Online updating method and system for AI algorithm of industrial Internet of Things

Abstract

The invention discloses an online updating method and system for an industrial Internet of Things AI algorithm, and belongs to the technical field of AI algorithms, and the method comprises the following steps: collecting industrial information, and obtaining a multi-source data set; performing exception removal processing on the multi-source data set to obtain a normal data set; constructing an AI algorithm model through the normal data set, and deploying the AI algorithm model on an AI algorithm node; and analyzing the AI algorithm node, and updating the AI algorithm model based on an analysis result. According to the method, the weight is dynamically adjusted by counting the abnormal value proportion of the data source for nearly N times, that is, the historical abnormal rate of a certain sensor is high, the weight of the sensor can be reduced, the interference of noise on the whole data set is avoided, the weight dynamically changes along with the historical performance of the data source, the weighted multi-source data is closer to the actual data quality, and the data quality is improved. And the data reliability of subsequent model training is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of AI algorithms, and particularly relates to an online update method and system for an AI algorithm in an industrial Internet of Things. Background Art

[0002] In the industrial Internet of Things, multi-source data (such as sensor data, device logs, environmental parameters, etc.) usually has characteristics such as heterogeneity, high noise, and high real-time requirements. Traditional methods usually use fixed weights to weight multi-source data and cannot adapt to the historical anomaly rate differences of different data sources. For example, old sensors may have a high anomaly rate for a long time. However, if the weights are fixed, it may lead to underestimation of their valid data or misjudgment of abnormal values. Secondly, in industrial scenarios, some short-term fluctuations may be misjudged as anomalies, such as the instantaneous noise when the device starts. Direct deletion may cause loss of valid data and affect the integrity of model training. Summary of the Invention

[0003] In view of the above existing problems, the present invention is proposed.

[0004] Therefore, the technical problems solved by the present invention are: traditional methods adopt a fixed weight allocation method and a single threshold deletion mechanism.

[0005] To solve the above technical problems, the present invention provides the following technical solutions: an online update method for an AI algorithm in an industrial Internet of Things, which includes the following steps. Collect industrial information to obtain a multi-source data set; Perform anomaly removal processing on the multi-source data set to obtain a normal data set; Construct an AI algorithm model through the normal data set and deploy the AI algorithm model on an AI algorithm node; Analyze the AI algorithm node and update the AI algorithm model based on the analysis results.

[0006] As a preferred solution of an online update method for an AI algorithm in an industrial Internet of Things according to the present invention, wherein: the step of obtaining the multi-source data set includes, Build a data collection platform according to the industrial application environment, and collect different types of data sources in each working cycle to obtain initial data; And perform data cleaning on the initial data; Perform feature extraction and feature selection on the cleaned initial data to obtain a data feature set, and merge the data sources with similar features in the data feature set to obtain a multi-source data set.

[0007] As a preferred solution of an online update method for an AI algorithm in an industrial Internet of Things according to the present invention, wherein: the step of obtaining the normal data set includes, Preset weight values; weight the multi-source data set based on the weight values to obtain a weighted multi-source data set; detect outliers in the weighted multi-source data set; set an outlier threshold and judge the difference between the outlier and the outlier threshold; when the difference is less than the safety value, perform secondary analysis; when the difference is not less than the safety value, delete the corresponding outlier; Among them, the steps of the secondary analysis include: setting a quantity value; obtaining the quantity of differences less than the safety value in this group of multi-source data sets; when the obtained quantity is less than the set quantity value, do not delete the corresponding outlier; otherwise, delete the corresponding outlier; Through the secondary analysis mechanism, it is possible to avoid data deletion errors caused by misjudgment of single outliers, and improve the accuracy of outlier detection.

[0008] As a preferred solution of an online update method of an industrial Internet of Things AI algorithm according to the present invention, among them: when presetting the weight value, the following steps are included, Obtain the data type of the data source, and at the same time obtain the recent N data sources of different types; Calculate the outlier proportion of the recent N data sources of different types, and perform calculations, and output as the weight value; Calculate the weight based on the outlier proportion of historical data, give priority to trusting data sources with high stability, improve the credibility of multi-source data fusion, and the weight changes dynamically according to the historical performance of the data source, making the model pay more attention to highly reliable data and reducing the interference of low-quality data sources.

