A time series data playback method and device for an Internet of Things device

Abstract

The present application discloses a time series data playback method and device for an Internet of Things device, which relates to the field of data playback technology and is used to solve the problem that it is difficult to play back key features within a limited time in the existing method. The method includes: filtering the historical time series data set according to the preset data set conditions, and determining the device attribute joint list based on the index data corresponding to the historical time series data set; filtering the device attribute joint list based on the project requirement information to obtain the historical time series data set to be analyzed; obtaining the key points corresponding to each historical time series data set to be analyzed according to the identification strategy corresponding to each key characteristic; traversing the historical time series data set to be analyzed according to the preset sliding time window to determine the continuous time area; sorting based on the number of preset sliding time windows corresponding to each continuous time area to determine the display area, and generating the visualization content of the historical time series data set to be analyzed in the display area for playback.

Description

Technical Field

[0001] The present application relates to the field of data playback technology, and in particular to a time series data playback method and device for an Internet of Things device. Background Art

[0002] With the rapid development of Internet of Things technology, interconnection between various types of equipment has become a reality, which not only greatly improves production efficiency, but also promotes the rapid development of industrial automation and intelligent manufacturing. In this context, the generation of massive time series data has become a phenomenon that cannot be ignored. These data, as a direct reflection of the operating status of the equipment, contain information that is crucial to equipment status monitoring, fault analysis and historical trend analysis, and are particularly valuable in the field of industrial automation and intelligent manufacturing. Re-presenting historical time series data sets, that is, replaying time series data, can not only accurately reproduce the specific operations and states of the equipment in history, but also has a significant effect on analyzing historical problems and identifying the causes of faults. It provides a unique perspective for in-depth understanding of equipment behavior and optimization of business processes, and is of great significance to improving the ability of equipment status monitoring, fault analysis and historical trend analysis.

[0003] Most traditional time series data playback methods use linear playback of data points one by one. This method can still cope with the situation of small data volume, but when facing large data volume, a large amount of time series data brings information overload challenges to users. Based on linear playback of data points one by one, it is difficult to quickly identify and select valuable information in massive data. The efficiency and user experience of this traditional method are difficult to meet the needs of modern industry. In terms of in-depth analysis, existing products are usually limited to numerical playback of data points and lack the ability to conduct in-depth analysis of time series data. This leads to efficiency problems in key feature extraction and information screening. Existing methods are difficult to effectively retain useful information of data with different time granularities when playing back data points one by one, thus affecting the accuracy and efficiency of key feature extraction. In addition, the existing method of selecting feature data after feature extraction and replaying it is also difficult to replay within a limited time, which makes it difficult for users to efficiently browse historical time series data sets within a limited time. Summary of the invention

[0004] In order to solve the above technical problems, one or more embodiments of the present application provide a time series data playback method and device for an Internet of Things device.

[0005] One or more embodiments of the present application adopt the following technical solutions:

[0006] One or more embodiments of the present application provide a time series data playback method for an IoT device, the method comprising:

[0007] The data playback system screens the historical time series data set according to the preset data set conditions, and determines the device attribute joint list based on the index data corresponding to the historical time series data set;

[0008] Iteratively screening the device attribute joint list based on project requirement information to obtain a historical time series data set to be analyzed; wherein the project requirement information includes: analysis object information and time range information;

[0009] According to the identification strategy corresponding to each key characteristic, the key features of the historical time series data set to be analyzed are detected to obtain the key points corresponding to each of the historical time series data sets to be analyzed;

[0010] Traversing the historical time series data set to be analyzed according to a preset sliding time window to determine a continuous time region according to the number of key points in each sliding time window and the average number of key points in each sliding time window;

[0011] The continuous time regions are sorted based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine the longest continuous time region as the display region, so as to generate visualization content of the historical time series data set to be analyzed in the display region for playback.

[0012] Optionally, in one or more embodiments of the present application, the data playback system screens the historical time series data set and determines a joint list of device attributes corresponding to the historical time series data set according to a preset data set condition and a preset index format, specifically including:

[0013] Detect whether there are missing values ​​in each data set in the database, so as to determine the integrity test results of each data set according to the missing value test results;

[0014] If it is determined that the integrity monitoring result meets the preset integrity requirement, a trend analysis is performed on each data set based on the time sequence to determine the volatility detection result of each data set;

[0015] If it is determined that the volatility detection result meets the preset volatility requirement, obtaining the initial historical time series data corresponding to the data set;

[0016] Predicting the initial historical time series data according to the long short-term memory network to obtain a prediction result, and adding the prediction result to the initial historical time series data to obtain a historical time series data set;

[0017] The original index data of the historical time series data set is obtained to determine the Internet of Things devices and Internet of Things device attributes corresponding to the historical time series data set based on the index data, and a corresponding device attribute joint list is generated based on the Internet of Things devices and the Internet of Things device attributes; wherein the Internet of Things device corresponds to multiple Internet of Things device attributes.

[0018] Optionally, in one or more embodiments of the present application, the initial historical time series data is predicted according to the long short-term memory network to obtain a prediction result, and the prediction result is added after the initial historical time series data and before the historical time series data set is obtained. The method further includes:

[0019] Obtain a historical time series data set within a preset period and divide it into a training set, a validation set, and a test set;

[0020] Standardizing the historical time series data set in the training set to obtain a standardized historical time series data set, and dividing the standardized historical time series data set into an input sequence and an output sequence;

[0021] Constructing an initial long short-term memory network based on a deep learning framework, training the initial long short-term memory network according to the input sequence and the output sequence, and performing performance monitoring based on the validation set to obtain a trained long short-term memory network;

[0022] The trained long short-term memory network is evaluated based on the test set to obtain a long short-term memory network, so as to predict the initial historical time series data according to the long short-term memory network.

