An internet of things device data collection optimization method and system based on deep learning

Abstract

The present application relates to the technical field of data intelligent acquisition, in particular to an Internet of Things device data acquisition optimization method and system based on deep learning, comprising: acquiring multi-dimensional real-time operation data of an Internet of Things device, pre-processing the real-time operation data based on device identification to obtain standard operation data; performing parameter importance analysis on the standard operation data based on a deep learning algorithm to obtain a parameter acquisition priority; dynamically adjusting and verifying the parameter acquisition priority based on a working condition change law to obtain dynamic acquisition strategy data; extracting a working condition switching and parameter abnormal event set from the standard operation data, and iteratively optimizing and verifying the effect of the parameter abnormal event set based on the dynamic acquisition strategy data to obtain strategy iteration data. The present application improves the accuracy and effectiveness of device data acquisition through a dynamic acquisition scheme.

Description

Technical Field

[0001] This invention relates to the field of intelligent data acquisition technology, and in particular to a method and system for optimizing data acquisition of Internet of Things (IoT) devices based on deep learning. Background Technology

[0002] With the rapid development of the Industrial Internet of Things (IIoT), IoT devices are widely used in all aspects of industrial production. The collection and analysis of multi-dimensional operational data from these devices has become a core means to ensure stable production and improve operational efficiency. Especially in precision manufacturing equipment scenarios such as molding machines, the real-time nature and accuracy of equipment operating parameters (such as injection pressure, mold temperature, and cycle time) directly affect product molding quality and production efficiency. Therefore, extremely high requirements are placed on the accuracy and relevance of data collection.

[0003] Existing IoT device data acquisition technologies mostly employ fixed-frequency acquisition modes, failing to fully consider the differences in parameter variation patterns under various operating conditions, thus exhibiting significant limitations. Indiscriminate acquisition of all parameters leads to insufficient resources for core parameter acquisition and redundant acquisition of non-core parameters, resulting in wasted network bandwidth and storage resources, while also increasing data processing pressure. Furthermore, the acquired acquisition strategies lack dynamic adjustment capabilities, failing to adapt to different operating conditions such as device production / standby, and struggling to respond to sudden scenarios such as abnormal parameter fluctuations. This results in insufficient effectiveness and timeliness of data acquisition, failing to provide reliable data support for device operation and maintenance and production optimization.

[0004] Furthermore, existing technologies suffer from issues such as inconsistent formats and inaccurate outlier filtering in the data preprocessing stage, which can easily lead to biased results in subsequent data analysis. The lack of a scientific quantitative model for parameter importance assessment and the subjective and arbitrary prioritization of data collection further reduce the practical value of the collected data. Summary of the Invention

[0005] This invention uses deep learning to extract features and optimize strategies for device operation data, generating a dynamic data acquisition scheme to improve the accuracy and effectiveness of device data acquisition.

[0006] The technical solution proposed in this invention is: a data acquisition optimization method for IoT devices based on deep learning, the method comprising:

[0007] Collect multi-dimensional real-time operating data from IoT devices, preprocess the real-time operating data based on device identification, and obtain standard operating data;

[0008] Parameter importance analysis is performed on standard operating data based on deep learning algorithms to obtain parameter collection priorities;

[0009] Based on the changing patterns of operating conditions, the priority of parameter acquisition is dynamically adjusted and verified to obtain dynamic acquisition strategy data.

[0010] Extract the set of operating condition switching and parameter anomaly events from the standard operating data, and perform strategy iteration optimization and effect verification on the set of parameter anomaly events based on the dynamically acquired strategy data to obtain strategy iteration data;

[0011] Based on the system's interaction logic, a visualization interface is generated corresponding to the dynamically collected strategy data, strategy iteration data, and real-time running data, and the visualization interface is displayed in a linked manner.

[0012] Preferably, the specific process for obtaining the multi-dimensional real-time operational data is as follows:

[0013] The operating status and core parameters of IoT devices are collected from multiple dimensions to obtain raw data.

[0014] Based on the device business logic, the raw collected data is associated with device identifiers and encapsulated to obtain encapsulated operational data.

[0015] The standardized interface based on the Internet of Things system will encapsulate the operational data and upload it to the pre-set central data platform;

[0016] The system receives real-time operation messages from the central data platform via message subscription and parses these messages to obtain parsed operation data.

[0017] The parsed running data is correlated with the device running context to obtain multi-dimensional real-time running data.

[0018] Preferably, the specific process for obtaining the standard operating data is as follows:

[0019] The system categorizes and classifies equipment and manufacturers based on multi-dimensional real-time operational data, and adds equipment tags to the multi-dimensional real-time operational data based on the classification results.

[0020] The integrity of real-time operation data is verified based on device tags, and the data is deduplicated based on device identifiers and data time difference thresholds after integrity verification to obtain deduplicated operation data.

[0021] Linear interpolation is used to fill in missing values ​​of continuous parameters in the deduplication running data, and mode completion is used to fill in missing values ​​of discrete parameters, thus obtaining the completed running data;

[0022] Based on 3 The principle is to perform outlier detection on the completed operation data, obtain outlier data, and replace the outlier data with the mean of normal data in the same time period to obtain noise-reduced operation data;

[0023] All numerical parameters in the noise reduction operation data are rounded to two decimal places. Based on the preset field template, the noise reduction operation data is standardized to obtain standard operation data.

[0024] Preferably, the specific process for obtaining the parameter acquisition priority is as follows:

[0025] Using the device identifier as the primary key, extract the parameter operation dataset corresponding to each device from the standard operation data, and extract the parameter time series events from each parameter operation dataset to obtain the parameter event set;

[0026] The parameter event set is sorted according to time order to obtain the parameter event sequence set, and the correlation coefficient between each parameter and business indicator is calculated based on the core objectives of the business scenario.

