the initial feeding stage, the slope of the temperature stability in the intermediate processing stage is 0, and the slope of the temperature drop in the final stage; for the feed speed, calculate the change ratio of the feed speed in adjacent time periods, such as the feed speed increased by 20% from the first minute to the second minute. At the same time, similar operations are performed on parameters such as current and pressure, such as calculating the fluctuation frequency of the current and the rate of change of pressure in different stages. Integrating these newly derived data with the original data will obtain the third sample production parameter state path data.

[0093] For the second sample production parameter state path data, such as the data of production line N. First, perform shallow path derivative strengthening processing to generate the fourth sample production parameter state path data. For temperature parameters, calculate the average temperature change in each short period of time, such as how many degrees the temperature rises or falls on average every 3 minutes; for feed speed, calculate the maximum change in feed speed in each time period, for example, the maximum change in feed speed in a certain 5 minutes is 5 meters / minute; for current, calculate the difference between the average current and the standard current, and for pressure, calculate the fluctuation range, etc.

[0094] Then, the second sample production parameter state path data of production line N is subjected to deep path derivation and enhancement processing to generate the fifth sample production parameter state path data. This process will dig deeper relationships. For example, the server will analyze the delay relationship between temperature change and feed speed change, assuming how long the feed speed will change accordingly after the temperature rises, or whether there is a periodic correlation between current change and pressure change. Through complex algorithms and models, these hidden relationships are found and new data features are generated and added to the fifth sample production parameter state path data.

[0095] Then, the fifth sample production parameter state path data is processed by self-attention to generate the sixth sample production parameter state path data. The self-attention mechanism allows each element in the data to pay attention to the importance of other related elements. For example, among the parameters such as temperature, feed rate, current, and pressure, the self-attention process may determine that the temperature has a greater impact on the entire production state at a certain production stage, and then the data representation will be adjusted accordingly to highlight the importance of the temperature parameter at this stage. These adjusted results are integrated to obtain the sixth sample production parameter state path data.

[0096] Finally, the server loads the third sample production parameter status path data (data processed by production line K), the fourth sample production parameter status path data (data shallowly processed by production line N), the fifth sample production parameter status path data (data deeply processed by production line N), and the sixth sample production parameter status path data (data self-attention processed by production line N) into the abnormal state chain prediction network respectively.

[0097] For the third sample production parameter status path data, the network analyzes it based on the original data and derivative data contained therein, as well as the abnormal state chain label data carried. For example, based on data such as the temperature change slope and the feed rate change ratio of production line K, combined with known abnormal state chain labels (such as too high temperature and abnormally reduced feed rate), it predicts in which production stages this abnormal state chain is most likely to occur, and information such as the severity of the abnormal state in each stage. This is the first abnormal state chain prediction result.

[0098] For the fourth sample production parameter status path data, the network analyzes the derivative features of various production parameters therein, determines whether production line N is in a normal production state, and predicts whether each production parameter fluctuates within a normal range. This is the second abnormal state chain prediction result. At the same time, the network will divide the fourth sample production parameter status path data according to different production stages or different time spans, obtaining status path segments of multiple status spans. For example, the entire production process is divided into spans of every 10 minutes, obtaining status path segments of parameters such as temperature and feed rate within each span.

[0099] For the fifth sample production parameter status path data, the network will also analyze it in a similar manner to obtain status path segments of multiple status spans. Due to the deep path derivative strengthening process, these status path segments contain more information about deep-level relationships. For example, the influence of the correlation between temperature and feed rate on the production state within a certain status span.

[0100] For the sixth sample production parameter status path data, the network analyzes the important information highlighted after self-attention processing therein, obtaining status path segments of multiple status spans. The weights of each parameter in these status path segments are adjusted by self-attention, which can better reflect the key information in the production process. For example, in a certain key production stage, the temperature parameter is highlighted due to the self-attention mechanism, and the slightest change in temperature in the corresponding status path segment can be accurately analyzed, thus helping to more precisely determine whether the production state is normal or whether there is an abnormal state chain.