[0023] Optionally, in one or more embodiments of the present application, the device attribute joint list is iteratively screened based on the project requirement information to obtain the historical time series data set to be analyzed, specifically including:

[0024] Filter the device attribute joint list according to the analysis object information to obtain a data playback analysis object list;

[0025] Perform secondary screening on the historical time series data set corresponding to the data playback object list according to the time range information to obtain an initial historical time series data set to be analyzed;

[0026] Constructing index retrieval information corresponding to the initial historical time series data set to be analyzed based on the index data corresponding to the initial historical time series data set to be analyzed and the time range information;

[0027] According to the index retrieval information, the playback video and analysis results of the preset data storage module are retrieved to determine whether the initial historical time series data set to be analyzed exists;

[0028] If yes, the existing initial historical time series data set to be analyzed is deleted to obtain the historical time series data set to be analyzed;

[0029] If not, the initial historical time series data set to be analyzed is used as the historical time series data set to be analyzed.

[0030] Optionally, in one or more embodiments of the present application, according to the identification strategy corresponding to each key characteristic, the key features of the historical time series data set to be analyzed are detected to obtain the key points corresponding to each of the historical time series data sets to be analyzed, specifically including:

[0031] When the key feature to be analyzed of the historical time series data set to be analyzed is a stability feature, the coefficient of variation corresponding to the historical time series data set to be analyzed is obtained, so as to determine the first key point corresponding to the stability feature of the historical time series data set to be analyzed according to the coefficient of variation and a preset scoring standard;

[0032] When the key feature to be analyzed of the historical time series data set to be analyzed is an abnormal condition feature, a second key point corresponding to the abnormal condition feature of the historical time series data set to be analyzed is determined according to a preset threshold and an alarm mechanism corresponding to the historical time series data set to be analyzed; wherein the preset threshold and the alarm mechanism are set based on a data playback object corresponding to the historical time series data set to be analyzed;

[0033] When the key feature to be analyzed of the historical time series data set to be analyzed is a statistical feature, statistical calculation is performed on the historical time series data set to be analyzed to obtain a third key point corresponding to the statistical feature of the historical time series data set to be analyzed;

[0034] The first key point, the second key point and the third key point are summarized to obtain key points corresponding to each of the historical time series data sets to be analyzed.

[0035] Optionally, in one or more embodiments of the present application, the historical time series data set to be analyzed is traversed according to a preset sliding time window to determine a continuous time region according to the number of key points in each sliding time window and the average number of key points in each sliding time window, specifically including:

[0036] Determine a complete analysis time range corresponding to the historical time series data set to be analyzed, so as to traverse the historical time series data set to be analyzed within the complete analysis time range according to a preset sliding time window;

[0037] The number of key points corresponding to each key point type in each sliding time window during the traversal process is accumulated to obtain the number of key points in each sliding time window;

[0038] Determine the total number of the preset sliding time windows according to the complete analysis time range and the time length of the preset sliding time window, and determine the average number of key points in each of the preset sliding time windows according to the number of key points in each sliding time window and the total number;

[0039] According to the relationship between the number of current preset sliding time windows and the total number of preset sliding time windows, and the relationship between the number of key points in the preset sliding time windows and the average number, the continuous time region corresponding to the historical time series data set to be analyzed is determined.

[0040] Optionally, in one or more embodiments of the present application, according to the relationship between the number of current preset sliding time windows and the total number of preset sliding time windows, and the relationship between the number of key points in the preset sliding time window and the average number, determining the continuous time region corresponding to the historical time series data set to be analyzed specifically includes:

[0041] If it is determined that the number of the current preset sliding time windows is less than the total number of the preset sliding time windows, and the number of key points in the preset sliding time windows is greater than the average number, then the position of the preset sliding time window is recorded as the position of the starting sliding window, and the starting time point of the starting sliding window is determined;

[0042] Recording the starting time point of the starting sliding window, so as to search for the ending time point based on the starting time point;

[0043] If it is determined that the number of the current preset sliding time windows is less than the total number of the preset sliding time windows, and the number of key points in the preset sliding time windows is less than or equal to the average number, then the position of the preset sliding time window is recorded as the position of the end sliding window, and the starting time point of the end sliding window is determined to be the end time point corresponding to the starting time point of the start sliding window;

[0044] If it is determined that the number of the current preset sliding time windows is greater than the total number of the preset sliding time windows, recording the position of the last preset sliding time window as the position of the end sliding window, and determining the starting time point of the end sliding window as the end time point corresponding to the starting time point of the start sliding window;

[0045] According to the starting time point of the starting sliding window and the ending time point corresponding to the starting time point of the starting sliding window, the continuous time region corresponding to the historical time series data set to be analyzed is determined.

[0046] Optionally, in one or more embodiments of the present application, sorting is performed based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine the longest continuous time region as the display region, specifically including:

[0047] Sorting is performed based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine a preset number of continuous time regions with the longest length, and using the preset number of continuous time regions as display regions;

[0048] If the number of the continuous time areas is less than the preset number of segments, all the continuous time areas are used as display areas.