[0027] The update frequency of each parameter per unit time under normal production conditions is statistically analyzed, and the percentage of parameter update frequency is obtained through linear normalization.

[0028] The parameter importance score is calculated by weighting the correlation coefficient and the parameter update frequency ratio. The K-means clustering algorithm is used to classify the parameter importance score to obtain the classification results of core parameters and non-core parameters.

[0029] Assign collection priorities to different types of parameters based on business scenario requirements, and generate parameter collection priority data.

[0030] Preferably, the specific process for obtaining the dynamic acquisition strategy data is as follows:

[0031] Extract the operating condition switching events from the dataset of each parameter to obtain the operating condition event set;

[0032] The set of working condition events is sorted according to time order to obtain the set of working condition event sequences, and the frequency and fluctuation amplitude of each parameter under different working conditions are calculated.

[0033] Based on the change frequency and fluctuation amplitude, the frequency correction coefficient corresponding to the working condition is determined. Combined with the frequency adjustment coefficient corresponding to the parameter acquisition priority, the adjusted acquisition frequency of each parameter under different working conditions is calculated.

[0034] Several devices from different manufacturers were selected as pilot devices, and real-time data was collected at the adjusted collection frequency. The efficiency of the collected data and the transmission delay were calculated.

[0035] The acquisition strategy is validated based on preset efficiency thresholds and transmission delay thresholds. If the threshold requirements are not met, the frequency correction coefficient is fine-tuned and the trial is repeated until the requirements are met, thus obtaining dynamic acquisition strategy data.

[0036] Preferably, the specific process for obtaining the strategy iteration data is as follows:

[0037] Based on the dynamic acquisition strategy data and parameter anomaly event set, multi-dimensional time series features are extracted from standard operating data to obtain parameter time series features;

[0038] Based on the parameter time series characteristics and parameter anomaly event set, graph structure modeling and anomaly context modeling are performed on standard operating data to obtain parameter association graph structure;

[0039] The parameter anomaly event set is used as anomaly label, and the anomaly label is used as a supervision signal to train the LSTM neural network model to obtain the parameter anomaly analysis model.

[0040] Based on the parameter anomaly analysis model, the parameter correlation graph structure is forward-propagated to obtain the anomaly contribution weights.

[0041] The parameter acquisition strategy is iteratively adjusted based on the abnormal contribution weight, and the effectiveness of the adjusted strategy is verified to obtain strategy iteration data.

[0042] Preferably, the strategy iteration and optimization process includes strategy version management and a failure rollback mechanism, the specific acquisition process of which is as follows:

[0043] Semantic version numbers are used to identify each version of the data collection strategy, and the creation time, modification content, responsible person and scope of effective devices of each version strategy are recorded to form a complete version iteration record;

[0044] Before strategy iteration, the currently effective collection strategy is automatically backed up as a backup version for rollback.

[0045] If the iterated strategy still fails to meet the preset indicators after several consecutive data collection and verifications, the automatic rollback mechanism will be triggered to immediately switch to the backup original strategy version to ensure that the data collection business is not interrupted.

[0046] After the rollback operation is completed, the reason for the iteration failure, the rollback time, and the information of the affected devices will be recorded in the system iteration log, and an alarm notification will be pushed to the operation and maintenance personnel for manual intervention and investigation.

[0047] Preferably, the strategy iterative optimization process also includes a device offline data acquisition and security protection mechanism, the specific acquisition process of which is as follows:

[0048] When the device is offline, the local caching function is automatically activated to collect and store core parameter data and corresponding timestamps at a preset frequency.

[0049] After the device is back online, cached data will be uploaded incrementally in timestamp order first. Once the upload is complete, the normal data collection rhythm will resume.

[0050] If the offline time exceeds the threshold, a sharded upload mechanism is adopted to avoid data transmission congestion, and the re-collected data needs to go through standardization processing and validity verification again;

[0051] The TLS1.3 encryption protocol is used during data transmission, and the AES-256 encryption algorithm is used for data storage; The system adopts the RBAC permission control model to distinguish the operation permissions of administrators, operation and maintenance personnel, and ordinary users. Secondary identity verification and operation logs need to be recorded for key policy adjustments and data query operations.

[0052] The present invention also provides an Internet of Things device data collection optimization system based on deep learning, and the system is used to execute the Internet of Things device data collection optimization method based on deep learning described above.

[0053] The present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the Internet of Things device data collection optimization method based on deep learning described above.

[0054] Advantages of the present invention:

[0055] 1. By establishing a secondary classification index, a standardized mapping dictionary, and a multi-dimensional data cleaning mechanism, problems such as inconsistent data formats, interference of redundant and invalid data, and index conflicts of Internet of Things device data are effectively solved, and the data standardization and availability are greatly improved. The high-quality data after standardization processing not only provides a reliable basis for subsequent parameter analysis and strategy formulation, but also realizes fast data retrieval and classification through hash indexing and uniqueness verification, significantly reducing the data management cost.

[0056] 2. Relying on the deep learning model to挖掘 the importance of parameters and the working condition change rules, a priority grading and dynamic frequency adjustment mechanism is constructed, breaking the limitations of the traditional fixed-frequency collection mode. The strategy of preferentially collecting core parameters and adjusting non-core parameters as needed not only avoids resource waste caused by indiscriminate collection, but also ensures the pertinence and timeliness of data collection through working condition adaptation and real-time correction, significantly improving the collection efficiency and resource utilization rate, and at the same time reducing the processing pressure of invalid data on the system.

