Online updating method and system of industrial internet of things AI algorithm

Abstract

The application discloses an online updating method and system of an industrial internet of things AI algorithm, and belongs to the technical field of AI algorithms, and comprises the following steps: collecting industrial information, and obtaining a multi-source data set; performing abnormality 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; analyzing the AI algorithm node, and updating the AI algorithm model based on the analysis result. The application dynamically adjusts the weight by counting the proportion of abnormal values of the data source in the last N times, that is, the weight of a sensor with a high historical abnormality rate is reduced, so that the noise of the sensor does not interfere with the overall data set, and the weight dynamically changes with the historical performance of the data source, so that the weighted multi-source data is closer to the actual data quality, and the data reliability of subsequent model training is improved.

Description

Technical Field

[0001] This invention relates to the field of AI algorithm technology, specifically to an online update method and system for industrial IoT AI algorithms. Background Technology

[0002] In the Industrial Internet of Things (IIoT), multi-source data (such as sensor data, equipment logs, and environmental parameters) are typically characterized by heterogeneity, high noise levels, and high real-time requirements. Traditional methods often use fixed weights to weight multi-source data, which cannot adapt to the historical differences in anomaly rates across different data sources. For example, older sensors may have a consistently high anomaly rate, but if the weights are fixed, their effective data may be underestimated or outliers may be misclassified. Secondly, in industrial scenarios, certain transient fluctuations may be misjudged as anomalies, such as the instantaneous noise during equipment startup. Directly deleting such data may result in the loss of effective data, affecting the integrity of model training. Summary of the Invention

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

[0004] Therefore, the technical problem solved by this invention is that traditional methods use a fixed weight allocation method and a single threshold deletion mechanism.

[0005] To address the aforementioned technical problems, this invention provides the following technical solution: an online update method for an industrial IoT AI algorithm, comprising the following steps:

[0006] Collect industrial information and obtain multi-source datasets;

[0007] Anomaly removal is performed on the multi-source dataset to obtain a normal dataset;

[0008] AI algorithm models are built using normal datasets and then deployed on AI algorithm nodes.

[0009] Analyze the AI ​​algorithm nodes and update the AI ​​algorithm model based on the analysis results.

[0010] As a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the step of acquiring the multi-source dataset includes:

[0011] A data acquisition platform is built based on the industrial application environment, and initial data is obtained by collecting data from different types of data sources in each work cycle.

[0012] And perform data cleaning on the initial data;

[0013] Feature extraction and feature selection are performed on the cleaned initial data to obtain a data feature set. Data sources with similar features in the data feature set are merged to obtain a multi-source dataset.

[0014] In a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the step of obtaining the normal dataset includes:

[0015] Preset weight values; weight the multi-source dataset based on the weight values ​​to obtain a weighted multi-source dataset; detect outliers in the weighted multi-source dataset; set an outlier threshold and determine the difference between the outlier and the outlier threshold; if the difference is less than the safety value, perform a second analysis; if the difference is not less than the safety value, delete the corresponding outlier.

[0016] The secondary analysis steps include: setting a quantity value; obtaining the number of differences in the multi-source dataset that are less than the safety value; if the obtained number is less than the set quantity value, then the corresponding outlier is not deleted; otherwise, the corresponding outlier is deleted.

[0017] By employing a secondary analysis mechanism, we can avoid accidental data deletion due to misjudgment of a single outlier, thereby improving the accuracy of anomaly detection.

[0018] As a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the preset weight value includes the following steps:

[0019] Get the data type of the data source, and obtain the near Nth data source of different types;

[0020] Calculate the percentage of outliers from the most recent N data sources of different types, and output the weight values.

[0021] The weights are calculated based on the proportion of outliers in historical data, prioritizing data sources with high stability to improve the credibility of multi-source data fusion. Furthermore, the weights change dynamically with the historical performance of the data sources, making the model focus more on high-reliability data and reducing interference from low-quality data sources.