[0049] Optionally, in one or more embodiments of the present application, generating visualization content of the historical time series data set to be analyzed in the display area for playback specifically includes:

[0050] Inputting the historical time series data set to be analyzed in the display area into a preset data playback engine, so as to convert the historical time series data set to be analyzed in the display area into visualization content according to the preset data playback engine; wherein the visualization content includes: static charts and dynamic videos;

[0051] The static charts and dynamic videos of the historical time series data set to be analyzed are played back based on the preset playback script.

[0052] One or more embodiments of the present application provide a time series data playback device for an Internet of Things device, the device comprising:

[0053] at least one processor; and,

[0054] a memory communicatively connected to the at least one processor; wherein,

[0055] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned methods.

[0056] At least one of the above technical solutions adopted in the embodiments of the present application can achieve the following beneficial effects:

[0057] The historical time series data set is filtered according to the preset data set conditions, and the device data joint list is determined according to the index data corresponding to the historical time series data set, so that the historical time series data set is converted into high-quality data that meets the data analysis and playback standards. Based on the analysis object information and time range information, the historical time series data set to be analyzed is obtained, which can preliminarily locate the historical time series data set of interest to the current project needs, avoiding the problem of low data playback efficiency caused by massive data analysis. According to the identification strategy corresponding to each key feature, the key features of the historical time series data set to be analyzed are detected, so as to obtain the key points corresponding to each historical time series data set to be analyzed, which can comprehensively and flexibly capture the key points in the historical time series data set. By determining the continuous time area by presetting the sliding time window, a limited number of key time areas can be further refined for key display, avoiding the problem of information overload, enabling users to grasp the key points faster and improve the efficiency of information acquisition. The playback process can be converted into intuitive visual content through the data playback engine, and data playback can be guided by the preset playback script, which improves the practicality and adaptability of data playback. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. In the drawings:

[0059] Figure 1 A schematic diagram of a method for replaying time series data of an Internet of Things device provided in an embodiment of the present application;

[0060] Figure 2 A schematic diagram of the overall architecture of time series data playback of an Internet of Things device provided in an embodiment of the present application;

[0061] Figure 3 A schematic diagram of the structure of a time series data playback device for an Internet of Things device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0062] The embodiments of the present application provide a method and device for replaying time series data of an Internet of Things device.

[0063] In order to enable those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this application.

[0064] like Figure 1 As shown, the present application embodiment provides a flow chart of a method for replaying time series data of an IoT device. Figure 1 It can be seen that in one or more embodiments of the present application, a time series data playback method of an IoT device includes:

[0065] S101: The data playback system filters a historical time series data set according to a preset data set condition, and determines a joint list of device attributes based on index data corresponding to the historical time series data set.

[0066] The historical time series data sets in real scenarios have been collected by the IoT platform and stored in the corresponding database. Therefore, in order to be able to replay the required historical time series data sets, the data playback system in the embodiment of the present application needs to synchronize these data to the data playback system. In order to convert the historical time series data sets into high-quality data that meets the data analysis and playback standards, the embodiment of the present application will filter the historical time series data sets according to the preset data set conditions, and determine the device data joint list according to the index data corresponding to the historical time series data sets. For example: Assuming that all relevant data has been properly stored in the database, the following conditions must be met when synchronizing the data: First, if Figure 2 The data set of the IoT devices shown must be continuous, without any missing values ​​or breakpoints, to ensure data integrity; timestamps should be evenly distributed or follow a specific time series pattern to ensure data consistency; data sets should show a certain volatility or variability, that is, data points on the time axis should show dynamic changes rather than static or constant; input data should be stored one by one according to the index format composed of devices and device attributes for subsequent query and analysis processes. Among them, the devices in the index refer to various industrial machines and equipment connected to the IoT platform, such as CNC machine tools, robots, conveyors, pumps, compressors, electricity meters, water meters, gas meters, etc. The device attributes in the index refer to the time series data types collected by sensors, which reflect the operating status of the equipment, covering multiple dimensions such as temperature, humidity, pressure, flow, vibration, speed, acceleration, etc.

[0067] Specifically, in one or more embodiments of the present application, the data playback system screens the historical time series data set and determines the device attribute joint list corresponding to the historical time series data set according to the preset data set condition and the preset index format, which specifically includes the following process:

[0068] First, detect whether there are missing values ​​in each data set in the database, and then determine the integrity test results of each data set according to the missing value test results. If it fails, it means that the data quality is poor and cannot meet the requirements of analysis and playback. At this time, it is necessary to process the original data and re-test it, or the user can choose to skip a certain data that does not meet the requirements and only input certain data that meet the standards to continue the process. If it is determined that the integrity monitoring result meets the preset integrity requirements, then trend analysis is performed on each data set based on the time sequence to determine the volatility test results of each data. If the volatility test result meets the preset volatility requirements, then the initial historical time series data corresponding to the data set is obtained. In order to supplement the initial historical time series data, in the embodiment of the present application, the initial historical time series data is predicted according to the long short-term memory network to obtain the prediction results, and the prediction results are supplemented after the initial historical time series data to obtain the historical time series data set. Then the original index data of the historical time series data set is obtained, so as to determine the IoT device and IoT device attributes corresponding to the historical time series data set according to the index data, and generate a corresponding device attribute joint list based on the IoT device and the IoT device attributes. It should be noted that an IoT device corresponds to multiple IoT device attributes, and in this list, the user is allowed to select the device and its device attributes that he or she wishes to perform playback analysis on by checking the box.