[0057] 3. Integrate precise instruction issuance, multi-terminal linked display, strategy iterative optimization, and complete operation and maintenance guarantee (failure rollback, offline re-collection, security protection) to form a full-link closed-loop management system. It not only realizes real-time visualization monitoring of the collected data and dynamic iteration of the strategy, ensuring the stable and efficient operation of the device, but also improves the system reliability and fault tolerance through log tracing, version management, and exception response mechanisms, providing comprehensive data support for device operation and maintenance optimization and system upgrade. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 A flowchart illustrating a deep learning-based optimization method for data acquisition in IoT devices;

[0059] Figure 2 This is a flowchart illustrating the strategy transformation process of a deep learning-based optimization method for data acquisition in IoT devices. Detailed Implementation

[0060] The following description is intended to disclose the present invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art. The basic principles of the invention defined in the following description can be applied to other embodiments, modifications, improvements, equivalents, and other technical solutions that do not depart from the spirit and scope of the invention.

[0061] It is understood that the term "a" should be understood as "at least one" or "one or more," that is, in one embodiment, the number of an element can be one, while in another embodiment, the number of the element can be multiple, and the term "a" should not be understood as a limitation on the number.

[0062] like Figure 1 and Figure 2 As shown, multi-dimensional real-time operational data from IoT devices is collected, classified, standardized and mapped for parameters, outlier removed, and format unified, outputting a standardized multi-dimensional data table. This effectively solves the problems of inconsistent device data formats and interference from redundant and invalid data, providing a high-quality and standardized data foundation for subsequent data applications. Based on the standardized multi-dimensional data table, deep learning is used to extract core parameter features of the devices, generate a parameter collection priority list, and initially determine the collection strategy. This achieves precise hierarchical parameter collection, avoiding resource waste caused by indiscriminate collection and improving the targeting and efficiency of data collection. Combining the collection priority configuration table with historical time-series data, the system mines the patterns of parameter changes under operating conditions, dynamically adjusts and verifies the collection strategy, and outputs a dynamic collection strategy table. This adapts the collection strategy to changes in device operating conditions, ensuring the real-time nature and effectiveness of data collection and further optimizing resource allocation. Collection commands are issued according to the dynamic collection strategy table, and real-time data is displayed through multiple front-end interfaces. Strategy iteration is triggered based on operating condition changes or anomaly prompts. This achieves precise execution of collection commands and real-time visual monitoring of data. At the same time, the collection effect is continuously optimized through dynamic strategy iteration, ensuring stable and efficient device operation.

[0063] Furthermore, multi-dimensional real-time operational data from IoT devices is collected, and after classification, parameter standardization mapping, outlier removal, and format unification processing, a standardized multi-dimensional data table is output. The specific process is as follows:

[0064] Collect multi-dimensional real-time operational data from IoT devices, specifically including: Device ID (a unique identifier for the IoT device, data type: string), Device Name (the official name of the device, clearly defining its functional attributes, data type: string), Manufacturer Category (the name of the device manufacturer, used to differentiate parameter characteristics of different brands of devices, data type: string), Device Classification (categorized by device function and department, used for classified management, data type: string), IP Address (the unique network address the device uses for data transmission and command issuance, data type: IPv4 format string), Operating Status (the current operating mode of the device, reflecting the device's workload, data type: enumerated value), Raw Parameter Identifier (the parameter code output from the device's underlying layer, used by the system to identify raw data fields, data type: string), and Core Operating Parameters (key indicators reflecting the core operating status of the device, including cycle time (the time it takes for the device to complete one full production process)). Time (data type: numeric, unit: seconds); Injection pressure (pressure value during equipment injection, affecting product molding quality, data type: numeric, unit: MPa); Mold temperature (real-time temperature of the equipment mold, a key environmental parameter for product molding, data type: numeric, unit: °C); Temperature of each section (zone temperature of the equipment heating section, refining temperature monitoring dimensions, data type: numeric, unit: °C); Injection speed (progress speed of the equipment injection process, including maximum injection speed, pressure holding switching speed, etc., data type: numeric, unit: mm / s); Mold opening and closing time (time taken to open and close the equipment mold, reflecting the efficiency of the equipment's mechanical actions, data type: numeric, unit: ms); Network latency (NL) (time difference between data transmission from the equipment to the system, reflecting network transmission status, data type: numeric, unit: ms); Data update timestamp (specific time of equipment parameter collection and upload, used to mark data timeliness, data type: time format string).

[0065] After data collection, the data undergoes multi-step processing. First, a secondary classification index is established based on equipment category (molding machine / molding group) and manufacturer category. Then, a mapping table between equipment number, category, and manufacturer is constructed using the following formula:

[0066] ;

[0067] in For the first The original data of the device For the first Classification of equipment For the first Manufacturer categories of Taiwanese equipment. For the first The serial number of the equipment. This indicates string concatenation. A hash function (using the SHA-256 algorithm) is used to generate a unique classification index, enabling fast data retrieval and categorization. The hash value is a 16-bit random number. In the event of a hash collision, double hash verification is used (if the first hash result is a collision, a new random number is generated and the SHA-256 is recalculated). At the same time, a mapping table between the hash value and the original information is established, and a uniqueness check is performed every day at midnight to ensure that the index is unique and conflict-free. The hash value length is fixed at 32 bytes, and uniqueness verification is performed through cyclic redundancy check (CRC32) to avoid duplicate indexes.

[0068] Establish a mapping dictionary between original parameter identifiers, standardized parameter names, and parameter units. Determine the standardized name and unit by matching the feature fields in the original parameter identifier using regular expressions, and complete the mapping according to the formula: ,in For the first Original parameter identifier, For the corresponding standardized parameter name, The corresponding parameter unit.

[0069] Use 3 The principle is to identify and remove invalid data, and to determine outliers according to the formula:

[0070] ;

[0071] in For the first The first piece of equipment The original values ​​of the parameters, This is the average of the historical data for this parameter. This is the standard deviation of the historical data for this parameter; outliers are directly removed. Removed outliers must be recorded in the anomaly database, including the original data value, collection time, device number, and anomaly type. This database allows for tracing and querying by time range and device number, providing data support for equipment fault diagnosis.