[0101] In a possible implementation manner, step S130 includes:

[0102] Load the fourth sample production parameter status path data into the guiding machine learning network corresponding to the guided machine learning network to generate a third abnormal state chain prediction result for the fourth sample production parameter status path data.

[0103] Regarding the abnormal state chain and the target abnormal state label characterized in the third abnormal state chain prediction result as the abnormal state chain label data corresponding to the second sample production parameter status path data, and based on the first abnormal state chain prediction result, the second abnormal state chain prediction result, the abnormal state chain label data corresponding to the first sample production parameter status path data, and the abnormal state chain label data corresponding to the second sample production parameter status path data, optimizing the abnormal state chain prediction network includes:

[0104] Regarding the abnormal state chain and the target abnormal state label characterized in the third abnormal state chain prediction result as the abnormal state chain label data corresponding to the fourth sample production parameter status path data, and based on the first abnormal state chain prediction result and the corresponding abnormal state chain label data of the third sample production parameter status path data, the second abnormal state chain prediction result and the corresponding abnormal state chain label data of the fourth sample production parameter status path data, as well as the state path segments of multiple state spans of the fourth sample production parameter status path data, the state path segments of multiple state spans of the fifth sample production parameter status path data, and the state path segments of multiple state spans of the sixth sample production parameter status path data, optimize the abnormal state chain prediction network.

[0105] In this embodiment, when loading the second sample production parameter status path data into the guided machine learning network to generate the prediction result of the third abnormal status chain, the specific operation is as follows. Taking the previously mentioned production line N as an example, the server loads the fourth sample production parameter status path data obtained through shallow path derivation and enhancement processing into the guided machine learning network corresponding to the guided machine learning network. This guided machine learning network has its own algorithm and model structure. When the fourth sample production parameter status path data enters the network, the network will analyze various production parameter features therein. For example, for the temperature parameter, it will check information such as the average change amplitude and the maximum change amount of the temperature at different stages; for the feeding speed, it will pay attention to the maximum change amount of the feeding speed and the change trend at different time periods, etc. Based on these features, the network predicts the possible abnormal status chain according to the internally set rules and the previously learned patterns. For example, the network may predict an abnormal status chain such as "abnormal fluctuation of feeding speed - unstable temperature" based on the sudden increase in the feeding speed within a certain time period and the small fluctuation of the temperature, and give corresponding confidence levels for each possible abnormal status (such as abnormal fluctuation of feeding speed, unstable temperature). The result including the abnormal status chain and the confidence levels of each status is the prediction result of the third abnormal status chain of the fourth sample production parameter status path data.