[0069] In this process, the historical time series data set is screened and determined by presetting the data set conditions, so that abnormal data or unstable data and other data with data quality problems can be identified when the data is synchronized to the data playback system, so as to improve the quality of subsequent playback. The initial historical time series data is predicted using the long short-term memory network, and the prediction results are added to the data set. This can not only enrich the content of the data set, but also make up for the shortcomings of historical data to a certain extent, and improve the integrity and continuity of the data. In addition, by generating a joint list of device attributes, users are allowed to select the devices and their device attributes that they want to replay and analyze according to their needs. This flexible data selection mechanism greatly improves the pertinence and practicality of data playback, and users can quickly locate the data points of interest for analysis.

[0070] Further, in one or more embodiments of the present application, the initial historical time series data is predicted according to the long short-term memory network to obtain a prediction result, and the prediction result is added after the initial historical time series data and before the historical time series data set is obtained. The method also includes the following process:

[0071] First, a historical time series data set within a preset period is obtained and divided into a training set, a validation set, and a test set. Then, the historical time series data set in the training set is standardized to obtain a standardized historical time series data set, and the standardized historical time series data set is divided into an input sequence and an output sequence. An initial long short-term memory network is constructed based on a deep learning framework, so that the initial long short-term memory network is trained according to the input sequence and the output sequence, and performance monitoring is performed based on the validation set to obtain a trained long short-term memory network. In order to make the prediction result more accurate, the trained long short-term memory network is evaluated according to the test set in an embodiment of the present application to obtain a long short-term memory network, so as to predict the initial historical time series data according to the long short-term memory network. In this process, the historical time series data set within the preset period is divided into a training set, a validation set, and a test set, which ensures the systematicness and scientificity of the model training. The training set is used to train the model, the validation set is used to monitor the model performance during the training process to prevent overfitting, and the test set is used to finally evaluate the accuracy and generalization ability of the model. This division method helps to obtain a more reliable and stable prediction model. Standardizing the historical time series data sets in the training set helps eliminate the dimensional differences between different data and improve the convergence speed and prediction accuracy of the model. Through rigorous model training, performance monitoring and evaluation processes, the long short-term memory network model obtained has higher prediction accuracy and stability. This means that when predicting the initial historical time series data based on this model, more accurate and reliable prediction results can be obtained, thereby further improving the integrity and continuity of the historical time series data set.

[0072] In a certain application scenario of this application, taking the carbon dioxide concentration monitoring in the intelligent environmental monitoring equipment as an example, the prediction process of the long short-term memory network is explained:

[0073] S1 Data collection and organization: Continuously monitor the indoor carbon dioxide concentration, record one data point every hour, collect historical time series data sets for the past year, and organize them into a format suitable for long short-term memory network input.

[0074] S2 Dataset division: Divide the data into training set, validation set and test set. For example, the data of the first 10 months can be used as the training set, the data of the next month can be used as the validation set, and the data of the last month can be used as the test set.

[0075] S3 data preprocessing: Standardize the carbon dioxide concentration data to make its mean 0 and standard deviation 1 to speed up model training and improve performance. Split the standardized time series data into input sequence and output sequence.

[0076] S4 Build LSTM model: Use deep learning frameworks such as TensorFlow or PyTorch to build the model. In this example, an LSTM layer with 64 hidden units and a fully connected layer Dense layer are used as the output layer. Select a suitable loss function such as mean square error MSE and optimizer such as Adam to compile the model.

[0077] S5 Model training and evaluation: Use the training set data to train the model, set the number of training rounds to 50, and the batch size to 32. During the training process, the model continuously adjusts the weights to minimize the loss function, and uses the validation set to monitor the model performance to prevent overfitting. Use the test set data to evaluate the trained model, and calculate the model's loss function value and other evaluation indicators on the test set, such as the mean absolute error MAE.

[0078] S6 Future CO2 Concentration Prediction: Prepare an input sequence to predict the CO2 concentration in the next 24 hours. If the time step is 3, the CO2 concentration data at the current time point and the previous two time points can be used as the input sequence. Input the prepared input sequence into the trained long short-term memory network model to obtain the predicted CO2 concentration value. After the prediction is completed, these data will be merged with the existing initial historical time series data to form a new continuous time series data set, and will be added to the data storage module after the initial historical time series data according to the index. For example, the original historical data of the lateral vibration attribute of the fan equipment is from 0 to 20 points. The long-term and short-term prediction network will generate time series data from 20 to 24 points after prediction. After adding it to the data storage module according to the index of "fan-lateral vibration", the two are merged into data from 0 to 24 points.

[0079] S102: Iteratively screening the device attribute joint list based on project requirement information to obtain a historical time series data set to be analyzed; wherein the project requirement information includes: analysis object information and time range information.

[0080] After obtaining the device attribute joint list based on the above step S101, in order to meet the user's research needs, before starting the data analysis process, select specific devices and device attributes as analysis targets. In the embodiment of the present application, the device attribute joint list will be iteratively screened according to the project requirement information to obtain the historical time series data set to be analyzed. It should be noted that the project requirement information includes: analysis object information and time range information.