[0009] As a preferred solution of an online update method of an industrial Internet of Things AI algorithm according to the present invention, among them: the steps of constructing the AI algorithm model include, Divide the normal data set into a training set and a test set according to a preset ratio through a cross-validation algorithm; Among them, the training set and the test set are respectively used for training the AI algorithm model and testing the AI algorithm model to obtain the AI algorithm model.

[0010] As a preferred solution of an online update method of an industrial Internet of Things AI algorithm according to the present invention, among them: the steps of analyzing the AI algorithm node include, Determine the index system according to the actual needs of the user, and determine the factor set that affects the index system; Calculate the influence degree of each sub-factor of the factor set on the index system; Evaluate based on the influence degree of each sub-factor on the index system to obtain the online update result of the AI algorithm model; Optimize the model of the AI algorithm model based on the online update result.

[0011] As a preferred solution of an online update method for an AI algorithm in industrial Internet of Things according to the present invention, when evaluating the influence degree of each sub-factor on the index system, the following steps are included: Through the index system of actual requirements, obtain the sub-factor with the greatest influence degree corresponding to the index system, and adjust the weight of this sub-factor to the maximum; then obtain the factor with the second greatest influence degree, and adjust the weight of this sub-factor to the second largest; repeat this method until all sub-factors are cycled once. Adjust the weight according to the influence degree of the sub-factor on the index, so that the AI algorithm model pays more attention to key factors, improves the pertinence of the update, and ensures that the core factor dominates the model behavior through the priority sorting of weight adjustment.

[0012] Another object of the present invention is to provide an online update system for an AI algorithm in industrial Internet of Things.

[0013] To solve the above technical problems, the present invention provides the following technical solution: An online update system for an AI algorithm in industrial Internet of Things, including a data acquisition module, a model construction module, and a model update module; The data acquisition module is responsible for collecting industrial information, obtaining a multi-source data set, and performing anomaly removal processing on the multi-source data set to obtain a normal data set; The model construction module is used to construct an AI algorithm model and deploy the constructed AI algorithm model to an AI algorithm node; The model update module is used to analyze the AI algorithm node and update the AI algorithm model based on the analysis result.

[0014] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor, when executing the computer program, implements the steps of the online update method for an AI algorithm in industrial Internet of Things as described above.

[0015] The present invention provides a computer-readable storage medium, on which a computer program is stored, and the computer program, when executed by a processor, implements the steps of the online update method for an AI algorithm in industrial Internet of Things as described above.

[0016] The beneficial effects of the present invention: By statistically calculating the proportion of abnormal values in the data source in the recent N times, the weight is dynamically adjusted. That is, if the historical abnormal rate of a certain sensor is relatively high, its weight will be reduced to avoid the interference of its noise on the overall data set, and the weight changes dynamically with the historical performance of the data source, making the weighted multi-source data closer to the actual data quality and improving the data reliability of subsequent model training.

[0017] When the difference between the outlier and the threshold is less than the safety value, instead of directly deleting it, the number of similar situations in this group of data is counted. That is, if only a few samples in a certain batch of data are close to the threshold, it may be normal fluctuations and the data is retained; if the number exceeds the set threshold, it is determined as abnormal and deleted to avoid the loss of valid data caused by single fluctuations or noise, ensure the integrity of the training data, and especially ensure the coverage ability of the model for the real working conditions in industrial scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0019] Figure 1 FIG. is the overall flowchart of an online update method for an industrial Internet of Things AI algorithm provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0020] In order to make the above objects, features, and advantages of the present invention more obvious and understandable, the following will describe the specific embodiments of the present invention in detail with reference to the drawings of the specification. Obviously, the described embodiments are some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0021] Embodiment 1, referring to Figure 1 , which is the first embodiment of the present invention. This embodiment provides an online update method for an industrial Internet of Things AI algorithm, including: S1. Collect industrial information and obtain a multi-source data set.

[0022] The steps for obtaining the multi-source data set include Build a data acquisition platform according to the industrial application environment, and collect different types of data sources in each working cycle to obtain initial data; In this embodiment, for example, in an automobile manufacturing factory, collect data of production line equipment, environment, and sensors; install temperature sensors (monitoring equipment temperature), pressure sensors (monitoring hydraulic system pressure), vibration sensors (monitoring robotic arm vibration), PLC controllers (recording equipment status codes), environmental temperature and humidity sensors, etc., and deploy a SCADA system (supervisory control and data acquisition system) as a data transfer station.