[0022] As a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the AI ​​algorithm model construction step includes:

[0023] The normal dataset is divided into training and test sets according to a preset ratio using a cross-validation algorithm.

[0024] The training set and the test set are used for training the AI ​​algorithm model and testing the AI ​​algorithm model, respectively, to obtain the AI ​​algorithm model.

[0025] In a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the step of analyzing the AI ​​algorithm nodes includes:

[0026] The indicator system is determined based on the actual needs of users, and the set of factors that affect the indicator system is determined.

[0027] Calculate the degree of influence of each sub-factor of the factor set on the indicator system;

[0028] The impact of each sub-factor on the indicator system is evaluated to obtain the online update results of the AI ​​algorithm model;

[0029] The AI ​​algorithm model is fine-tuned based on the online update results.

[0030] As a preferred embodiment of the online update method for an industrial IoT AI algorithm described in this invention, the evaluation of the influence of each sub-factor on the indicator system includes the following steps:

[0031] Based on the indicator system of actual needs, identify the sub-factor with the greatest influence corresponding to the indicator system and adjust its weight to the maximum; then identify the next most influential factor and adjust its weight to the second largest; repeat this process until all sub-factors have been cycled through.

[0032] By adjusting the weights according to the degree of influence of sub-factors on the indicators, the AI ​​algorithm model can focus more on key factors, improve the relevance of updates, and ensure that core factors dominate the model behavior through the priority ranking of weight adjustments.

[0033] Another objective of this invention is to provide an online update system for industrial IoT AI algorithms.

[0034] To address the aforementioned technical problems, this invention provides the following technical solution: an online update system for industrial IoT AI algorithms, comprising a data acquisition module, a model building module, and a model update module;

[0035] The data acquisition module is responsible for collecting industrial information, obtaining multi-source datasets, and performing anomaly removal processing on the multi-source datasets to obtain normal datasets.

[0036] The model building module is used to build AI algorithm models and deploy the built AI algorithm models to AI algorithm nodes;

[0037] The model update module is used to analyze AI algorithm nodes and update the AI ​​algorithm model based on the analysis results.

[0038] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, characterized in that the processor executes the computer program to implement the steps of the online update method for an industrial Internet of Things AI algorithm.

[0039] The present invention provides a computer-readable storage medium having a computer program stored thereon, characterized in that, when the computer program is executed by a processor, it implements the steps of the online update method for an industrial Internet of Things AI algorithm.

[0040] The beneficial effects of this invention are:

[0041] By statistically analyzing the proportion of outliers from the most recent N data sources, the weights are dynamically adjusted. If a sensor has a high historical outlier rate, its weight will be reduced to avoid its noise interfering with the overall dataset. Furthermore, the weights change dynamically with 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.

[0042] When the difference between an outlier and the threshold is less than the safety value, it is not directly deleted. Instead, the number of similar cases in the data set 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 judged as an anomaly and deleted. This avoids the loss of effective data due to a single fluctuation or noise, and ensures the integrity of the training data, especially in industrial scenarios to ensure the model's coverage of real working conditions. Attached Figure Description

[0043] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0044] Figure 1 The above is a flowchart of an online update method for an industrial IoT AI algorithm provided in one embodiment of the present invention. Detailed Implementation

[0045] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0046] Example 1, referring to Figure 1This is the first embodiment of the present invention, which provides an online update method for an industrial IoT AI algorithm, including:

[0047] S1. Collect industrial information and obtain multi-source datasets.

[0048] The steps for obtaining multi-source datasets include:

[0049] A data acquisition platform is built based on the industrial application environment, and initial data is obtained by collecting data from different types of data sources in each work cycle.

[0050] In this embodiment, for example in an automobile manufacturing plant, data from production line equipment, environment, and sensors are collected; temperature sensors (to monitor equipment temperature), pressure sensors (to monitor hydraulic system pressure), vibration sensors (to monitor robotic arm vibration), PLC controllers (to record equipment status codes), environmental temperature and humidity sensors, etc. are installed, and a SCADA system (data acquisition and monitoring system) is deployed as a data transfer station.