[0072] All numerical parameters are rounded to two decimal places using the following formula:

[0073] ;

[0074] in These are the standardized parameter values. For high-precision measurement parameters (such as mold temperature), three decimal places are retained before standardization for intermediate calculations, and two decimal places are retained in the final output to avoid precision loss; a numerical range verification rule is established, and values ​​exceeding the reasonable physical range are checked even if no error is triggered. Any abnormal data must be marked as suspicious and manually reviewed.

[0075] The final result is a standardized multidimensional data table. Table fields include device number, device name, manufacturer category, device classification, IP address, operating status, original parameter identifier, standardized parameter name, parameter unit, parameter value, and data update timestamp. A sharded storage strategy is adopted, with data partitioned according to device category and time range. Data from the past 30 days is stored in a high-performance database (such as Redis) for real-time querying, while historical data is archived to a distributed file system (such as HDFS) to ensure data read / write efficiency and storage security. Data transmission uses the TLS 1.3 encryption protocol, and storage uses the AES-256 encryption algorithm, with keys rotated quarterly. Access control adopts the RBAC (Role-Based Access Control) model, differentiating between administrators, maintenance personnel, and ordinary users, refining data operation permission boundaries, and requiring secondary authentication and operation logging for critical operations.

[0076] Furthermore, based on standardized multidimensional data tables, deep learning is used to extract core parameter features of the device, match and generate a parameter acquisition priority list, and initially determine the acquisition strategy. The specific process is as follows:

[0077] Feature extraction and analysis of molding machine operation data in a standardized multidimensional data table are performed using a deep learning model to construct a parameter importance evaluation model. The formula is as follows:

[0078] ;

[0079] in For the first The importance score of each parameter For the first Parameters and Molding Unit Production Business Scenarios The correlation coefficient (with values ​​ranging from [0,1]). For the first The percentage of data update frequency for each parameter under normal production conditions (value range [0,1]). A CNN-LSTM hybrid model is adopted, with CNN layers used to extract local features of the parameters and LSTM layers used to capture time series dependencies. The model training adopts cross-validation (5-fold cross-validation) to ensure the model's generalization ability; the training data must include equipment operation data from at least 3 different manufacturers and 5 different operating conditions, with a sample size of no less than 500,000.

[0080] The results are obtained through business scenario correlation analysis and feature correlation calculation, and the specific steps are as follows:

[0081] Define the production business scenarios of molding units The core management objectives (such as product qualification rate, production efficiency, and equipment operation stability) are defined with corresponding sets of key business indicators. (such as pass rate, output per unit time, and frequency of equipment failure).

[0082] Parameter data and corresponding business indicator data for the past 30 days were extracted from a standardized multidimensional data table to construct a dataset that associates parameters with business indicators. This ensures that each parameter data corresponds to a unique measured value of a business indicator. Data from abnormal periods of business indicators (such as equipment failure and maintenance periods or raw material replacement periods) were removed. A small number of missing data were supplemented using linear interpolation to ensure the completeness of the dataset. The training set and the test set were divided in a 7:3 ratio for the validation of the correlation model.

[0083] The Pearson correlation coefficient formula was used to calculate the first... Parameters and various business indicators ( correlation coefficient The formula is:

[0084] ;

[0085] in For the sample size, For the first The parameter of the first Each sample value For the first The sample mean of each parameter, For the first The first business indicator Each sample value For the first The sample mean of each business indicator. Multicollinearity among parameters is detected using the variance inflation factor (VIF). Redundant parameters were removed at 10:00 AM. The Spearman rank correlation coefficient was used to supplement the nonlinear relationship analysis. An autoregressive integral moving average (ARIMA) model was employed to handle the autocorrelation of time series data, improving the accuracy of correlation assessment. The correlation coefficient threshold was determined by... Statistical validation of the test (significance level) =0.05), to ensure the objectivity of the correlation determination.

[0086] Take the first The maximum value of the correlation coefficients between each parameter and all business indicators is used as... ,Right now This ensures that the coefficients reflect the strongest correlation between the parameters and the business scenario.

[0087] The parameters are obtained through statistical analysis and normalization of update frequency. The specific steps are as follows:

[0088] Extract parameter data under normal production conditions from a standardized multidimensional data table, remove outliers and missing values, and obtain the first... Valid time series data for each parameter Corresponding timestamp sequence .

[0089] Calculate the time interval between two adjacent valid data points ( The average number of updates per unit of time (1 hour) (Unit: times / hour); For abnormal update interval data caused by network fluctuations (such as intervals less than 10ms or greater than 10 minutes), a 3x median method is used for correction to ensure the accuracy of frequency statistics.

[0090] Calculate the average number of updates per unit time for all parameters to obtain the maximum value. and minimum value Using the linear normalization formula to Convert to a value within the range [0,1], that is: If all parameters are updated at the same frequency ( ),but .

[0091] Obtained through the above process and Then, the importance scores for all parameters are calculated. Therefore, the scoring threshold is determined. The specific determination process is as follows:

[0092] Parameter data under normal production conditions for nearly 30 days were extracted from a standardized multidimensional data table. Valid sample sets covering all equipment types and all time periods were selected to ensure that the samples are representative and that the sample size is no less than 1,000 groups.

[0093] Based on the above parameter importance assessment model, the importance of all parameters in the sample set is calculated. Values ​​are used to form an importance score set. ( (Total number of parameters).

[0094] right Statistical analysis was performed to calculate the frequency and cumulative distribution of the ratings. Unsupervised clustering of the ratings was then conducted using the K-means clustering algorithm (with a cluster size of 2), resulting in two cluster centers. and ( ).