[0023] And perform data cleaning on the initial data; If a sensor has no data during a certain acquisition cycle (such as a temperature sensor failure), it is filled with linear interpolation of the previous valid value; for example, the reading of a temperature sensor is 150 °C, which is much higher than the normal range of 50 - 80 °C. Extract key features from the cleaned data and screen important features. Perform feature extraction and feature selection on the initial cleaned data to obtain a data feature set. Merge data sources with similar features in the data feature set to obtain a multi-source data set. Use conventional analysis of variance (ANOVA) to screen features that have a great impact on equipment failure prediction. Finally, merge data sources with similar features to reduce redundancy and improve the efficiency of subsequent processing.

[0024] S2. Perform anomaly processing on the multi-source data set to obtain a normal data set.

[0025] Obtain the data type of the data source, and at the same time obtain the nearly N times of data sources of different types; Calculate the proportion of outliers in the nearly N times of data sources of different types, and perform calculations, and the output is the weight value.

[0026] In an optional embodiment, a certain CNC machine tool processing center needs to monitor the tool wear state in real time to prevent workpiece scrapping caused by tool breakage. The system integrates three types of sensors, namely a temperature sensor for monitoring the contact temperature between the tool and the workpiece; a vibration sensor for detecting the vibration amplitude of the spindle; and a current sensor for collecting the current of the spindle motor.

[0027] The data source types are temperature, vibration, and current respectively; in the nearly N times of data, collect the latest 100 sampling data of each type of sensor. That is, in this embodiment, N = 100. The obtained number of anomalies is 12 times for temperature, so the proportion is 12%; 25 times for vibration, so the proportion is 25%; 5 times for current, so the proportion is 5%. The lower the anomaly proportion, the higher the data reliability and the greater the weight; the calculation formula is: ; It can be obtained that the weight value is 0.29 for temperature, 0.14 for vibration, and 0.57 for current.

[0028] Perform weighting on the multi-source data set based on the weight value to obtain a weighted multi-source data set; Enlarge or reduce the multi-source data set according to the weight, that is, temperature value × 0.29, vibration value × 0.14, current value × 0.57.

[0029] Detect outliers in the weighted multi-source data set; Set an anomaly threshold and judge the difference between the outlier and the anomaly threshold; When the difference is less than the safety value, secondary analysis is performed; when the difference is not less than the safety value, the corresponding outlier is deleted. In this embodiment, for example, during temperature detection, the detected outliers are: 80.2, 81.5, 88.6, 80.7, 90.4, 88.9, 89.1, 88.3, 91.1, 91.6, 79.1, 90.9.

[0030] The set outlier threshold is 75. By calculating the difference between each outlier and the outlier threshold in turn, if the set safety value is 5, it is determined that there is only one group with a difference less than the safety value, which is 79.1, and the remaining outliers are all deleted.

[0031] Among them, the steps of the secondary analysis include: Set a quantity value; Obtain the number of differences less than the safety value in this group of multi-source data sets; When the obtained number is less than the set quantity value, the corresponding outlier is not deleted; otherwise, the corresponding outlier is deleted.

[0032] In this embodiment, the set quantity value is 3, and after the previous step, there is only 1 remaining difference quantity. Therefore, the corresponding outlier is not deleted, that is, 79.1 is retained; For example, in the previous step, there are 4 remaining difference quantities, such as: 79.1, 80.2, 80.7, 81.5; therefore, the remaining difference quantity is not less than the set quantity value, and all 4 of these outliers are deleted.

[0033] S3. Construct an AI algorithm model through the normal data set and deploy the AI algorithm model on the AI algorithm node.

[0034] The construction steps of the AI algorithm model include, Divide the normal data set into a training set and a test set according to a preset ratio through the cross-validation algorithm; Among them, the training set and the test set are respectively used for training the AI algorithm model and testing the AI algorithm model to obtain the AI algorithm model.

[0035] In this embodiment, to avoid future information leakage of time series data, the data is divided into data blocks according to processing batches. The first 7 processing batch data, about 10,000 pieces per batch, are used as the training set; the last 3 processing batch data are used as the test set.

[0036] S4. Analyze the AI algorithm node and update the AI algorithm model based on the analysis results.