[0106] Next, when using the abnormal state chain characterized in the third abnormal state chain prediction result and the target abnormal state label as the abnormal state chain label data corresponding to the fourth sample production parameter state path data to further optimize the abnormal state chain prediction network. The server regards the abnormal state chain of "abnormal fluctuation of feeding speed - unstable temperature" mentioned in the previously obtained third abnormal state chain prediction result and the target abnormal state labels therein (such as abnormal fluctuation of feeding speed, unstable temperature) as the new abnormal state chain label data of the fourth sample production parameter state path data of production line N. Then, based on multiple results, the abnormal state chain prediction network is optimized. First, it is the first abnormal state chain prediction result of the third sample production parameter state path data (from production line K) and its corresponding abnormal state chain label data. For example, the first abnormal state chain prediction result of production line K shows that there is a situation of "too high temperature and abnormal decrease in feeding speed" at a certain stage and gives the prediction situation of each link, while the corresponding abnormal state chain label data clarifies that this abnormal state chain actually exists. There is a certain difference between the two, and this difference can be used to adjust the network. Second, it is the second abnormal state chain prediction result of the fourth sample production parameter state path data (production line N) and its newly determined abnormal state chain label data. For example, the second abnormal state chain prediction result may originally think that production line N is in normal production, but the new abnormal state chain label data indicates that there is a situation of "abnormal fluctuation of feeding speed - unstable temperature". The difference between them is also the basis for optimizing the network. In addition, there are also state path segments of multiple state spans of the fourth sample production parameter state path data. For example, after dividing the production process of production line N into spans of every 10 minutes, the state path segments of parameters such as temperature and feeding speed within each span. The parameter change situations and mutual relationships in these segments can provide more detailed information about the production state. The state path segments of multiple state spans of the fifth sample production parameter state path data (production line N after deep path derivative strengthening processing) contain deeper production parameter relationships, such as the impact of the deep-level association between temperature and feeding speed within a certain span on the production state. The state path segments of multiple state spans of the sixth sample production parameter state path data (production line N after self-attention processing) highlight the important parameters in the key production stages. For example, at a certain key stage, the temperature parameter is focused on due to the self-attention mechanism, and its change situation in the state path segment can better reflect the production state. The server synthesizes this information and uses a suitable algorithm (such as an algorithm based on error backpropagation) to adjust parameters such as the neuron weights in the abnormal state chain prediction network.For example, if there is a deviation between the prediction of excessive temperature in the third sample production parameter status path data and the actual label data, and it is found that the relationship between temperature and feed rate in a state path segment within a certain state span of the fourth sample production parameter status path data does not match the prediction, the server will adjust the neuron weights related to temperature and feed rate processing in the network based on these situations, thereby optimizing the abnormal state chain prediction network to enable it to more accurately judge the abnormal state chain in subsequent predictions.

[0107] In a possible implementation manner, based on the first abnormal state chain prediction result of the third sample production parameter status path data and the corresponding abnormal state chain label data, the second abnormal state chain prediction result of the fourth sample production parameter status path data and the corresponding abnormal state chain label data, as well as the state path segments of multiple state spans of the fourth sample production parameter status path data, the state path segments of multiple state spans of the fifth sample production parameter status path data, and the state path segments of multiple state spans of the sixth sample production parameter status path data, optimizing the abnormal state chain prediction network includes:

[0108] Based on the first abnormal state chain prediction result of the third sample production parameter status path data and the corresponding abnormal state chain label data, determine a first training error parameter.

[0109] Based on the second abnormal state chain prediction result of the fourth sample production parameter status path data and the corresponding abnormal state chain label data, determine a second training error parameter.

[0110] For the state path segments of multiple state spans corresponding to the fourth sample production parameter status path data and the fifth sample production parameter status path data respectively, determine the fusion calculation result of the feature distance training error parameters between the state path segments of the same state span as the third training error parameter.

[0111] For the state path segments of multiple state spans corresponding to the fifth sample production parameter status path data and the sixth sample production parameter status path data respectively, determine the fusion calculation result of the feature distance training error parameters between the state path segments of the same state span as the fourth training error parameter.

[0112] Based on the first training error parameter, the second training error parameter, the third training error parameter, and the fourth training error parameter, determine a global training error parameter, so as to optimize the neuron weight information of the abnormal state chain prediction network based on the global training error parameter.

[0113] In this embodiment, first, a first training error parameter is determined based on the first abnormal state chain prediction result of the third sample production parameter status path data and the corresponding abnormal state chain label data. Taking the case of production line K as an example, the abnormal state chain label data corresponding to the third sample production parameter status path data indicates that there is an abnormal state chain of "excessive temperature and abnormally reduced feeding speed" at a certain specific stage, and the accurate situation of each state is clarified. The first abnormal state chain prediction result predicts this abnormal state chain. Suppose the predicted probability of "excessive temperature" in the prediction is 0.8, while in the label data this state is bound to occur (probability is 1), and the predicted probability of "abnormally reduced feeding speed" is 0.7, with the actual probability being 1. The first training error parameter is determined by calculating the difference between the prediction result and the label data. The mean square error (MSE) calculation method can be used. Calculate the square of the difference between the predicted probability and the actual probability in the "excessive temperature" state, and then add the square of the difference between the predicted probability and the actual probability in the "abnormally reduced feeding speed" state to obtain a part of the first training error parameter. After comprehensively considering the errors of other relevant states, the first training error parameter is finally determined.

