A method, device, equipment and storage medium for processing production data

Abstract

The present disclosure provides a method, apparatus, device and storage medium for production data processing, which relates to the field of computer data processing, and in particular to the field of production data processing. The specific implementation scheme is: collecting the production data generated in a specified scenario and the target production data and potential target production data whose output results of the production model are within the target threshold range from the production data; then, aggregating and clustering the above-mentioned collected production data to obtain training data that can be used to train the production model. Since these training data contain data generated in a specified scenario and target production data and potential target production data determined according to the target threshold range, the production model can be trained in a more targeted manner, and by collecting potential target production data, the distribution range of the training data can be further expanded, thereby enhancing the generalization ability of the production model.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technology, and further to the field of deep learning, industrial Internet, industrial production safety, data backflow, and industrial inspection technology, and in particular to a method, device, equipment and storage medium for production data processing. Background Art

[0002] With the continuous development and maturity of deep learning technology, traditional industrial scenarios have gradually applied advanced deep learning technology, using mature deep learning technology to conduct detection and analysis of scenarios such as industrial plant safety prevention and production operation standard inspections.

[0003] The current main technical approach in this field is to train deep learning models using collected production data, then apply these trained deep learning models to specific production scenarios. Safety production scenarios are often open-set environments, which places high demands on the generalization capabilities of deep learning models for practical applications. Therefore, actively feeding data generated in production environments back to the model training end to provide data support for improving deep learning model performance will help improve the generalization capabilities of deep learning models and enhance the overall effectiveness of industrial safety production analysis.

[0004] Among related technologies, the main method is to recirculate data based on error samples. This involves initially screening production data based on image detection results, manually identifying samples with image detection errors, then labeling these samples that were manually identified as errors. Finally, targeted optimization is performed based on these error-detected and relabeled samples to improve the model's ability to identify error samples. However, this method only recirculates error samples, and the number and types of recirculated samples are relatively small. This makes it difficult to further improve the generalization ability of deep learning models in complex environments and large amounts of data. Summary of the Invention

[0005] The present disclosure provides a method, apparatus, device and storage medium for processing production data.

[0006] According to one aspect of the present disclosure, a method for processing production data is provided, comprising: determining production data generated in a specified scenario from production data input into a production model to obtain first data; determining production data that makes an output result of the production model fall within a target threshold range from the production data input into the production model to obtain second data; and aggregating and clustering the first data and the second data to obtain third data.

[0007] According to another aspect of the present disclosure, a device for processing production data is provided, including: a scenario data determination module, used to determine production data generated under a specified scenario from production data input into a production model, to obtain first data; a target production data determination module, used to determine production data that makes the output result of the production model fall within a target threshold range from the production data input into the production model, to obtain second data; and an aggregation and clustering module, used to aggregate and cluster the first data and the second data, to obtain third data.

[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform any of the above-mentioned methods for processing production data.

[0009] According to another aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to enable a computer to execute the above-mentioned production data processing method.

[0010] According to another aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements any of the above-mentioned production data processing methods when executed by a processor.

[0011] The present disclosure provides a method, apparatus, device, and storage medium for processing production data. This method collects production data generated in a specified scenario and target production data and potential target production data whose output results of a production model fall within a target threshold range from production data. The collected production data is then aggregated and clustered to obtain training data that can be used to train the production model. Because this training data contains data generated in the specified scenario and target production data and potential target production data determined based on the target threshold range, the production model can be trained more specifically. The collection of potential target production data further expands the distribution of training data, enhancing the generalization capability of the production model.

[0012] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] The accompanying drawings are provided to facilitate a better understanding of the present invention and do not constitute a limitation of the present disclosure.