[0037] The steps of analyzing the AI algorithm node include, Determine the index system according to the actual needs of the user, and determine the factor set that affects the index system; Among them, the index system includes: the accuracy rate of the AI algorithm model, the response time of the AI algorithm model, and the operating cost of the AI algorithm model. The factor set includes the AI algorithm model structure, the AI algorithm model algorithm, and the AI algorithm model scale; Calculate the influence degree of each sub-factor in the factor set on the index system; Through the index system of actual needs, obtain the sub-factor with the greatest influence degree corresponding to the index system, and adjust the weight of this sub-factor to the maximum; Then obtain the factor with the second greatest influence degree, and adjust the weight of this sub-factor to the second largest; Repeat this method until all sub-factors are looped through; Obtain the online update result of the AI algorithm model; perform model tuning on the AI algorithm model based on the online update result.

[0038] In an alternative embodiment, for example, a certain e-commerce warehouse uses 100 logistics robots for goods sorting and needs to plan the optimal path in real time. The following problems are currently faced: congestion during peak hours, resulting in backlogs of goods due to delayed robot responses; fast battery consumption, complex path algorithms increasing the CPU burden and shortening the battery life; slow adaptation to new shelves, and after adding new shelves, the error rate of the original model planning increases.

[0039] First, define the index system and the factor set. The accuracy rate of the AI algorithm model ≥ 98%, the response time of the AI algorithm model ≤ 200 ms, and the operating cost of the AI algorithm model, that is, the CPU occupancy ≤ 15%. The AI algorithm model structure is the number of neural network layers and the attention mechanism; the AI algorithm model algorithms are the A* algorithm, the Dijkstra algorithm, and the deep learning model. The AI algorithm model scale includes the number of parameters.

[0040] Through historical data statistics, the influence of each factor on the index is as follows: the AI algorithm model structure is expressed as increasing the number of neural network layers → the accuracy rate +10%, but the response time +30 ms, and the CPU occupancy +5%; the AI algorithm model algorithm is expressed as switching to the A* algorithm → the response time -50 ms, but the accuracy rate -3%; the AI algorithm model scale is expressed as quantifying and compressing the model → the CPU occupancy -8%, but the accuracy rate -2%. Therefore, it is obtained that the accuracy rate is mainly affected by the model structure; the response time is mainly affected by the algorithm type; and the operating cost is mainly affected by the model scale.

[0041] The problem in actual use is that after adding new shelves, the planning accuracy rate drops to 95%, but the response time and CPU are normal. Therefore, give priority to increasing the weight of the factor that has the greatest influence on the accuracy rate, that is, the AI algorithm model structure.

[0042] Finally, add one layer of attention mechanism to the original three-layer neural network; continue to use the deep learning model; and mildly quantize the newly added layer.

[0043] In summary, by statistically analyzing the proportion of outliers in the recent N data sources and dynamically adjusting the weights, that is, if a sensor has a relatively high historical outlier rate, its weight will be reduced to avoid the interference of its noise on the overall data set, and the weights change dynamically according to the historical performance of the data sources, making the weighted multi-source data closer to the actual data quality and improving the data reliability for subsequent model training.

[0044] When the difference between the outlier and the threshold is less than the safety value, it is not directly deleted, but the number of similar situations in this group of data is counted. That is, if only a few samples in a batch of data are close to the threshold, it may be normal fluctuation and the data is retained; if the number exceeds the set threshold, it is determined as an outlier and deleted to avoid the loss of valid data caused by single fluctuations or noise, ensuring the integrity of the training data, especially in industrial scenarios to ensure the coverage ability of the model for the actual working conditions.

[0045] Example 2 is the second example of the present invention, which is different from the previous two examples in that: If the described function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0046] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, which can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch and execute instructions from the instruction execution system, apparatus, or device), or used in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device.

[0047] More specific examples (a non-exhaustive list) of computer-readable media include the following: electrical connections (electronic devices) having one or more wirings, portable computer disk cartridges (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). Additionally, the computer-readable media can even be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other media, followed by editing, interpretation, or otherwise processing it in a suitable manner if necessary, and then storing it in a computer memory.