[0114] Next, a second training error parameter is determined based on the second abnormal state chain prediction result of the fourth sample production parameter status path data and the corresponding abnormal state chain label data. For the fourth sample production parameter status path data of production line N, assume that the newly determined abnormal state chain label data indicates the situation of "slightly abnormal fluctuation of feeding speed and slightly unstable temperature". In the second abnormal state chain prediction result, the predicted probability of "slightly abnormal fluctuation of feeding speed" is 0.1, and the actual probability may be 0.3. The predicted probability of "slightly unstable temperature" is 0.2, and the actual probability is 0.4. Similarly, using the mean square error calculation method, calculate the sum of the squares of the differences between the predicted probability and the actual probability in these two states, and then consider other relevant states to determine the second training error parameter.

[0115] Then, for the state path segments of multiple state spans corresponding to the fourth sample production parameter status path data and the fifth sample production parameter status path data respectively, determine the third training error parameter. For example, divide the production process of production line N into state spans of every 10 minutes to obtain their respective state path segments. For a certain same state span, such as the state span of the 3rd 10 minutes, in the state path segment of the fourth sample production parameter status path data, the numerical change of the temperature parameter, the numerical value of the feeding speed, and their simple relationship are one situation. In the state path segment of the fifth sample production parameter status path data, due to the deep path derivative reinforcement processing, in addition to the numerical changes of the temperature and the feeding speed, it also includes more features such as the deeper correlation relationship between the temperature and the feeding speed. By calculating the feature distance between these two state path segments, for example, using the Euclidean distance or other appropriate distance measurement methods, calculate the distance of the numerical change of the temperature parameter, the distance of the numerical change of the feeding speed, and the distance between other features, and then fuse and calculate these distances according to a certain weight (this weight can be determined according to the actual situation or experience) to obtain the feature distance training error parameter under this state span. Perform such calculations for all the same state spans, and then fuse these results as the third training error parameter.

[0116] The process of determining the fourth training error parameter for the state path segments of multiple state spans corresponding to the fifth sample production parameter status path data and the sixth sample production parameter status path data respectively is similar. For example, in the state span of the 5th 10 minutes, the state path segment of the fifth sample production parameter status path data contains the production parameter relationship after deep path derivative reinforcement processing, and the state path segment of the sixth sample production parameter status path data contains the important parameter relationship highlighted after self-attention processing. Calculate the feature distance between these two state path segments, such as the difference in the importance change of the temperature after self-attention processing and its relationship under deep path derivative reinforcement processing, the feature difference of the feeding speed in the two cases, etc., fuse the calculation results of these distances according to a certain weight, perform such operations for all the same state spans, and then fuse to obtain the fourth training error parameter.

[0117] Finally, determine the global training error parameter based on the first training error parameter, the second training error parameter, the third training error parameter, and the fourth training error parameter. For example, the global training error parameter can be simply obtained by adding these four training error parameters, or different weights can be assigned according to their importance during the entire training process, and then the weighted sum is calculated to obtain the global training error parameter. After obtaining the global training error parameter, the server uses this parameter to optimize the neuron weight information of the abnormal state chain prediction network. An optimization algorithm based on gradient descent can be adopted to calculate the gradient of each neuron weight according to the global training error parameter, and then update the weights at a certain learning rate, so that the network can more accurately predict the abnormal state chain in the next prediction. For example, if a certain neuron is related to the processing of temperature parameters and a relatively large error related to temperature is found when calculating the global training error parameter, then when updating the weights, the weights of this neuron will be adjusted in the direction of reducing this error, thereby improving the accuracy of the entire abnormal state chain prediction network in predicting the abnormal state chain.