[0014] Figure 1 is a flowchart of a method for implementing production data processing according to the first embodiment of the present disclosure;

[0015] Figure 2 is a flowchart of a method for implementing production data processing according to the second embodiment of the present disclosure;

[0016] Figure 3 is a flowchart of a method for implementing production data processing according to the third embodiment of the present disclosure;

[0017] Figure 4 This is a schematic diagram of the process of collecting production data in the detection module in the third embodiment of the present disclosure;

[0018] Figure 5 This is a schematic diagram of the process of collecting production data in the attribute module in the third embodiment of the present disclosure;

[0019] Figure 6 This is a schematic diagram of the process of collecting production data in the tracking and smoothing module in the third embodiment of the present disclosure;

[0020] Figure 7 This is a schematic diagram of the process of collecting production data in the quality module in the third embodiment of the present disclosure;

[0021] Figure 8 This is a schematic diagram of the process of collecting production data in the alarm module in the third embodiment of the present disclosure;

[0022] Figure 9 is a schematic diagram of a process for clustering production data according to the third embodiment of the present disclosure;

[0023] Figure 10 This is a schematic structural diagram of a device for processing production data according to the first embodiment of the present disclosure;

[0024] Figure 11 The block diagram is a block diagram of an electronic device for implementing the method of the embodiment of the present disclosure. DETAILED DESCRIPTION

[0025] The following description of exemplary embodiments of the present disclosure is made in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0026] Figure 1 The process of implementing the method for processing production data according to an embodiment of the present disclosure is shown. Figure 1The method includes: operation S110, determining the production data generated in a specified scenario from the production data input to the production model to obtain first data; operation S120, determining the production data that makes the output result of the production model fall within the target threshold range from the production data input to the production model to obtain second data; operation S130, aggregating and clustering the first data and the second data to obtain third data

[0027] In operation S110, the production data input into the production model may be original production data collected in the production environment; or production data that has been pre-processed by cleaning, denoising, and filtering; or vectorized production data obtained after feature extraction, etc.

[0028] The first data is obtained by determining the production data generated in a specified scenario from the production data input into the production model. This usually refers to pre-analyzing the data generated in a specific scenario to obtain some characteristics of these data that distinguish them from data generated in other scenarios or certain rules they themselves have, and then judging whether the production data input into the production model is the production data generated in the specified scenario based on the above characteristics or rules.

[0029] For example, if a new product is being introduced to a production line, but the production data for that product does not yet exist in actual production, the production data generated from a specific scenario in a non-production environment, such as a pilot production line, can be used as input for the production model.

[0030] Typically, the designated scenario is the production model's intended optimization scenario, such as a scenario with limited data or low prediction accuracy. This allows you to optimize the production model for that specific scenario, enhancing its effectiveness in that specific scenario.

[0031] In operation S120, the output of the production model is a predicted value calculated based on the input data and the model algorithm. It may be a Boolean value, a score, or a set of values and the probability of each value, etc. The target domain value interval corresponds to a numerical range of the output of the production model. For example, if the output of the production model is a Boolean value, the target domain value interval may be 1 or 0; if the output of the production model is a score, the target domain value interval is a numerical range, such as greater than n, less than m, or greater than or equal to n and less than or equal to m, where n and m are specific numerical values; if the output of the production model is a set of values and the probability of each value, the threshold range is also a set of values and the value range of each value.

[0032] Generally, different target threshold intervals represent different types of data. In the present disclosure, the target threshold interval is not only used to collect production data of the target type, but also used to collect potential target production data that is relatively close to the target type data.

[0033] For example, suppose the production model is used to score the degree of production compliance. Its output is the compliance level of the production process, where scores less than 60 are considered non-compliant. If the goal of this training model is to improve the recognition rate of non-compliant production processes, production data with compliance levels less than 65 can be collected. In addition to non-compliant data with scores less than 60, compliant data near the compliance threshold can also be collected. This can appropriately expand the range of candidate training data and improve the generalization and adaptability of the production model.

[0034] Since the production data collected in the above operations S110 and S120 may contain a large amount of redundant data, these redundant data will not only waste computing resources, but may also make the subsequent data processing volume abnormally large, making the processing speed unacceptable, or even resulting in no results.

[0035] To this end, in operation S130, the collected production data is summarized and clustered, and similar data is merged according to the clustering results to save computing resources and reduce the amount of data for subsequent operations without affecting data distribution.

[0036] Clustering can also group data with the same or similar business attributes into a single category, allowing for more accurate data capture of specific business attributes. For example, clustering production data can yield data for different types of computing tools, such as sports cars, trucks, and cars. This allows for subsequent processing, such as adding appropriate classification labels, without the need for manual classification and labeling.