[0095] Threshold determination: The midpoint between the two cluster centers is used as the initial threshold. Based on the production business scenario requirements of the molding unit, the rationality of the division between high-scoring parameters and low-scoring parameters under this threshold is verified. First, the parameters are... The parameters are tentatively set as the high-scoring parameter set. The parameters are tentatively set as the low-scoring parameter set; verify whether the high-scoring parameter set includes core production control indicators such as injection pressure, mold temperature, and cycle time, and whether the low-scoring parameter set does not contain any key auxiliary indicators necessary for production monitoring (such as key parameters affecting the safe operation of equipment); if this condition is met, then If it does not meet the requirements, then proceed as follows: The step size is fine-tuned to adjust the threshold, and the above provisional set and verification process is repeated until the partitioning results meet the above requirements, and finally the scoring threshold is determined. Establish a classification rule knowledge base, combining industry standards and an expert review mechanism (inviting at least 3 senior process engineers to participate in the review), to reduce the space for subjective judgment and ensure classification consistency; the rule knowledge base supports version management and change tracking.

[0096] Based on the final determined scoring threshold Clearly define the rules for classifying parameter types: when When, it is determined to be a core parameter (including injection pressure, mold temperature, cycle time, etc.); when When this occurs, it is determined to be a non-core parameter (including maximum injection speed, mold opening and closing time, network latency, etc.).

[0097] Next, based on the business scenario requirements of the molding unit, and following the rule of "prioritizing the collection of core parameters and then collecting non-core parameters," a parameter collection priority list for each device is generated. The priority mapping rule is implemented according to the formula:

[0098] ;

[0099] in For the first The collection priority of each parameter, the core parameter set is as follows The parameter set is divided into two parts: the non-core critical parameter set, which includes parameters directly related to equipment operating efficiency (such as maximum injection speed and mold opening / closing time), and the non-core general parameter set, which includes other non-core parameters (such as network latency). Each quarter, based on changes in equipment operating data and business indicators, parameter priorities are reassessed, and the core parameter set is adjusted by adding or deleting parameters. The adjustments must be approved by the process department before taking effect.

[0100] Finally, based on the parameter acquisition priority list, the acquisition frequency adjustment logic is configured, and a mapping relationship between priority and acquisition frequency adjustment coefficient is established. The formula is as follows:

[0101] ;

[0102] in A sampling frequency adjustment coefficient is set for the corresponding priority level. This coefficient enables dynamic configuration of the sampling frequency (the configuration entry is implicitly provided in the front-end "Machine Configuration" module). The adjusted sampling frequency = base sampling frequency × adjustment coefficient. A distributed lock (RedisRedlock) is used to control concurrent sampling tasks across multiple devices. Processing threads are allocated by hash partitioning based on device ID to avoid resource contention. A task queue buffer mechanism is set up to control the number of concurrent requests to not exceed the system threshold (default 1000 devices / second). An overload protection mechanism is established; when the system load exceeds 85%, the sampling frequency of non-core parameters is automatically reduced, prioritizing the sampling of core parameters.

[0103] The final output is a device acquisition priority configuration table. The data table fields are arranged in a fixed order as follows: device number, IP address, standardized parameter name, parameter unit, parameter type (core / non-core), and correlation coefficient. Update frequency percentage The system includes importance scoring, collection priority (high / medium / low), and collection frequency adjustment coefficients. Before output, completeness checks (ensuring no fields are missing) and reasonableness checks (ensuring the collection frequency is within the device's supported range) are required. Configuration tables that fail checks must be regenerated. Configuration tables are stored in JSON format and support both incremental updates and full replacements.

[0104] Furthermore, by combining the data acquisition priority configuration table with historical time-series data, the parameters under different operating conditions are analyzed to identify patterns, dynamically adjust and validate the acquisition strategy, and output a dynamic acquisition strategy table. The specific process is as follows:

[0105] We selected effective, multi-dimensional, real-time operational data from the past 30 days under normal production conditions from historical time-series data, removed outliers and missing data, and categorized and organized the data by equipment number and parameter type to construct a five-dimensional time-series dataset encompassing equipment, parameters, time, values, and operating conditions. This ensured that each data point contained a clear operating condition status identifier (production / standby) and timestamp information. For operating conditions with limited data (such as maintenance conditions), we employed Synthetic Minority Oversampling Technique (SMOTE) to generate virtual samples, ensuring a balanced data volume across all operating conditions. The time-series data was segmented, with each segment containing one hour of continuous data to facilitate pattern detection.

[0106] An LSTM (Long Short-Term Memory) model is used to learn features from the preprocessed time-series dataset, uncovering the variation patterns of parameters under different operating conditions. The training data requirements are a time span ≥ 90 days, a sample size ≥ 100,000, covering all operating scenarios. An early stopping mechanism is employed to prevent overfitting. The model hyperparameters are optimized using grid search (2-4 hidden layers, learning rate 0.001-0.01, batch size 32-128). The model input consists of continuous time-series parameter data and corresponding operating condition labels, and the output is a feature vector of parameter variation trends. Specifically, the frequency and amplitude of parameter changes under different operating conditions are calculated using the following formula:

[0107] Frequency of change calculation: ,in, For the first These parameters are under operating conditions. ( A value of 1 indicates the production condition. Taking 2 to represent the frequency of change per unit time under standby conditions, For working conditions The effective number of changes of the lower parameter For working conditions Duration (in hours);

[0108] Fluctuation range calculation: ,in, For the first These parameters are under operating conditions. The fluctuation range below For working conditions The maximum value of the following parameter, For working conditions The minimum value of the parameter. For periodically fluctuating parameters (such as injection pressure), the fluctuation coefficient (fluctuation amplitude / mean) is used instead of the absolute fluctuation amplitude to more accurately reflect the stability of the parameter.