[0118] Furthermore, Figure 2 shows a schematic hardware structure diagram of a power line on-line production monitoring system 100 for implementing the method provided in the embodiments of the present application. As Figure 2 shown, the power line on-line production monitoring system 100 may include at least one processor 102 (the processor 102 may include, but is not limited to, a processing device such as a microprocessor MCU or a programmable logic device FPGA), a memory 104 for storing data, and a transmission device 106 and a controller 108 for communication functions. Those of ordinary skill in the art can understand that Figure 2 the structure shown is only schematic and does not limit the structure of the power line on-line production monitoring system 100. For example, the power line on-line production monitoring system 100 may also include more or fewer components than Figure 2 shown, or have a different configuration from Figure 2 shown.

[0119] The memory 104 can be used to store software programs and modules of application software, such as the program instructions corresponding to the method embodiments described above in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, to implement the above-mentioned on-line production monitoring method for power lines. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some instances, the memory 104 may further include a memory remotely disposed relative to the processor 102, and these remote memories can be connected to the on-line production monitoring system 100 for power lines through a network. Examples of the above-mentioned network include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof.

[0120] The transmission device 106 is used to obtain or send data via a network. Specific examples of the above-mentioned network may include a wireless network provided by a communication provider of the on-line production monitoring system 100 for power lines. In one instance, the transmission device 106 includes a network adapter, which can be connected to other network devices through a base station and thus can communicate with the Internet. In one instance, the transmission device 106 can be a radio frequency module, which is used to communicate with the Internet wirelessly.

[0121] It should be noted that: the above-mentioned sequence of the embodiments of the present application is only for description and does not represent the superiority or inferiority of the embodiments. And the above-mentioned specific embodiments of the present application have been described. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain implementations, multitasking and parallel processing are also possible or may be advantageous.

[0122] The various embodiments in the embodiments of the present application are all described in a progressive manner. For the parts that are consistent and similar among the various embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the above different embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and reference can be made to the partial description of the method embodiments for the relevant parts.

[0123] Those of ordinary skill in the art can understand that all or part of the steps to implement the above embodiments can be completed by hardware, or can be completed by a program instructing relevant hardware. The above program can be stored in a computer-readable storage medium, and the above-mentioned storage medium can be a read-only memory, a magnetic disk, an optical disk, or the like.

Claims

1. A method for online production monitoring of power lines, characterized in that: The method comprises: Acquire first sample production parameter state path data and second sample production parameter state path data of each reference power cord production line, wherein the first sample production parameter state path data carries abnormal state chain label data, and the second sample production parameter state path data does not carry abnormal state chain label data; The first sample production parameter state path data and the second sample production parameter state path data are respectively loaded into an abnormal state chain prediction network as a guided machine learning network, and a first abnormal state chain prediction result of the first sample production parameter state path data and a second abnormal state chain prediction result of the second sample production parameter state path data are generated; Loading the second sample production parameter state path data into the guiding machine learning network corresponding to the guided machine learning network, generating a third abnormal state chain prediction result of the second sample production parameter state path data, wherein the third abnormal state chain prediction result represents the confidence of the abnormal state chain and the target abnormal state label; If the confidence is greater than a predefined threshold value corresponding to the target abnormal state label, the third abnormal state chain prediction result is determined to be a trust result, and the abnormal state chain and the target abnormal state label represented in the third abnormal state chain prediction result are used as the abnormal state chain label data corresponding to the second sample production parameter state path data, so as to optimize the abnormal state chain prediction network based on the first abnormal state chain prediction result, the second abnormal state chain prediction result, the abnormal state chain label data corresponding to the first sample production parameter state path data, and the abnormal state chain label data corresponding to the second sample production parameter state path data; Obtain target production parameter state path data, load the target production parameter state path data into the optimized abnormal state chain prediction network, and generate an abnormal state chain prediction result output by the abnormal state chain prediction network.