[0081] Specifically, in one or more embodiments of the present application, the device attribute joint list is iteratively screened based on the project requirement information to obtain the historical time series data set to be analyzed, specifically including:

[0082] The device attribute joint list is screened according to the analysis object information to obtain the data playback analysis object list. For example: according to the formed device and device attribute joint list, the device and device attribute object to be analyzed are selected by checking, and a data playback analysis object list is formed. Then, the historical time series data set corresponding to the data playback object list is screened again according to the time range information to obtain the initial historical time series data set to be analyzed. According to the index data and time range information corresponding to the initial historical time series data set to be analyzed, the index retrieval information corresponding to the initial historical time series data set to be analyzed is constructed. According to the index retrieval information, the playback video and analysis results of the preset data storage module are retrieved to determine whether the initial historical time series data set to be analyzed exists. If it exists, it means that the user has completed an analysis of the same content once, and there is no need to analyze it again. The system will explain to the user that "the data playback analysis result already exists" and directly call the existing analysis result. Therefore, the existing initial historical time series data set to be analyzed is deleted, and the remaining initial historical time series data set to be analyzed is used as the subsequent historical time series data set to be analyzed. On the contrary, if the initial historical time series data set to be analyzed does not exist, the initial historical time series data set to be analyzed will be used as the historical time series data set to be analyzed. It should also be noted that in this process, in order to ensure the reliability and effectiveness of the analysis results, the number of data points in the selected time range will be strictly checked. For example: Based on actual engineering experience in a certain application scenario, the number of data points must be at least 1,000 to avoid serious deviations in the analysis results. If the number of data points is insufficient, the system will prompt the user to reselect the time range. For example, continuing with the previous example, it is identified that the time range of 12 o'clock to 18 o'clock for the analysis of "fan-lateral vibration" is only 800 data points, which is less than 1,000, so the user needs to reselect the time range to continue the subsequent steps. Similarly, it can be understood that in different scenarios, the number of data points can be selected based on different actual engineering experience and user needs.

[0083] S103: According to the identification strategy corresponding to each key characteristic, key features of the historical time series data set to be analyzed are detected to obtain key points corresponding to each of the historical time series data sets to be analyzed.

[0084] Since the key to data analysis is to accurately identify important time nodes with different characteristics from a vast ocean of data, in the embodiment of the present application, in order to enable the historical time series data set extracted from the massive data to be played back under the limited playback time limit, the embodiment of the present application will detect key features of the historical time series data set to be analyzed according to the identification strategy corresponding to each key characteristic, so as to obtain the key points corresponding to each historical time series data set to be analyzed.

[0085] Specifically, in one or more embodiments of the present application, according to the identification strategy corresponding to each key characteristic, the key features of the historical time series data set to be analyzed are detected to obtain the key points corresponding to each historical time series data set to be analyzed, which specifically includes the following process:

[0086] like Figure 2 As shown in the figure, seasonal decomposition or trend decomposition will be performed on the historical time series data set to be analyzed based on the STL algorithm to determine the seasonal characteristics or trend characteristics of the historical time series data set to be analyzed. In this process, for the extraction of trend components, a rough trend estimate is obtained by preliminary smoothing through local weighted regression, and then the trend estimate is continuously adjusted through inner loop iteration to minimize the variability of the remaining components. Finally, the outer loop adjusts the robustness weights of the data points according to the characteristics of the remaining components to obtain a robust trend component. For the extraction of seasonal components, seasonal subsequences are first constructed by grouping according to seasonal cycles. Then, each subsequence is smoothed to remove noise and outliers and obtain an average pattern. Finally, the smoothed subsequences are combined and adjusted to accurately reflect periodic changes. For the extraction of random components, the remaining part obtained after subtracting the trend and seasonal components from the original time series represents random fluctuations and noise that cannot be explained. For example, in the temperature monitoring system of intelligent buildings, temperature data is collected every minute for a month by collecting it once a day, and the STL decomposition method is used to reveal the periodicity and trend of temperature data. Then, the seasonal cycle is first set to 24 hours to match the characteristics of the data collected daily. In the STL parameter configuration, the period parameter is set to 24 hours to reflect the temperature change cycle within a day; the season parameter is set to 7 as the length of seasonal smoothing to capture temperature fluctuations; and the trend parameter is set according to 1 to 1.5 times of season, with a default value of 9 to depict long-term trends. The decomposition results will show the trend part, seasonal changes, and random fluctuations of the temperature. By observing these components, we can identify patterns in the data, such as whether the temperature has a trend of rising or falling at a specific time of the day, whether the seasonal changes are obvious, etc.

[0087] When the key feature to be analyzed of the historical time series data set to be analyzed is a stability feature, the discrete points corresponding to the historical time series data set to be analyzed are obtained, so as to determine the first key point corresponding to the stability feature of the historical time series data set to be analyzed according to the discrete points and the preset scoring criteria. Discrete points can significantly change the statistical characteristics of the time series data, such as the mean and variance. If these discrete points appear frequently, they may cause the mean and standard deviation of the data to change over time, thereby destroying the stability of the data. Among them, the preset scoring criteria are discrete points. If the value of the data point is more than three times the standard deviation away from the mean, these data points can be regarded as discrete points. In addition, the coefficient of variation is also a statistic for measuring the degree of data dispersion. Coefficient of variation = (standard deviation / mean value) × 100%. By directly observing the value of the coefficient of variation, an intuitive judgment can be made on the stability of the data. For example, in a certain scenario, a set of common scoring standards and grading in the time series data analysis of IoT devices can be formulated, such as the coefficient of variation is less than 0.2 when it is very stable, the coefficient of variation is less than 0.6 when it is stable, or the coefficient of variation is greater than 0.6 when it is unstable.