[0037] Thus, the disclosed embodiment collects production data generated under a specified scenario from the production data in operation S110, and collects target production data and potential target production data whose output results of the production model fall within a target threshold range in operation S120. Subsequently, in operation S130, the collected production data is aggregated and clustered to obtain training data for training the production model. Because this training data contains data generated under the specified scenario and target production data and potential target production data determined based on the target threshold range, more targeted training of the production model is possible. Furthermore, the collection of potential target production data further expands the distribution of training data, enhancing the generalization capability of the production model.

[0038] However, in most cases, the amount of production data even after being processed in operation S130 is very large.

[0039] To this end, Figure 2 In another embodiment of the present disclosure, after operation S230, an additional operation S240 is further performed, wherein the third data is filtered based on an active learning strategy to obtain fourth data to be labeled, wherein the active learning strategy includes at least one of the following query strategies:

[0040] The Uncertainty Sampling query strategy is used to select production data that is difficult for the production model to distinguish and hand it over to professionals for labeling, thereby achieving the ability to improve the algorithm's performance at a faster speed;

[0041] A query-by-committee query strategy that uses multiple sub-models to vote to select indistinguishable production data;

[0042] A query strategy based on expected model change is used to select production data that causes the largest gradient change;

[0043] A query strategy based on expected error reduction is used to select production data that reduces the loss function the most;

[0044] A query strategy based on variance reduction is used to select production data that reduces variance the most;

[0045] Density-weighted methods are used to select different strategies based on business scenarios, thereby selecting high-density data that is difficult to distinguish.

[0046] In this way, the amount of data to be labeled can be greatly reduced, the efficiency of training data can be improved, and the labor cost of labeling data can be reduced.

[0047] It should be noted that Figure 1 and Figure 2 The above embodiments shown are only basic embodiments of the method for processing production data disclosed herein. Implementers may further refine and expand the above embodiments and apply them in various applicable application scenarios to form new embodiments.

[0048] Figures 3 to 9 Another embodiment of the present disclosure is shown, which is applied to the scenario of safety and compliance testing of industrial production processes.

[0049] In this application scenario, the embodiment of the present disclosure applies a safety compliance detection model to detect images collected during the production process to determine whether there are any violations in the production process.

[0050] For example, images of workers' work scenes are collected, and it is determined based on the images whether the workers are wearing safety helmets. If not, it is considered that there are violations in the production process, and the corresponding images are identified as violation images.

[0051] In the disclosed embodiment, the production data refers to the images collected during the production process, and the collection environment of these images is a typical open environment. The target to be analyzed (for example, the head of the operator) is very susceptible to changes in various factors such as background, distance, sudden / interference objects, and the collected sample set may also have unpredictable situations. For example, a flying insect flies into the camera, making it impossible to capture the entire production process.

[0052] In the embodiment of the present disclosure, the security compliance detection model uses a model reasoning process including multiple processing stages when performing security compliance detection. Each processing stage corresponds to a processing module, for example, Figure 3 The detection module, attribute module, tracking smoothing module, quality module and alarm module are shown.

[0053] The detection module is implemented by the target detection sub-model, which is used to detect the target object (for example, a worker or a vehicle). The output of the target detection sub-model is the location of the target object in the image and the probability of the target image appearing at that location. The probability of the target image appearing at that location is the detection score of the target detection sub-model.

[0054] The attribute module is implemented through the attribute sub-model. It is used to determine the safety compliance attributes of the target object, such as the worker, based on the target object obtained by the detection module. For example, whether the target object is wearing a hard hat or protective clothing. The score output by the attribute sub-model represents the probability of meeting the safety compliance attributes and represents the worker's safety compliance level.

[0055] The tracking smoothing module is implemented through the tracking smoothing processing sub-model. It is used to eliminate or process the non-smooth frames with inconsistent attributes between the previous and next frames in the detection module based on the output of the detection module and the output of the attribute module, so that the final production data has consistent attributes between the previous and next frames and is logical.

[0056] The quality module is implemented through the quality sub-model. The output of the quality sub-model is a quality score, which is used to indicate the quality of the block diagram of the target object output by the target detection sub-model.

[0057] The alarm module is implemented through the alarm sub-model, which is used to make safety compliance judgments on the production data processed by the above processing module. If the production data is safety non-compliant (illegal image), an alarm will be issued and a non-compliant image will be given.

[0058] After the above processing flow, the security compliance detection model can send the detected violation images and alarm events to relevant personnel for timely processing.