2. The method for online production monitoring of power lines according to claim 1, characterized in that: The second sample production parameter state path data is any one of a training data sequence consisting of a plurality of sample production parameter state path data that do not carry abnormal state chain label data; The method further comprises: Using the guiding machine learning network to obtain a plurality of abnormal state chain prediction results corresponding to the plurality of sample production parameter state path data; Determine the confidences respectively corresponding to the target abnormal state labels in the multiple abnormal state chain prediction results, and the number of triggering times of triggering the target abnormal state labels in the multiple abnormal state chain prediction results; The threshold value corresponding to the target abnormal state label is determined based on the confidences respectively corresponding to the target abnormal state labels in the multiple abnormal state chain prediction results and the number of triggering times of triggering the target abnormal state label in the multiple abnormal state chain prediction results.

3. The method for online production monitoring of power lines according to claim 1, characterized in that: The first sample production parameter state path data is any one of a training data sequence consisting of a plurality of sample production parameter state path data each carrying abnormal state chain label data, wherein the abnormal state chain label data includes an abnormal state label and an abnormal state chain; If the confidence is greater than a predefined threshold value corresponding to the target abnormal state label, determining that the third abnormal state chain prediction result is a trust result includes: If the confidence is greater than a predefined threshold value corresponding to the target abnormal state label, extracting corresponding first abnormal state chain path data from the second sample production parameter state path data based on the abnormal state chain corresponding to the target abnormal state label in the second sample production parameter state path data; Loading the first abnormal state chain path data into a set graph autoencoder to generate a first graph autoencoder vector; Acquire one or more sample production parameter state path data corresponding to the target abnormal state label in the training data sequence; Based on the abnormal state chains respectively corresponding to the target abnormal state tags in the one or more sample production parameter state path data, extracting corresponding one or more second abnormal state chain path data from the one or more sample production parameter state path data respectively; Loading the one or more second abnormal state chain path data into the graph autoencoder respectively to generate one or more second graph autoencoder vectors; determining a target second graph autoencoder vector based on the one or more second graph autoencoder vectors; If the characteristic distance between the first image auto-encoder vector and the target second image auto-encoder vector is less than a set distance, the third abnormal state chain prediction result is determined to be a trust result.

4. The method for online production monitoring of power lines according to claim 1, characterized in that: The optimizing the abnormal state chain prediction network based on the first abnormal state chain prediction result, the second abnormal state chain prediction result, the abnormal state chain label data corresponding to the first sample production parameter state path data, and the abnormal state chain label data corresponding to the second sample production parameter state path data includes: Determining a first training error parameter based on the first abnormal state chain prediction result and the abnormal state chain label data corresponding to the first sample production parameter state path data; Determining a second training error parameter based on the second abnormal state chain prediction result and the abnormal state chain label data corresponding to the second sample production parameter state path data; A global training error parameter is determined based on the first training error parameter and the second training error parameter to optimize neuron weight information of the abnormal state chain prediction network based on the global training error parameter.

5. The method for online production monitoring of power lines according to claim 4, characterized in that: The method further comprises: Based on the neuron weight information of the abnormal state chain prediction network, the neuron weight information of the guidance machine learning network is optimized.

6. The method for online production monitoring of power lines according to claim 1, characterized in that: The step of loading the first sample production parameter state path data and the second sample production parameter state path data into the abnormal state chain prediction network as the guided machine learning network respectively includes: The first sample production parameter state path data and the second sample production parameter state path data are respectively subjected to shallow path derivation reinforcement processing, and the first sample production parameter state path data after the path derivation reinforcement processing and the second sample production parameter state path data after the path derivation reinforcement processing are respectively loaded into the abnormal state chain prediction network which serves as the guided machine learning network.