[0088] When the key feature to be analyzed of the historical time series data set to be analyzed is an abnormal condition feature, the second key point corresponding to the abnormal condition feature of the historical time series data set to be analyzed is determined according to the preset threshold and alarm mechanism corresponding to the historical time series data set to be analyzed. It should be noted that the preset threshold and alarm mechanism are set based on the data playback object corresponding to the historical time series data set to be analyzed. In this process, that is, in the process of abnormal condition analysis, the threshold of abnormal value detection and the alarm mechanism of abnormal condition can be set for the selected data playback analysis object according to the needs. Detect the abnormal values ​​in the historical data and the predicted data within the selected time range, and extract the occurrence time point and end time point of the alarm event according to the user-configured alarm mechanism. For example, the temperature range of the normal production environment in a warehouse is 18°C ​​to 24°C, and the humidity is 40% to 60%. Any reading beyond this threshold range should be considered abnormal. The alarm strategy is to start the alarm when the abnormal value threshold is reached, end the alarm when it returns to the normal range, and do not repeat the alarm within 5 minutes after the alarm starts.

[0089] When the key feature to be analyzed of the historical time series data set to be analyzed is a statistical feature, statistical calculation is performed on the historical time series data set to be analyzed to obtain the third key point corresponding to the statistical feature of the historical time series data set to be analyzed. For example: calculate the standard deviation, range, maximum value, minimum value, and average value of the data within the selected time range.

[0090] Summarize the first key point, the second key point and the third key point to obtain the key points corresponding to each historical time series data set to be analyzed. That is to say, the discrete points, maximum points, minimum points, abnormal value points, alarm occurrence points and end points obtained in the above process are all used as key points. Each key point represents its unique type characteristics, and they play a vital role in data analysis and monitoring. This process can comprehensively and flexibly capture the key points in the historical time series data set by identifying different types of key features such as seasonality, trend characteristics, stability characteristics, abnormal condition characteristics, and statistical characteristics. By decomposing and extracting different components of the data, the inherent laws and characteristics of the data are clearer and more explainable. This helps users better understand the data, discover patterns and trends in the data, and make more informed decisions.

[0091] S104: Traversing the historical time series data set to be analyzed according to a preset sliding time window to determine a continuous time region according to the number of key points in each sliding time window and the average number of key points in each sliding time window.

[0092] The shortcomings of the prior art in feature extraction, especially the lack of a sophisticated and targeted screening mechanism, have resulted in the inability to effectively highlight the most critical data features within a limited playback time. Therefore, in order to further extract a limited number of key time areas from a large amount of key information for key display, it is necessary to divide these key points into different areas and display the detailed situations in a limited number of the most important areas. Therefore, in the embodiment of the present application, the historical time series data set to be analyzed is traversed according to a preset sliding time window to determine the continuous time area based on the number of key points in each sliding time window and the average number of key points in each sliding time window.

[0093] Specifically, in one or more embodiments of the present application, the historical time series data set to be analyzed is traversed according to the preset sliding time window to determine the continuous time region according to the number of key points in each sliding time window and the average number of key points in each sliding time window, specifically including the following process:

[0094] First, determine the complete analysis time range corresponding to the historical time series data set to be analyzed, and traverse the historical time series data set to be analyzed within the complete analysis time range according to the preset sliding time window. For example, assuming that the size of the preset sliding time window is L, and 100 data points are used as a sliding time window, then the sliding time window is used to traverse the entire analysis time range, and the number of key points in each sliding window is recorded as , which needs to be explained is Indicates Then, the number of key points corresponding to each key point type in each sliding time window during the traversal process is accumulated to obtain the number of key points in each sliding time window. .

[0095] At the same time, according to the complete analysis time range and the time length L of the preset sliding time window, the total number N of sliding time windows that can be preset is determined, so as to determine the average number of key points in each preset sliding time window based on the number of key points in each sliding time window and the total number of sliding windows. ,Right now Represents the average number of key points in each preset sliding time window. Then, based on the relationship between the number of current preset sliding time windows and the total number of preset sliding time windows, and the relationship between the number of key points in the preset sliding time window and the average number, the continuous time area corresponding to the historical time series data set to be analyzed is determined. In the above method, by traversing the historical time series data set through the preset sliding time window, it is possible to achieve refined screening of data features. The number of key points in each sliding window reflects the activity and importance of the data in this time period, which helps to extract the most critical time area from massive data. Within the limited playback or display time, by determining the continuous time area and focusing on the detailed situations in these areas, the most critical data features can be effectively highlighted. This method avoids information overload, enables users to grasp the key points more quickly, and improves the efficiency of information acquisition.

[0096] Further, in one or more embodiments of the present application, according to the relationship between the number of current preset sliding time windows and the total number of preset sliding time windows, and the relationship between the number of key points in the preset sliding time windows and the average number, the continuous time region corresponding to the historical time series data set to be analyzed is determined, specifically including the following process:

[0097] If it is determined that the number of the current preset sliding time windows is less than the total number of preset sliding time windows, and the number of key points in the preset sliding time window is greater than the average number, then the position of the preset sliding time window is recorded as the position of the starting sliding window, and the starting time point of the starting sliding window is determined. That is, all time windows are traversed in chronological order. Stop traversal when . Start judging from the first sliding window to determine whether If not, continue to judge the next sliding time window. If satisfied, record the position of the starting sliding window And the starting time point of the sliding window Only when a starting point is recorded After that, we start looking for its corresponding end point If it is determined that the number of the current preset sliding time windows is less than the total number of the preset sliding time windows, and the number of key points in the preset sliding time windows is less than or equal to the average number, then the position of the current preset sliding time window is recorded as the position of the end sliding window, and the end time point of the end sliding window is determined. , continue to judge the next time window until Stop when , and record the position of this sliding window And the starting time of the sliding window ,in Only when a pair of starting points and end point After finding all of them, start looking for the next starting point.