[0059] See also Figure 3 In the disclosed embodiments, the collection of production data is performed simultaneously with the application of a safety compliance detection model to the images collected during the production process. The solid line represents the process flow of industrial safety production analysis, while the dotted line represents the summary of data collected at each stage. The main process includes:

[0060] Operation S3010: inputting a video frame;

[0061] The video frames here are images of the production workshop captured by cameras in the production environment, and each video frame is equivalent to an image.

[0062] Operation S3020: collecting data in the detection module;

[0063] The specific process of collecting data in the detection module is as follows Figure 4 As shown, it mainly includes:

[0064] Step S410, receiving input data (video frame);

[0065] Step S420, obtaining a detection score by detecting the sub-model;

[0066] Step S430, determine whether it is an emergency scene, if so, proceed to step S450, if not, proceed to step S440;

[0067] Here, the emergency scenario refers to an abnormal production scenario, and the data generated in the emergency scenario will also be abnormal production data.

[0068] Since normal production data under normal production conditions is usually used when training the safety and compliance detection model, the safety and compliance detection model will frequently and continuously issue alarms when detecting abnormal production data generated in emergency scenarios.

[0069] Therefore, when continuous alarms are received during the production process, the device number and time period that continuously sends the alarms can be recorded as conditions for determining whether the scenario that generates the production data is an emergency scenario.

[0070] For example, you can obtain the device number and time period of the current production data. If the device number of the production data is the same as the device number recorded in the emergency scenario and the acquisition time of the current production data also coincides with the time period in the emergency scenario, then it means that the current production data is data generated in the emergency scenario.

[0071] Emergency scenarios caused by the same reason often last for a period of time, and it is usually not necessary to retain all production data collected during this period.

[0072] Therefore, this embodiment also sets a selection strategy for saving production data in emergency scenarios. For example, a limit is set on the number of images to be saved. After determining that the current production data is produced in an emergency scenario, the production data that meets the selection strategy is selected as the scene data for the emergency scenario. This can avoid data disasters and make data collection more efficient.

[0073] Step S440: determine whether the detection score is within the valid range. If so, proceed to step S450; if not, terminate this execution.

[0074] The valid interval of the detection score here (target threshold interval) mainly refers to the dual threshold area formed by the first threshold above the set threshold and the second threshold below the set threshold, where the set threshold refers to a probability value that can be considered to have detected the target.

[0075] In the disclosed embodiments, a first threshold value above a set threshold value and a second threshold value below a set threshold value may be configured (including setting or modifying) based on the needs of production model optimization. This may enable the model to converge faster, achieve the desired accuracy, or realize the desired improvement in generalization capability.

[0076] Step S450: save the original image.

[0077] Since the data collection environment of the embodiment of the present disclosure is an open set environment, its detection results will be affected by factors such as background and distance. Therefore, in order to better analyze the detection results, the production data collected in the detection module is the collected original image.

[0078] In this way, the production data collected in the detection module can be obtained.

[0079] Operation S3030: collecting data in the attribute module;

[0080] The specific process of collecting data in the attribute module is as follows Figure 5 As shown, it mainly includes:

[0081] Step S510, receiving input data;

[0082] The input data of the attribute module is a box diagram obtained by cutting out the area where the target object is located according to the output of the detection module.

[0083] Step S520, obtaining attribute scores through the attribute sub-model;

[0084] Among them, attributes refer to safety and compliance attributes, such as a condition for judging safety and compliance: whether the clothing of workers in a production environment meets safety standards, etc.; and attribute scores refer to the probability that the input data meets the corresponding safety and compliance attributes.

[0085] Step S530, determine whether it is an emergency scene, if so, proceed to step S550, if not, proceed to step S540;

[0086] The determination here is similar to step S430, so it will not be repeated here.

[0087] Step S540: determine whether the attribute score is within the valid range. If so, proceed to step S550; if not, terminate this execution.

[0088] The valid interval here is also defined by a double threshold. In the attribute module, the specified threshold usually represents the attribute score that can be determined as non-compliant.

[0089] Step S550: save the block diagram.

[0090] Since the attribute module only needs to be responsible for the block diagram obtained by cutting out the target object, in step S550, only the block diagram containing the target object needs to be saved.

[0091] In this way, the production data collected in the attribute module can be obtained.