7. The method for online production monitoring of power lines according to claim 1, characterized in that: The first sample production parameter state path data and the second sample production parameter state path data are respectively loaded into the abnormal state chain prediction network as the guided machine learning network, and a first abnormal state chain prediction result of the first sample production parameter state path data and a second abnormal state chain prediction result of the second sample production parameter state path data are generated, including: Performing shallow path derivation and enhancement processing on the first sample production parameter state path data to generate third sample production parameter state path data, wherein the third sample production parameter state path data carries abnormal state chain label data corresponding to the first sample production parameter state path data; Performing shallow path derivation and enhancement processing on the second sample production parameter state path data to generate fourth sample production parameter state path data; Performing deep path derivation and enhancement processing on the second sample production parameter state path data to generate fifth sample production parameter state path data; Performing self-attention processing on the fifth sample production parameter state path data to generate sixth sample production parameter state path data; The third sample production parameter state path data, the fourth sample production parameter state path data, the fifth sample production parameter state path data and the sixth sample production parameter state path data are respectively loaded into the abnormal state chain prediction network as the guided machine learning network to generate a first abnormal state chain prediction result of the third sample production parameter state path data, a second abnormal state chain prediction result of the fourth sample production parameter state path data, state path segments of multiple state spans of the fourth sample production parameter state path data, state path segments of multiple state spans of the fifth sample production parameter state path data and state path segments of multiple state spans of the sixth sample production parameter state path data.

8. The method for online production monitoring of power lines according to claim 7, characterized in that: The step of loading the second sample production parameter state path data into the guiding machine learning network corresponding to the guided machine learning network to generate a third abnormal state chain prediction result of the second sample production parameter state path data includes: Loading the fourth sample production parameter state path data into a guiding machine learning network corresponding to the guided machine learning network, and generating a third abnormal state chain prediction result of the fourth sample production parameter state path data; The method uses the abnormal state chain and the target abnormal state label represented in the third abnormal state chain prediction result as the abnormal state chain label data corresponding to the second sample production parameter state path data, so as to optimize the abnormal state chain prediction network based on the first abnormal state chain prediction result, the second abnormal state chain prediction result, the abnormal state chain label data corresponding to the first sample production parameter state path data, and the abnormal state chain label data corresponding to the second sample production parameter state path data, including: The abnormal state chain and the target abnormal state label represented in the third abnormal state chain prediction result are used as the abnormal state chain label data corresponding to the fourth sample production parameter state path data, so as to optimize the abnormal state chain prediction network based on the first abnormal state chain prediction result and the corresponding abnormal state chain label data of the third sample production parameter state path data, the second abnormal state chain prediction result and the corresponding abnormal state chain label data of the fourth sample production parameter state path data, and the state path segments of multiple state spans of the fourth sample production parameter state path data, the state path segments of multiple state spans of the fifth sample production parameter state path data, and the state path segments of multiple state spans of the sixth sample production parameter state path data.

9. The method for online production monitoring of power lines according to claim 8, characterized in that: The method optimizes the abnormal state chain prediction network based on the first abnormal state chain prediction result of the third sample production parameter state path data and the corresponding abnormal state chain label data, the second abnormal state chain prediction result of the fourth sample production parameter state path data and the corresponding abnormal state chain label data, and the state path segments of multiple state spans of the fourth sample production parameter state path data, the state path segments of multiple state spans of the fifth sample production parameter state path data, and the state path segments of multiple state spans of the sixth sample production parameter state path data, including: Determine a first training error parameter based on the first abnormal state chain prediction result of the third sample production parameter state path data and the corresponding abnormal state chain label data; Determine a second training error parameter based on the second abnormal state chain prediction result of the fourth sample production parameter state path data and the corresponding abnormal state chain label data; For the state path segments of the plurality of state spans respectively corresponding to the fourth sample production parameter state path data and the fifth sample production parameter state path data, determining a fusion calculation result of the characteristic distance training error parameter between the state path segments of the same state span as the third training error parameter; For the state path segments of the multiple state spans respectively corresponding to the fifth sample production parameter state path data and the sixth sample production parameter state path data, determine a fusion calculation result of the characteristic distance training error parameter between the state path segments of the same state span as a fourth training error parameter; A global training error parameter is determined based on the first training error parameter, the second training error parameter, the third training error parameter, and the fourth training error parameter to optimize neuron weight information of the abnormal state chain prediction network based on the global training error parameter.

10. A power line online production monitoring system, characterized in that: The power line online production monitoring system includes a processor and a readable storage medium, wherein the readable storage medium stores a program, and when the program is executed by the processor, the power line online production monitoring method according to any one of claims 1 to 9 is implemented.

Images