[0109] Dynamic adjustment of data acquisition strategy: Based on the parameter change patterns and the data acquisition priority configuration table, establish a mapping relationship between operating conditions, priorities, and acquisition frequency, and dynamically adjust the acquisition frequency of each parameter according to the following formula: ,in, For the first These parameters are under operating conditions. The adjusted sampling frequency This is the baseline sampling frequency (default value is 1 time / second). For the first The sampling frequency adjustment coefficients corresponding to each parameter For working conditions The frequency correction factor below.

[0110] A priority-based decision-making mechanism is used to resolve conflicts arising from multiple conditions (first categorizing by change frequency, then further subdividing by fluctuation amplitude). The specific rules are as follows: When If times / hour ( For parameters (historical average) ,like but ,like but ;when times / hour If times / hour but ,like but ,like but ;when ,like but ,like but ,like but .

[0111] The adjusted data collection strategy was applied to pilot equipment (three molding machines from different manufacturers) to collect data in real time for 24 hours. The validity and timeliness of the collected data were verified using the following formula:

[0112] Data validity: ,in, For the first The efficiency of data collection for each parameter. To effectively collect data, For the total number of data collections, ultimately ensuring ;

[0113] Data timeliness: ,in, For the first Average transmission delay of each parameter For the system to receive the first The timestamp of each data item Generate the first for the device The timestamp of each piece of data ultimately ensures If the verification result does not meet the above requirements, proceed as follows: The frequency correction coefficient of the corresponding parameter for step size fine-tuning Repeat the pilot data collection and verification process until all parameters meet the requirements of validity and timeliness.

[0114] Through the above processing, the final output is a dynamic data acquisition strategy table. The table fields are arranged in a fixed order as follows: device number, IP address, standardized parameter name, parameter unit, parameter type, acquisition priority, baseline acquisition frequency, frequency adjustment coefficient, operating condition type (production / standby), frequency correction coefficient, adjusted acquisition frequency, parameter change frequency, parameter fluctuation amplitude, and verification result (pass / fail). Semantic version numbers (e.g., V1.0.0) are used to manage strategy versions, recording version iteration records (creation time, modified content, responsible person), and supporting version backtracking queries. The currently effective strategy is automatically backed up before iteration. If verification fails after iteration (three consecutive acquisitions failing to meet the indicators), an automatic rollback to the backup version is triggered, and the reason for the failure is recorded in the iteration log, notifying maintenance personnel to intervene. The rollback process must be completed within 10 seconds to ensure uninterrupted data acquisition.

[0115] Furthermore, data collection instructions are issued according to the dynamic data collection strategy table, and real-time data is displayed through multiple front-end interfaces. Strategy iterations are triggered based on operating condition changes or anomaly prompts. The specific process is as follows:

[0116] First, based on the dynamic data acquisition strategy table, a one-to-one command mapping relationship is established between device number and IP address. For each device, a structured acquisition command is generated. The command content clearly includes a list of parameters to be acquired, the corresponding adjusted acquisition frequency, data upload format specifications (such as the number of bits to retain in the numerical value, timestamp format), and transmission protocol requirements (TCP / IP). To ensure the uniqueness of commands and full-process traceability, each command is assigned a unique identifier code. The encoding rule is a hash value combining the device number, parameter name, operating condition type, and command issuance timestamp. A fixed-length unique identifier is generated through a hash algorithm to avoid command confusion or duplicate issuance. The architecture employs a front-end / back-end separation approach. The front-end (Vue3 framework) interacts with the back-end service (SpringBoot microservice) via RESTful API. The back-end adopts a layered design, including an interface layer (receiving front-end requests), a business logic layer (handling collection strategies and command generation), a data access layer (reading and writing to the database), and a device communication layer (interacting with IoT devices). Real-time data push is achieved through WebSocket to ensure that the front-end interface data is updated in real time. Nginx is used as a reverse proxy to achieve load balancing and static resource caching. A message queue (RabbitMQ) is introduced to handle command issuance and data upload, avoiding system congestion.

[0117] The system sends collection commands to the corresponding device's control module based on IP address. After sending, it monitors the command status in real time, distinguishing between three states: pending response, responded, and failed. For commands that fail to send, it automatically retryes at 10-second intervals. If the retry fails after 3 attempts, it triggers a device offline alarm, which is simultaneously pushed to the front-end data monitoring module and recorded in the system log, facilitating timely troubleshooting of network or device faults by maintenance personnel. When a device is offline, it starts a local cache (capacity ≥ 100,000 records, storage duration ≥ 72 hours) to cache collected data and timestamps. After the device comes back online, it prioritizes uploading cached data (incrementally uploading in timestamp order), and resumes normal collection after the upload is complete. If the offline time exceeds 24 hours, it triggers a chunked upload mechanism (10,000 data records per chunk) to avoid data transmission congestion. Supplemented data must be re-processed and verified to ensure data integrity. Key parameters (core parameters) during offline periods use a local anomaly detection mechanism, triggering a local device alarm when they exceed safe limits.

[0118] After receiving the acquisition command, the device control module strictly follows the adjusted acquisition frequency specified in the command to collect parameter data. During the acquisition process, standardized processing rules are automatically adopted to remove outliers and unify the numerical format, ensuring the standardization of uploaded data. After data acquisition is completed, the data is uploaded to the system along with the corresponding command's unique identifier and data generation timestamp. After receiving the data, the system performs integrity verification. Once it confirms that the data fields are complete and the format meets the requirements, it synchronously pushes the data to multiple front-end interfaces for linked display, ensuring the consistency and real-time performance of data across all interfaces. Interface data synchronization rules are established to ensure that the display delay of the same parameter on different interfaces is ≤300ms; cross-interface parameter filtering and positioning are supported, and clicking on a parameter on a certain interface will automatically highlight the corresponding data on other related interfaces; the front-end interface adopts a responsive design, adapting to desktop, tablet, and mobile devices to meet the monitoring needs of different scenarios.