[0092] Operation S3040: collecting data in the tracking smoothing module;

[0093] The specific process of collecting data in the tracking smoothing module is as follows Figure 6 As shown, it mainly includes:

[0094] Step S610, receiving input data;

[0095] The input data received by the tracking smoothing module are the original image saved by the detection module and the block diagram saved in the attribute module.

[0096] Step S620, obtaining a smoothing queue through a tracking smoothing module;

[0097] Combining the attribute results and attribute scores obtained in the attribute module, a smoothing analysis is performed on the previous and next frames in the original image and frame queues. Frames with inconsistent attributes are identified as non-smoothed frames and removed. After this smoothing analysis and processing, a smoothed queue containing only smoothed frames is obtained.

[0098] Step S630, determine whether the current frame is in the smoothing queue, if so, proceed to step S640, if not, end this execution;

[0099] Step S640: determine whether the attributes of the previous and next frames are consistent. If so, return to step S630 to continue acquiring the next frame. If not, proceed to step S650.

[0100] Step S650: save the frame image and the original image, and then continue to acquire the next frame.

[0101] In this way, the production data collected in the tracking smoothing module can be obtained.

[0102] Operation S3050: collecting data in the quality module;

[0103] The specific process of collecting data in the attribute module is as follows Figure 7 As shown, it mainly includes:

[0104] Step S710, receiving input data;

[0105] The input data is the original image or box image in the smoothing queue.

[0106] Step S720, obtaining a quality score through the quality sub-model;

[0107] The quality score represents the quality of the captured video frame image.

[0108] Step S730, determine whether the quality score is within the valid range, if so, proceed to step S740, if not, end this execution;

[0109] The valid interval here is also defined by a double threshold. In the attribute module, the specified threshold usually represents the quality score that can be determined as poor quality.

[0110] Step S740: save the original image or block diagram.

[0111] In this way, the production data collected in the quality module can be obtained.

[0112] Operation S3060: collecting data in the alarm module;

[0113] The specific process of collecting data in the alarm module is as follows Figure 8 As shown, it mainly includes:

[0114] Step S810, receiving input data;

[0115] The input data of the alarm module is the original image or block diagram collected by the above-mentioned detection module, attribute module, tracking smoothing module and quality module.

[0116] Step S820, obtaining alarm information through the alarm module;

[0117] Identify and judge the original image or block diagram to determine whether there is any illegal operation in the original image or block diagram. If so, an alarm will be issued, and the reason for the alarm and the original image or block diagram will be given.

[0118] Step S830: Ask the customer to confirm the warning information;

[0119] Step S840, determine whether it is a false alarm, if yes and it is a false alarm, proceed to step S850, if yes and it is a false alarm, proceed to step S860, if no, terminate this execution;

[0120] Among them, false detection refers to the situation where the target object is not contained in the area output by the target detection; false alarm refers to the situation where the target object is contained, but the judgment on whether the target object is in violation is incorrect.

[0121] Step S850: It is a false alarm, and the block diagram is saved;

[0122] Step S860: It is a false detection, and the original image is saved;

[0123] Step S870: perform data aggregation.

[0124] In this way, production data (original image or block diagram) that results in false detection or false alarm when the safety compliance detection model is applied in actual production can be obtained.

[0125] Operation S3070: collecting and obtaining a set of production data;

[0126] The production data collected in the detection module, attribute module, tracking smoothing module, quality module and alarm module are merged to obtain the production data collection.

[0127] Operation S3080: transmitting data using a data transmission module;

[0128] In order to ensure the performance of the online model and reasonably allocate computing power, the embodiment of the present disclosure uploads the production data preliminarily screened by the security and compliance detection model through the data transmission module to support subsequent data labeling, data screening, model iteration and other processing.

[0129] Specifically, to avoid peak hours for on-site tasks, a scheduled task is created to automatically upload the day's preliminary screening data to an online storage space at a fixed time each day. The server then sends a request to process the data in the storage space, performs a series of image preprocessing on it, and then sends it to the clustering module.

[0130] Operation S3090: clustering the production data using a clustering module;

[0131] Since the data collected in the above steps is large in volume and has data redundancy, the embodiment of the present disclosure uses a clustering module to cluster the production data. Figure 9 As shown, it mainly includes:

[0132] Step S910, receiving input data;

[0133] The input data received by the clustering module is production data that has been preliminarily screened by the safety compliance detection model and obtained after a series of graphic preprocessing. These data include both block diagrams and original images. If it is a block diagram, proceed to step S920; if it is an original image, proceed to step S950.