[0048] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one or a combination of the following techniques well known in the art can be used: discrete logic circuits having logic gate circuits for implementing logical functions on data signals, application specific integrated circuits having suitable combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0049] Example 3 is the third embodiment of the present invention. This embodiment provides an online update system for an AI algorithm in an industrial Internet of Things, including a data acquisition module, a model construction module, and a model update module; The data acquisition module is responsible for collecting industrial information, obtaining a multi-source data set, and performing anomaly removal processing on the multi-source data set to obtain a normal data set; The model construction module is used to construct an AI algorithm model and deploy the constructed AI algorithm model to an AI algorithm node; The model update module is used to analyze the AI algorithm node and update the AI algorithm model based on the analysis result.

[0050] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not restrictive. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An online update method for an industrial Internet of Things AI algorithm, characterized by: The following steps are included: Collect industrial information and obtain multi-source data sets; Perform outlier processing on multi-source data sets to obtain normal data sets; Build an AI algorithm model using a normal data set and deploy the AI ​​algorithm model on an AI algorithm node; Analyze the AI ​​algorithm nodes and update the AI ​​algorithm model based on the analysis results.

2. The online update method of an industrial Internet of Things AI algorithm as claimed in claim 1, characterized in that: The step of acquiring the multi-source data set includes: Build a data collection platform based on the industrial application environment, and collect different types of data sources in each work cycle to obtain initial data; And perform data cleaning on the initial data; Feature extraction and feature selection are performed on the cleaned initial data to obtain a data feature set, and data sources with similar features in the data feature set are merged to obtain a multi-source data set.

3. The online update method of an industrial Internet of Things AI algorithm as described in claim 2 is characterized in that: The step of acquiring the normal data set includes: Preset weight value; Weighting the multi-source data set based on the weight value to obtain a weighted multi-source data set; Detect outliers on weighted multi-source datasets; Set an abnormal threshold and determine the difference between the abnormal value and the abnormal threshold; When the difference is less than the safety value, a secondary analysis is performed; When the difference is not less than the safety value, the corresponding outlier value is deleted; Wherein, the steps of the secondary analysis include: Set the quantity value; Get the number of differences in this group of multi-source data sets that are less than the safety value; When the acquired quantity is less than the set quantity value, the corresponding abnormal value will not be deleted; otherwise, the corresponding abnormal value will be deleted.

4. The online update method of an industrial Internet of Things AI algorithm as claimed in claim 3, characterized in that: When the weight value is preset, the following steps are included: Get the data type of the data source, and get the recent N data sources of different types; Calculate the proportion of outliers in the recent N data sources of different types, perform calculations, and output as weight values.

5. The online update method of the industrial Internet of Things AI algorithm as claimed in claim 4, characterized in that: The steps of constructing the AI ​​algorithm model include: The normal data set is divided into a training set and a test set according to a preset ratio through a cross-validation algorithm; Among them, the training set and test set are used for AI algorithm model training and AI algorithm model testing respectively to obtain the AI ​​algorithm model.

6. The online update method of an industrial Internet of Things AI algorithm as claimed in claim 4, characterized in that: The steps to analyze AI algorithm nodes include: Determine the indicator system according to the actual needs of users and determine the set of factors that affect the indicator system; Calculate the influence of each sub-factor of the factor set on the indicator system; Evaluate the impact of each sub-factor on the indicator system and obtain the online update results of the AI ​​algorithm model; Perform model tuning on the AI ​​algorithm model based on online update results.

7. The online update method of the industrial Internet of Things AI algorithm as claimed in claim 4, characterized in that: When evaluating the impact of various factors on the indicator system, the following steps are included: Through the indicator system of actual needs, obtain the sub-factor with the greatest impact corresponding to the indicator system, and adjust the weight of the sub-factor to the maximum; Then obtain the factor with the second highest influence, and adjust the weight of this sub-factor to the second highest; Repeat this process until all sub-factors have been cycled through.

8. An online update system for an industrial Internet of Things AI algorithm, applying an online update method for an industrial Internet of Things AI algorithm as described in any one of claims 1 to 7, characterized in that: It includes a data acquisition module, a model building module, and a model updating module; The data acquisition module is responsible for collecting industrial information, obtaining multi-source data sets, and performing anomaly removal processing on the multi-source data sets to obtain normal data sets; The model building module is used to build an AI algorithm model and deploy the built AI algorithm model to the AI ​​algorithm node; The model updating module is used to analyze the AI ​​algorithm nodes and update the AI ​​algorithm model based on the analysis results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the online update method of the industrial Internet of Things AI algorithm described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of an online update method of an industrial Internet of Things AI algorithm described in any one of claims 1 to 7 are implemented.

Images