[0098] If it is determined that the number of the current preset sliding time windows is greater than the total number of preset sliding time windows, then the traversal process is stopped directly, and the position of the last preset sliding time window is recorded as the position of the end sliding window, and the end time point of the end sliding window is determined. Then, according to the starting time point and the end time point, the continuous time area corresponding to the historical time series data set to be analyzed is determined. For example: when analyzing the operating data of a certain power system, through the above traversal and judgment process, it is accurately found that the number of key points from the 5th sliding window is greater than the average value until the number of key points in the 12th sliding window is less than the average value, then based on the determined starting time point and end time point, an important continuous area can be determined. Among them, it should be noted that only when the first data point of the 5th sliding window and the first data point of the 12th sliding window are found, the next starting point can be found. For example, in the 20th sliding window, the number of key points is greater than the average value again, and the first data point of the 20th sliding window is a new starting point.

[0099] S105: Sorting is performed based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine the longest continuous time region as the display region, so as to generate visualization content of the historical time series data set to be analyzed in the display region for playback.

[0100] After the continuous time regions are acquired based on the above step S104, the embodiment of the present application will sort all the selected continuous regions according to the number of preset sliding time windows corresponding to each continuous time region, so as to determine the longest continuous time region as the display area, thereby facilitating the generation of visualization content of the historical time series data set to be analyzed in the display area for playback.

[0101] Specifically, in one or more embodiments of the present application, sorting is performed based on the number of preset sliding time windows corresponding to each continuous time region to determine the longest continuous time region as the display region, specifically including:

[0102] The continuous time regions are sorted based on the number of preset sliding time windows corresponding to each continuous time region to determine the longest preset number of continuous time regions, and the preset number of continuous time regions are used as display regions. If the number of continuous time regions is less than the preset number of segments, all continuous time regions are used as display regions.

[0103] Furthermore, when the existing method of combining generative AI technology and precise prompt technology to replay the historical time series data set to be analyzed based on dynamic video, the generation technology is usually limited to only being able to produce a maximum of 1 minute of coherent video content, which may not fully meet the user's demand for video length. Secondly, at the commercial application level, there are still hidden dangers such as privacy leakage, data pollution and property rights protection. Therefore, in one or more embodiments of the present application, the visualization content of the historical time series data set to be analyzed in the display area is generated for playback, specifically including:

[0104] First, the historical time series data set to be analyzed in the display area is input, and the preset data playback engine is input to convert the historical time series data set to be analyzed in the display area into visualization content according to the preset data playback engine; it should be noted that the visualization content includes: static charts and dynamic videos. Then, based on the preset playback script, the static charts and dynamic videos of the historical time series data set to be analyzed are played back. It should be noted here that static charts contain four visualization elements: line charts, bar charts, tables, and box plots. Four static line charts are used to respectively show an overview of the time series data within the time range and the trend components, seasonal components, and random components therein. Users can focus on the time range they are interested in by dragging the time axis attached to the bottom of the line chart, so as to observe and analyze the data of a specific time period in depth. In the time series data overview, historical data uses a white background, and predicted data uses a blue background. On this basis, the areas that are screened out and need to be displayed are highlighted with a red background in the line chart and the time axis; the table is used to show the stability judgment results, standard deviation, range, maximum value, minimum value, and average value, and the box plot is used to describe the characteristics within the selected time range. The two together show the statistical characteristics of the time series data within the selected time range. The bar chart is used to display the number of discrete points, abnormal value points, alarm occurrence points and end points, and maximum value points.

[0105] The dynamic video imitates the form of PPT slide show, automatically showing the selected area range in sequence. In the video, the dynamic line chart depicts the data fluctuations of the display area over time. The key points in the display area are clearly distinguished by labels of different colors, and the timeline below the video is assisted in an intuitive way: the red background highlights the specific position of each display area on the timeline, while the gray area clearly indicates the position of the current display area on the timeline. At the same time, a detailed table is synchronized with the video, which not only lists the types of key points and the specific time they appear, but also ensures that the audience can obtain the latest data information through real-time scrolling and refreshing.

[0106] Specifically, the process can be transformed into intuitive visualization content through the data playback engine, using static charts to comprehensively display the macro characteristics of the analyzed data, and dynamic videos to display the data evolution process of the selected areas. According to the preset playback script, after the user clicks the "Start" button, the system first displays a static chart for 15 seconds so that the user can have a clear understanding of the data overview. Then, the system will smoothly transition and switch to dynamic video display. The video is displayed smoothly with a duration of 30 seconds per area to ensure that users can track the dynamic changes of the data. In addition, the speed of video playback will be intelligently adjusted according to the number of data points in each area, that is, the playback rate in each area will be different to ensure that the number of data points displayed per second matches the data density of the area, thereby providing users with a detailed and easy-to-understand data evolution perspective. The generated playback video is stored in the data storage module according to the index of "device + device attribute + analysis time range". The user will perform detection in the subsequent process. If the detection exists, the time for data analysis and playback video generation can be saved, saving the user waiting time. In order to avoid excessive storage space occupation, we have implemented a regular data management strategy. According to this strategy, we will perform periodic cleanup of stored analysis results, playback videos, and other data. Specifically, all data and video content that has been stored for more than one week will be automatically marked and cleared by the system.