[0051] And perform data cleaning on the initial data;

[0052] If a sensor has no data in a certain acquisition cycle (e.g., due to a temperature sensor malfunction), the data is filled by linear interpolation of the previous valid value; for example, if a temperature sensor reading is 150℃, which is much higher than the normal range of 50-80℃, key features are extracted from the cleaned data, and important features are selected.

[0053] Feature extraction and selection are performed on the cleaned initial data to obtain a data feature set. Data sources with similar features in the feature set are merged to obtain a multi-source dataset. Conventional analysis of variance (ANOVA) is used to screen features that have a significant impact on equipment failure prediction. Finally, data sources with similar features are merged to reduce redundancy and improve subsequent processing efficiency.

[0054] S2. Perform anomaly removal processing on the multi-source dataset to obtain a normal dataset.

[0055] Get the data type of the data source, and obtain the near Nth data source of different types;

[0056] Calculate the percentage of outliers from the Nth data source of different types, and output the weight values.

[0057] In one optional embodiment, a CNC machining center needs to monitor the tool wear status in real time to prevent workpiece scrap due to tool breakage. The system integrates three types of sensors: a temperature sensor to monitor the contact temperature between the tool and the workpiece; a vibration sensor to detect the amplitude of spindle vibration; and a current sensor to collect the spindle motor current.

[0058] The data sources are temperature, vibration, and current. From the most recent N data points, the most recent 100 samples from each sensor type are collected; in this embodiment, N=100. The number of anomalies detected are as follows: temperature 12 times (12%), vibration 25 times (25%), and current 5 times (5%).

[0059] The lower the percentage of anomalies, the higher the data reliability and the greater the weight; the calculation formula is:

[0060] ;

[0061] It can be concluded that the weighting value is 0.29 for temperature, 0.14 for vibration, and 0.57 for current.

[0062] The multi-source dataset is weighted based on the weight values ​​to obtain the weighted multi-source dataset;

[0063] The multi-source dataset is scaled up or down according to weights, i.e., temperature value × 0.29, vibration value × 0.14, and current value × 0.57.

[0064] Detect outliers in the weighted multi-source dataset;

[0065] Set an anomaly threshold and determine the difference between an outlier and the outlier threshold;

[0066] If the difference is less than the safety value, a second analysis is performed; if the difference is not less than the safety value, the corresponding outlier is deleted.

[0067] In this embodiment, for example, during temperature detection, the detected abnormal values ​​are: 80.2, 81.5, 88.6, 80.7, 90.4, 88.9, 89.1, 88.3, ​​91.1, 91.6, 79.1, and 90.9.

[0068] The abnormal threshold is set to 75. The difference between all abnormal values ​​and the abnormal threshold is calculated sequentially. If the safety value is set to 5, only one group is found to have a difference less than the safety value, namely 79.1. All the remaining abnormal values ​​are deleted.

[0069] The secondary analysis steps include:

[0070] Set quantity value;

[0071] Find the number of differences in this multi-source dataset that are less than the safe value;

[0072] If the quantity obtained is less than the set quantity value, the corresponding abnormal value will not be deleted; otherwise, the corresponding abnormal value will be deleted.

[0073] In this embodiment, the set quantity is 3, and after the previous step, there is only 1 remaining difference. Therefore, the corresponding outlier is not deleted, that is, 79.1 is retained.

[0074] For example, in the previous step, there were 4 remaining differences, such as 79.1, 80.2, 80.7, and 81.5. Therefore, the number of remaining differences is not less than the set value, so all 4 outliers are deleted.

[0075] S3. Build an AI algorithm model using a normal dataset and deploy the AI ​​algorithm model on the AI ​​algorithm node.

[0076] The steps for constructing the AI ​​algorithm model include:

[0077] The normal dataset is divided into training and test sets according to a preset ratio using a cross-validation algorithm.