[0134] Step S920: If the input data is a block diagram, it is directly input into the feature extraction model for feature extraction;

[0135] Step S930, clustering the extracted features using the dbscan algorithm to obtain block diagram clusters;

[0136] Step S940: if it is an original image, crop the original image to obtain a target image;

[0137] Step S950: sending the target image to the feature extraction module for feature extraction;

[0138] Step S960, clustering the extracted features using the dbscan algorithm to obtain clusters of the target image;

[0139] Step S970: Then, the clusters of the original image are obtained according to the mapping relationship.

[0140] In this way, clustered production data is obtained. Then, based on the clustering results, some typical production data can be selected from each cluster as candidate frames to be labeled.

[0141] After the above clustering operation, the redundant data of production data can be greatly reduced, and the amount of data to be processed subsequently can be reduced.

[0142] Operation S3100: selecting a frame to be labeled based on an automatic learning strategy;

[0143] Since data labeling is costly and time-consuming, in the embodiment of the present disclosure, a batch of production data that is relatively helpful for subsequent model training and iteration is selected from massive production data through active learning to be labeled, so as to reduce labeling costs and save time.

[0144] The core of the active learning strategy lies in the selection of query strategies. One or more query strategies may be selected from the multiple query strategies described in the aforementioned operation S240 according to business scenario requirements.

[0145] Operation S3110: annotating the frame to be annotated to form training data;

[0146] At this point, the number of frames to be annotated is greatly reduced after the automatic learning strategy is applied, and manual or automatic annotation methods can be used to annotate and obtain training data. For example, a small number of frames can be manually annotated first, and then a labeling model can be trained based on these frames. This labeling model can then be used to annotate the remaining frames to be annotated.

[0147] Operation S3120: Targeted optimization of the model.

[0148] Once labeled training data is obtained, the existing security and compliance detection model can be trained using the training data to generate a trained production model. The trained security and compliance detection model can then be applied to actual production data for security and compliance testing. Because the trained security and compliance detection model is optimized for emergencies, non-compliant production data, and near-non-compliant production data, it can capture more emergencies and non-compliant events, achieve higher accuracy, and achieve better application results.

[0149] In the disclosed embodiment, data collection is embedded into multiple sub-models in multiple processing stages of the security compliance detection model, so that more numbers and types of data can be collected from each processing stage. Compared with the end-to-end one-step method of collecting production data, the generalization ability of the security compliance detection model can be better improved, and it is easy to trace back to a specific stage.

[0150] In the disclosed embodiments, a data collection mechanism for emergency scenarios is introduced, enabling targeted collection of data generated during emergency scenarios. This provides effective data support for root cause analysis of emergency scenarios. In some cases, the same root cause can trigger multiple emergency scenarios. If root cause analysis can be performed, the security compliance detection model can quickly converge on the various emergency scenarios caused by the same root cause, thereby reducing the number of emergency scenarios that actually need to be processed and making subsequent emergency scenario processing more efficient.

[0151] In the disclosed embodiment, the valid interval also includes target production data (e.g., non-compliant production data) and potential target production data at each processing stage. This greatly expands the number and types of candidate training data, improving the safety and compliance detection model's detection rate for non-compliant issues and enhancing its generalization capabilities.

[0152] In the disclosed embodiment, the collection of production data is carried out simultaneously with the actual application of the security compliance detection model, which can greatly simplify the entire production data backflow processing process, accelerate the convergence of the model, and shorten the model iteration cycle.

[0153] Furthermore, the present disclosure also provides a device for secure data processing, such as Figure 10 As shown, the device 100 includes: a scenario data determination module 1001, which is used to determine the production data generated in a specified scenario from the production data of the input production model to obtain first data; a target production data determination module 1002, which is used to determine the production data that makes the output result of the production model fall within the target threshold range from the production data of the input production model to obtain second data; and an aggregation and clustering module 1003, which is used to aggregate and cluster the first data and the second data to obtain third data.