[0119] The system continuously monitors equipment operating status and parameter data, and initiates iterative optimization of strategies through multi-dimensional trigger conditions to ensure that the data acquisition strategy always adapts to the equipment's operating needs. It monitors the operating condition identifiers uploaded by the equipment in real time. When it detects a switch in equipment operating condition from the current state to another state (such as from production to standby, or vice versa), it automatically extracts the corresponding data acquisition parameter configuration for the new operating condition from the dynamic data acquisition strategy table. This includes the adjusted acquisition frequency and frequency correction coefficient for each parameter under this operating condition. It quickly generates new acquisition instructions and sends them to the equipment, overriding the acquisition strategy corresponding to the original operating condition. Simultaneously, it records the operating condition switch time and strategy update content in the iteration log, achieving seamless integration between operating condition switching and strategy adjustment. The system operates according to a set 3... The anomaly detection rules continuously verify uploaded parameter values. When a parameter value is detected to be outside the normal range (i.e., determined to be abnormal data), an anomaly alert is immediately triggered, notifying operations and maintenance personnel in the front-end data monitoring module. Simultaneously, a temporary policy adjustment is automatically initiated, temporarily raising the collection priority of the abnormal parameter by one level (core high priority remains unchanged, non-core medium priority is raised to high priority, and non-core low priority is raised to medium priority), correspondingly increasing the collection frequency to 1.5 times the original frequency, and generating a temporary collection command. After continuously collecting data according to the temporary policy for 10 minutes, if the parameter anomaly is resolved, the original collection policy is automatically restored; if the anomaly persists, the temporary collection policy is retained and... Manual intervention notifications are pushed out to maintenance personnel to troubleshoot equipment faults. Every day from 00:00 to 02:00 (during low equipment load operation), the system automatically initiates a periodic strategy iteration process. This process extracts all data collected from the equipment throughout the day, recalculates the frequency and amplitude of changes in each parameter under different operating conditions, updates the corresponding frequency correction coefficients using the strategy adjustment rules, and generates an updated strategy table. The new strategy table undergoes validity verification to ensure that parameter collection effectiveness is no less than 95% and the average transmission latency is less than 100ms. After successful verification, the new strategy is automatically deployed to the corresponding equipment at 02:30 the following day, completing the periodic optimization of the strategy and ensuring its adaptability to the long-term operating patterns of the equipment. Iteration effect evaluation indicators are established, including collection efficiency improvement rate, data effectiveness improvement rate, and equipment fault warning accuracy rate. Monthly statistics and evaluations of iteration effects are compiled, generating an optimization report. For strategies with poor iteration effects, a manual optimization process is initiated, where the technical team analyzes the causes and adjusts the strategy parameters.

[0120] All strategy iteration operations are recorded in detail in the system iteration log in a unified format. The log content includes the iteration timestamp, the device number involved, the parameter name, the operating status before and after the iteration, the strategy parameters before and after the adjustment (such as the frequency correction coefficient and the collection frequency), and the iteration execution result (success / failure), forming a complete strategy iteration traceability chain. The system log supports searching and querying by time range, device number, iteration type, and other conditions, which facilitates the review of the strategy adjustment effect by operation and maintenance personnel and provides data support for subsequent collection strategy optimization. The log uses distributed storage (Elasticsearch) and supports multi-dimensional searching and querying by time range, device number, iteration type, version number, and other conditions. The log retention period is 1 year (logs exceeding the retention period are automatically archived to cold storage, with an archiving period of ≥3 years); a log export function is provided (supporting Excel and PDF formats) for easy auditing and review; a log security mechanism is established, requiring permission verification for log access, and critical operation logs (such as strategy rollback and permission changes) are encrypted and stored.

[0121] Through the above-described end-to-end processing, a closed-loop management system is achieved, from the precise issuance of collection commands and the real-time multi-terminal linkage display of data to the dynamic iteration and optimization of strategies. This effectively ensures the real-time, effectiveness, and adaptability of equipment parameter collection. The output strategy iteration logs and updated dynamic collection strategy tables can serve as core foundational data for equipment operation and maintenance optimization and subsequent system upgrades.

[0122] The processes described above with reference to the flowcharts in the embodiments disclosed in this invention can be implemented as computer software programs. The embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs the functions defined in the methods of this application. It should be noted that the computer-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wire segments, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless segments, wire segments, optical fibers, RF, etc., or any suitable combination thereof.

[0123] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0124] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The purpose of the present invention has been fully and effectively achieved. The functions and structural principles of the present invention have been shown and explained in the embodiments. Without departing from the stated principles, the implementation of the present invention may have any variations or modifications.

Claims

1. A method for optimizing data acquisition in IoT devices based on deep learning, characterized in that, The method includes: Collect multi-dimensional real-time operating data from IoT devices, preprocess the real-time operating data based on device identification, and obtain standard operating data; Parameter importance analysis is performed on standard operating data based on deep learning algorithms to obtain parameter collection priorities; Based on the changing patterns of operating conditions, the priority of parameter acquisition is dynamically adjusted and verified to obtain dynamic acquisition strategy data. Extract the set of operating condition switching and parameter anomaly events from the standard operating data, and perform strategy iteration optimization and effect verification on the set of parameter anomaly events based on the dynamically collected strategy data to obtain strategy iteration data; Based on the system interaction logic, a visualization interface is generated corresponding to the dynamically collected strategy data, strategy iteration data, and real-time running data, and the visualization interface is displayed in a linked manner. The specific process for obtaining the data using the dynamic acquisition strategy is as follows: Extract the operating condition switching events from the dataset of each parameter to obtain the operating condition event set; The set of working condition events is sorted according to time order to obtain the set of working condition event sequences, and the frequency and fluctuation amplitude of each parameter under different working conditions are calculated. Based on the change frequency and fluctuation amplitude, the frequency correction coefficient corresponding to the working condition is determined. Combined with the frequency adjustment coefficient corresponding to the parameter acquisition priority, the adjusted acquisition frequency of each parameter under different working conditions is calculated. Several devices from different manufacturers were selected as pilot devices, and real-time data was collected at the adjusted collection frequency. The efficiency of the collected data and the transmission delay were calculated. The acquisition strategy is validated based on preset efficiency thresholds and transmission delay thresholds. If the threshold requirements are not met, the frequency correction coefficient is fine-tuned and the trial is repeated until the requirements are met, thus obtaining dynamic acquisition strategy data.

2. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 1, characterized in that, The specific process for obtaining the multi-dimensional real-time runtime data is as follows: The operating status and core parameters of IoT devices are collected from multiple dimensions to obtain raw data. Based on the device business logic, the raw collected data is associated with device identifiers and encapsulated to obtain encapsulated operational data. The standardized interface based on the Internet of Things system will encapsulate the operational data and upload it to the pre-set central data platform; The system receives real-time operation messages from the central data platform via message subscription and parses these messages to obtain parsed operation data. The parsed running data is correlated with the device running context to obtain multi-dimensional real-time running data.

3. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 2, characterized in that, The specific process for obtaining the standard operating data is as follows: The system categorizes and classifies equipment and manufacturers based on multi-dimensional real-time operational data, and adds equipment tags to the multi-dimensional real-time operational data based on the classification results. The integrity of real-time operation data is verified based on device tags, and the data is deduplicated based on device identifiers and data time difference thresholds after integrity verification to obtain deduplicated operation data. Linear interpolation is used to fill in missing values ​​of continuous parameters in the deduplication running data, and mode completion is used to fill in missing values ​​of discrete parameters, thus obtaining the completed running data; Anomaly detection is performed on the completed operation data based on the 3σ principle to obtain abnormal data. The abnormal data is then replaced with the mean of the normal data in the same time period to obtain the noise-reduced operation data. All numerical parameters in the noise reduction operation data are rounded to two decimal places. Based on the preset field template, the noise reduction operation data is standardized to obtain standard operation data.

4. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 3, characterized in that, The specific process for obtaining the parameter acquisition priority is as follows: Using the device identifier as the primary key, extract the parameter operation dataset corresponding to each device from the standard operation data, and extract the parameter time series events from each parameter operation dataset to obtain the parameter event set; The parameter event set is sorted according to time order to obtain the parameter event sequence set, and the correlation coefficient between each parameter and business indicator is calculated based on the core objectives of the business scenario. The update frequency of each parameter per unit time under normal production conditions is statistically analyzed, and the percentage of parameter update frequency is obtained through linear normalization. The parameter importance score is calculated by weighting the correlation coefficient and the parameter update frequency ratio. The K-means clustering algorithm is used to classify the parameter importance score to obtain the classification results of core parameters and non-core parameters. Assign collection priorities to different types of parameters based on business scenario requirements, and generate parameter collection priority data.

5. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 4, characterized in that, The specific process for obtaining the strategy iteration data is as follows: Based on the dynamic acquisition strategy data and parameter anomaly event set, multi-dimensional time series features are extracted from standard operating data to obtain parameter time series features; Based on the parameter time series characteristics and parameter anomaly event set, graph structure modeling and anomaly context modeling are performed on standard operating data to obtain parameter association graph structure; The parameter anomaly event set is used as anomaly label, and the anomaly label is used as a supervision signal to train the LSTM neural network model to obtain the parameter anomaly analysis model. Based on the parameter anomaly analysis model, the parameter correlation graph structure is forward-propagated to obtain the anomaly contribution weights. The parameter acquisition strategy is iteratively adjusted based on the abnormal contribution weight, and the effectiveness of the adjusted strategy is verified to obtain strategy iteration data.

6. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 5, characterized in that, The strategy iteration and optimization process includes strategy version management and a failure rollback mechanism, the specific acquisition process of which is shown below: Semantic version numbers are used to identify each version of the data collection strategy, and the creation time, modification content, responsible person and scope of effective devices of each version strategy are recorded to form a complete version iteration record; Before strategy iteration, the currently effective collection strategy is automatically backed up as a backup version for rollback. If the iterated strategy still fails to meet the preset indicators after several consecutive data collection and verifications, the automatic rollback mechanism will be triggered to immediately switch to the backup original strategy version to ensure that the data collection business is not interrupted. After the rollback operation is completed, the reason for the iteration failure, the rollback time, and the information of the affected devices will be recorded in the system iteration log, and an alarm notification will be pushed to the operation and maintenance personnel for manual intervention and investigation.

7. The method for optimizing data acquisition of IoT devices based on deep learning according to claim 6, characterized in that, The strategy iterative optimization process also includes offline data acquisition and security protection mechanisms, the specific acquisition process of which is shown below: When the device is offline, the local caching function is automatically activated to collect and store core parameter data and corresponding timestamps at a preset frequency. After the device is back online, cached data will be uploaded incrementally in timestamp order first. Once the upload is complete, the normal data collection rhythm will resume. If the offline time exceeds the threshold, a segmented upload mechanism is used to avoid data transmission congestion. The supplementary data needs to be reprocessed and validated. The data transmission process uses the TLS 1.3 encryption protocol, and the data storage uses the AES-256 encryption algorithm. The system adopts the RBAC access control model to distinguish the operation permissions of administrators, maintenance personnel and ordinary users. Key policy adjustments and data query operations require secondary authentication and operation log recording.

8. A data acquisition and optimization system for IoT devices based on deep learning, characterized in that, The system is used to execute the deep learning-based IoT device data acquisition optimization method according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is executed by a processor to implement the deep learning-based data acquisition optimization method for Internet of Things devices as described in any one of claims 1-7.

Images