[0078] The training set and the test set are used for training the AI ​​algorithm model and testing the AI ​​algorithm model, respectively, to obtain the AI ​​algorithm model.

[0079] In this embodiment, to avoid leakage of future information from the time series data, the data is divided into blocks according to processing batches. The first 7 processing batches, each with approximately 10,000 records, are used as the training set; the last 3 processing batches are used as the test set.

[0080] S4. Analyze the AI ​​algorithm nodes and update the AI ​​algorithm model based on the analysis results.

[0081] The steps for analyzing AI algorithm nodes include:

[0082] The indicator system is determined based on the actual needs of users, and the set of factors that affect the indicator system is determined.

[0083] The indicator system includes: the accuracy 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 structure of the AI ​​algorithm model, the algorithm of the AI ​​algorithm model, and the scale of the AI ​​algorithm model.

[0084] Calculate the degree of influence of each sub-factor of the factor set on the indicator system;

[0085] By using an indicator system based on actual needs, identify the sub-factors with the greatest impact corresponding to the indicator system, and adjust the weight of these sub-factors to the maximum.

[0086] Next, identify the factors with the next highest influence and adjust the weight of this sub-factor to the second highest.

[0087] Repeat this process until all sub-factors have been cycled through;

[0088] Obtain the online update results of the AI ​​algorithm model; optimize the AI ​​algorithm model based on the online update results.

[0089] In one alternative embodiment, for example, an e-commerce warehouse uses 100 logistics robots for goods sorting, requiring real-time planning of optimal routes. Currently, the following problems are faced: congestion during peak hours, robot response delays leading to goods backlog; rapid battery consumption, complex path algorithms increasing CPU load and shortening battery life; slow adaptation to new shelving, with the error rate of the original planning model increasing after adding new shelving.

[0090] First, define the indicator system and factor set: the accuracy of the AI ​​algorithm model is ≥98%, the response time of the AI ​​algorithm model is ≤200ms, and the operating cost of the AI ​​algorithm model, i.e., CPU usage, is ≤15%. The structure of the AI ​​algorithm model includes the number of neural network layers and the attention mechanism; the AI ​​algorithm model includes A* algorithm, Dijkstra algorithm, and deep learning model; and the scale of the AI ​​algorithm model includes the number of parameters.

[0091] By analyzing historical data, the impact of various factors on the indicators is as follows: Increasing the number of neural network layers in an AI algorithm model results in a 10% increase in accuracy, but a 30ms increase in response time and a 5% increase in CPU usage. Switching to the A* algorithm reduces response time by 50ms but decreases accuracy by 3%. Quantizing and compressing the model reduces CPU usage by 8% but decreases accuracy by 2%. Therefore, accuracy is primarily affected by model structure; response time is primarily affected by algorithm type; and operating costs are primarily affected by model size.

[0092] The problem in practical use is that after adding new shelves, the planning accuracy drops to 95%, but the response time and CPU usage remain normal. Therefore, the priority should be to increase the weight of the factors that have the greatest impact on accuracy, namely the AI ​​algorithm model structure.

[0093] Ultimately, an attention mechanism was added to the original three-layer neural network; a deep learning model was continued; and the new layer was quantized.

[0094] In summary, by statistically analyzing the proportion of outliers from the most recent N data sources, the weights are dynamically adjusted. That is, if a sensor has a high historical outlier rate, its weight will be reduced to avoid its noise interfering with the overall dataset. Furthermore, the weights change dynamically with 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.

[0095] When the difference between an outlier and the threshold is less than the safety value, it is not directly deleted. Instead, the number of similar cases in the data set 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 judged as an anomaly and deleted. This avoids the loss of effective data due to a single fluctuation or noise, and ensures the integrity of the training data, especially in industrial scenarios to ensure the model's coverage of real working conditions.

[0096] Example 2 is the second embodiment of the present invention, which differs from the previous two embodiments in that:

[0097] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0098] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0099] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0100] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0101] Example 3 is the third embodiment of the present invention. This embodiment provides an online update system for industrial IoT AI algorithms, including a data acquisition module, a model building module, and a model update module.