[0154] In another embodiment of the present disclosure, the production model includes at least two sub-models, each sub-model corresponds to a processing stage, and accordingly, the scenario data determination module 1001 is also used to determine the production data generated under a specified scenario from the production data input to each sub-model to obtain first data; accordingly, the target production data determination module 1002 is also used to determine the production data whose corresponding sub-model score is within the target threshold range from the production data input to each sub-model to obtain second data.

[0155] In another embodiment of the present disclosure, the scenario data determination module 1001 includes: a scenario data determination submodule, which is used to determine the production data generated under the specified scenario from the input data; and a scenario data screening submodule, which is used to determine the production data that meets the data selection strategy from the production data generated under the specified scenario to obtain the first data.

[0156] In another embodiment of the present disclosure, the designated scenario is an emergency scenario. Accordingly, the scenario data determination module 1001 includes: an emergency scenario determination submodule, which is used to determine the device number and time period that generate the emergency scenario based on the alarm frequency and duration; and an emergency scenario data determination submodule, which is used to determine the production data generated under the emergency scenario from the production data of the input production model based on the device number and time period to obtain the first data.

[0157] In another embodiment of the present disclosure, the target threshold interval includes a dual threshold area formed by a first threshold above a set threshold and a second threshold below the set threshold, and the device 1000 also includes: a target threshold interval configuration module, which configures the first threshold area and the second threshold according to the needs of production model optimization.

[0158] In another embodiment of the present disclosure, the apparatus 100 further includes: a data screening module, which screens the third data based on an active learning strategy to obtain fourth data to be labeled.

[0159] In another embodiment of the present disclosure, the device 100 also includes: a data labeling module, used to label the fourth data to obtain training data; a model training module, used to use the training data to train the production model to obtain a trained production model; and a model replacement module, used to apply the trained production model to perform production model prediction on actual production data.

[0160] In the technical solutions disclosed herein, the acquisition, storage, and application of user personal information involved comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0161] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0162] Figure 11 A schematic block diagram of an example electronic device 800 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are provided as examples only and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0163] like Figure 11 As shown, the device 1100 includes a computing unit 1101, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1102 or a computer program loaded from a storage unit 1108 into a random access memory (RAM) 1103. Various programs and data required for the operation of the device 1100 can also be stored in the RAM 1103. The computing unit 1101, the ROM 1102, and the RAM 1103 are connected to each other via a bus 1104. An input / output (I / O) interface 1105 is also connected to the bus 1104.

[0164] Various components in device 1100 are connected to I / O interface 1105, including an input unit 1106, such as a keyboard and mouse; an output unit 1107, such as various types of displays and speakers; a storage unit 1108, such as a magnetic disk and optical disk; and a communication unit 1109, such as a network card, a modem, a wireless communication transceiver, etc. Communication unit 1109 allows device 1100 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0165] The computing unit 1101 can be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 1101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units that run machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 1101 performs the various methods and processes described above, such as the processing method of production data of the present application. For example, in some embodiments, the method of production data processing of the present disclosure can be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 1108. In some embodiments, part or all of the computer program can be loaded and / or installed on the device 1100 via the ROM 1102 and / or the communication unit 1109. When the computer program is loaded into the RAM 1103 and executed by the computing unit 1101, one or more steps of the method of production data processing described above can be performed. Alternatively, in other embodiments, the computing unit 1101 may be configured to execute the production data processing method of the present disclosure in any other appropriate manner (for example, by means of firmware).

[0166] Various embodiments of the systems and techniques described herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), system-on-chip systems (SOCs), programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0167] The program code for implementing the method of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device so that when the program code is executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The program code can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0168] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0169] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0170] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.

[0171] A computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The client-server relationship arises through computer programs running on the respective computers and having a client-server relationship with each other. The server may be a cloud server, a server in a distributed system, or a server integrated with a blockchain.

[0172] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this disclosure can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved. This is not a limitation herein.

[0173] The above specific embodiments do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the scope of protection of this disclosure.

Claims

1. A method for processing production data, comprising: determining production data generated in a specified scenario from production data input into the production model to obtain first data; The designated scenario is an emergency scenario; Determining, from the production data input to the production model, production data that causes an output result of the production model to be within a target threshold range, to obtain second data; Aggregating and clustering the first data and the second data to obtain third data; The third data is used to train the production model; The step of determining production data generated in a specified scenario from the production data input into the production model to obtain first data includes: Determine the device number and time period that caused the emergency scenario based on the alarm frequency and duration; According to the device number and the time period, production data generated in an emergency scenario is determined from the production data input into the production model to obtain first data.