[0102] The data acquisition module is responsible for collecting industrial information, obtaining multi-source datasets, and performing anomaly removal processing on the multi-source datasets to obtain normal datasets.

[0103] The model building module is used to build AI algorithm models and deploy the built AI algorithm models to AI algorithm nodes;

[0104] The model update module is used to analyze AI algorithm nodes and update the AI ​​algorithm model based on the analysis results.

[0105] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. An online update method for an industrial IoT AI algorithm, characterized in that: Includes the following steps, Collect industrial information and obtain multi-source datasets; Anomaly removal is performed on the multi-source dataset to obtain a normal dataset; AI algorithm models are built using normal datasets and then deployed on AI algorithm nodes. Analyze the AI ​​algorithm nodes and update the AI ​​algorithm model based on the analysis results; The steps for obtaining the normal dataset include: Preset weight values; The multi-source dataset is weighted based on the weight values ​​to obtain the weighted multi-source dataset. Detect outliers in the weighted multi-source dataset; Set an anomaly threshold and determine the difference between an outlier and the outlier threshold; If the difference is less than the safety value, a second analysis is performed; if the difference is not less than the safety value, the corresponding outlier is deleted. The secondary analysis steps include: Set quantity value; Find the number of differences in this multi-source dataset that are less than the safe value; If the quantity obtained is less than the set quantity value, the corresponding abnormal value will not be deleted; otherwise, the corresponding abnormal value will be deleted. When the weight value is preset, the following steps are included: Get the data type of the data source, and obtain the near Nth data source of different types; Calculate the percentage of outliers from the Nth data source of different types, and output the weight values.

2. The online update method for an industrial IoT AI algorithm as described in claim 1, characterized in that: The steps for obtaining the multi-source dataset include: A data acquisition platform is built based on the industrial application environment, and initial data is obtained by collecting data from different types of data sources in each work cycle. 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. Data sources with similar features in the data feature set are merged to obtain a multi-source dataset.

3. The online update method for an industrial IoT AI algorithm as described in claim 2, characterized in that: The steps for constructing the AI ​​algorithm model include: The normal dataset is divided into training and test sets according to a preset ratio using a cross-validation algorithm. The training set and the test set are used for training the AI ​​algorithm model and testing the AI ​​algorithm model, respectively, to obtain the AI ​​algorithm model.

4. The online update method for an industrial IoT AI algorithm as described in claim 3, characterized in that: The steps for analyzing AI algorithm nodes include: The indicator system is determined based on the actual needs of users, and the set of factors that affect the indicator system is determined. Calculate the degree of influence of each sub-factor of the factor set on the indicator system; The impact of each sub-factor on the indicator system is evaluated to obtain the online update results of the AI ​​algorithm model; The AI ​​algorithm model is fine-tuned based on the online update results.

5. The online update method for an industrial IoT AI algorithm as described in claim 4, characterized in that: When assessing the impact of each factor on the indicator system, the following steps are included: By using an indicator system based on actual needs, identify the sub-factors with the greatest impact corresponding to the indicator system, and adjust the weight of these sub-factors to the maximum. Next, identify the factors with the next 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.

6. An online update system for an industrial IoT AI algorithm, employing the online update method for an industrial IoT AI algorithm as described in any one of claims 1 to 5, characterized in that, It includes a data acquisition module, a model building module, and a model update module; The data acquisition module is responsible for collecting industrial information, obtaining multi-source datasets, and performing anomaly removal processing on the multi-source datasets to obtain normal datasets. The model building module is used to build AI algorithm models and deploy the built AI algorithm models to AI algorithm nodes; The model update module is used to analyze AI algorithm nodes and update the AI ​​algorithm model based on the analysis results.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the online update method for an industrial IoT AI algorithm according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the online update method for an industrial IoT AI algorithm according to any one of claims 1 to 5.

Images