2. The method according to claim 1, wherein The production model includes at least two sub-models, each sub-model corresponds to a processing stage, Accordingly, determining the production data generated in the specified scenario from the production data input into the production model to obtain the first data includes: Determining production data generated in a specified scenario from production data input into each sub-model to obtain first data; Accordingly, the step of determining, from the data input into the production model, production data that causes the output result of the production model to be within the target threshold range to obtain the second data includes: The production data input into each sub-model is determined as the production data in which the corresponding sub-model score is within the target threshold range to obtain the second data.

3. The method according to claim 1, wherein Determining production data generated in a specified scenario from production data input into a production model to obtain first data includes: Determining production data generated under a specified scenario from production data input into a production model; From the production data generated in the specified scenario, production data that meets the data selection strategy is determined to obtain first data.

4. The method according to claim 1, wherein The target threshold interval includes a dual-threshold region formed by a first threshold above a set threshold and a second threshold below the set threshold, and the method further includes: The first threshold and the second threshold are configured according to the needs of the production model optimization.

5. The method according to claim 1, further comprising: Based on the active learning strategy, the third data is screened to obtain fourth data to be labeled.

6. The method according to claim 5, wherein: The active learning strategy includes at least one of the following query strategies: A query strategy of uncertainty sampling, used to select production data that is difficult for the production model to distinguish; A committee-based query strategy for voting across multiple sub-models to select indistinguishable production data; A query strategy based on model change expectations is used to select production data that maximizes gradient changes. A query strategy based on error reduction is used to select the production data that reduces the loss function the most; A query strategy based on variance reduction is used to select the production data that reduces the variance the most; Density-weighted query strategy is used to select high-density and difficult-to-distinguish data.

7. The method according to claim 5, further comprising: Labeling the fourth data to obtain training data; Using the training data to train the production model to obtain a trained production model; Apply the trained production model to perform production model prediction on actual production data.

8. A device for processing production data, comprising: A scenario data determination module, configured to determine production data generated in a specified scenario from production data input into the production model to obtain first data; The designated scenario is an emergency scenario; a target production data determination module, configured to determine, from the production data input to the production model, production data that causes the output result of the production model to fall within a target threshold range, thereby obtaining second data; an aggregation and clustering module, configured to aggregate and cluster the first data and the second data to obtain third data; The third data is used to train the production model; The scene data determination module includes: The emergency scene determination submodule is used to determine the device number and time period that generates the emergency scene based on the alarm frequency and duration; The emergency scenario data determination submodule is used to determine the production data generated in the emergency scenario from the production data input into the production model according to the equipment number and time period to obtain first data.

9. The device according to claim 8, wherein The production model includes at least two sub-models, each sub-model corresponds to a processing stage, Accordingly, the scenario data determination module is further configured to determine the production data generated under the specified scenario from the production data input to each sub-model to obtain the first data; Correspondingly, the target production data determination module is further configured to determine, from the production data input into each sub-model, production data that makes the score of the corresponding sub-model fall within the target threshold range, to obtain second data.

10. The device according to claim 8, wherein The scene data determination module includes: A scenario data determination submodule is used to determine the production data generated under a specified scenario from the production data input into the production model; The scenario data screening submodule is used to determine the production data that meets the data selection strategy from the production data generated in the specified scenario to obtain the first data.

11. The device according to claim 8, wherein The target threshold interval includes a dual-threshold region formed by a first threshold above a set threshold and a second threshold below the set threshold, and the apparatus further includes: The target threshold interval configuration module configures the first threshold and the second threshold.

12. The apparatus according to claim 8, further comprising: The data screening module screens the third data based on an active learning strategy to obtain fourth data to be labeled.

13. The apparatus according to claim 12, further comprising: a data labeling module, configured to label the fourth data to obtain training data; A model training module, configured to train the production model using the training data to obtain a trained production model; The model replacement module is used to apply the trained production model to perform production model prediction on actual production data.

14. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

15. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to enable the computer to execute the method according to any one of claims 1 to 7.

16. A computer program product comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements the method according to any one of claims 1 to 7.

Images