[0107] like Figure 3 As shown, the present application embodiment provides a schematic diagram of the structure of a time series data playback device for an Internet of Things device. Figure 3 It can be seen that in one or more embodiments of the present application, a time series data playback device for an IoT device includes:

[0108] at least one processor; and,

[0109] a memory communicatively connected to the at least one processor; wherein,

[0110] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: execute any of the above-mentioned methods.

[0111] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0112] The above describes specific embodiments of the present application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0113] The above is only one or more embodiments of the present application and is not intended to limit the present application. For those skilled in the art, one or more embodiments of the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of one or more embodiments of the present application should be included in the scope of the claims of the present application.

Claims

1. A time series data playback method for an Internet of Things device, characterized in that: The method comprises: The data playback system screens the historical time series data set according to the preset data set conditions, and determines the device attribute joint list based on the index data corresponding to the historical time series data set; Iteratively screening the device attribute joint list based on project requirement information to obtain a historical time series data set to be analyzed; wherein the project requirement information includes: analysis object information and time range information; According to the identification strategy corresponding to each key characteristic, the key features of the historical time series data set to be analyzed are detected to obtain the key points corresponding to each of the historical time series data sets to be analyzed; Traversing the historical time series data set to be analyzed according to a preset sliding time window to determine a continuous time region according to the number of key points in each sliding time window and the average number of key points in each sliding time window; Sorting based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine the longest continuous time region as the display region, so as to generate visualization content of the historical time series data set to be analyzed in the display region for playback; Among them, according to the identification strategy corresponding to each key characteristic, the key features of the historical time series data set to be analyzed are detected to obtain the key points corresponding to each historical time series data set to be analyzed, which specifically includes: When the key feature to be analyzed of the historical time series data set to be analyzed is a stability feature, the coefficient of variation corresponding to the historical time series data set to be analyzed is obtained, so as to determine the first key point corresponding to the stability feature of the historical time series data set to be analyzed according to the coefficient of variation and a preset scoring standard; When the key feature to be analyzed of the historical time series data set to be analyzed is an abnormal condition feature, a second key point corresponding to the abnormal condition feature of the historical time series data set to be analyzed is determined according to a preset threshold and an alarm mechanism corresponding to the historical time series data set to be analyzed; wherein the preset threshold and the alarm mechanism are set based on a data playback object corresponding to the historical time series data set to be analyzed; When the key feature to be analyzed of the historical time series data set to be analyzed is a statistical feature, statistical calculation is performed on the historical time series data set to be analyzed to obtain a third key point corresponding to the statistical feature of the historical time series data set to be analyzed; Summarizing the first key point, the second key point and the third key point to obtain key points corresponding to each of the historical time series data sets to be analyzed; According to the number of key points in each sliding time window and the average number of key points in each sliding time window, the continuous time region is determined, including: If it is determined that the number of the current preset sliding time windows is less than the total number of preset sliding time windows, and the number of key points in the preset sliding time windows is greater than the average number, then the position of the preset sliding time window is recorded as the position of the starting sliding window, and the starting time point of the starting sliding window is determined; Recording the starting time point of the starting sliding window, so as to search for the ending time point based on the starting time point; If it is determined that the number of the current preset sliding time windows is less than the total number of preset sliding time windows, and the number of key points in the preset sliding time windows is less than or equal to the average number, then the position of the preset sliding time window is recorded as the position of the end sliding window, and the starting time point of the end sliding window is determined to be the end time point corresponding to the starting time point of the start sliding window; If it is determined that the number of the current preset sliding time windows is greater than the total number of preset sliding time windows, the position of the last preset sliding time window is recorded as the position of the end sliding window, and the starting time point of the end sliding window is determined to be the end time point corresponding to the starting time point of the start sliding window; According to the starting time point of the starting sliding window and the ending time point corresponding to the starting time point of the starting sliding window, the continuous time region corresponding to the historical time series data set to be analyzed is determined.

2. According to claim 1, a time series data playback method for an Internet of Things device is characterized in that: The data playback system screens the historical time series data set and determines the device attribute joint list corresponding to the historical time series data set according to the preset data set conditions and the preset index format, specifically including: Detect whether there are missing values ​​in each data set in the database, so as to determine the integrity test results of each data set according to the missing value test results; If it is determined that the integrity monitoring result meets the preset integrity requirement, a trend analysis is performed on each data set based on the time sequence to determine the volatility detection result of each data set; If it is determined that the volatility detection result meets the preset volatility requirement, obtaining the initial historical time series data corresponding to the data set; Predicting the initial historical time series data according to the long short-term memory network to obtain a prediction result, and adding the prediction result to the initial historical time series data to obtain a historical time series data set; The original index data of the historical time series data set is obtained to determine the Internet of Things devices and Internet of Things device attributes corresponding to the historical time series data set based on the index data, and a corresponding device attribute joint list is generated based on the Internet of Things devices and the Internet of Things device attributes; wherein the Internet of Things device corresponds to multiple Internet of Things device attributes.

3. The time series data playback method of an Internet of Things device according to claim 2 is characterized in that: Predicting the initial historical time series data according to the long short-term memory network to obtain a prediction result, and adding the prediction result after the initial historical time series data and before obtaining the historical time series data set, the method further includes: Obtain a historical time series data set within a preset period and divide it into a training set, a validation set, and a test set; Standardizing the historical time series data set in the training set to obtain a standardized historical time series data set, and dividing the standardized historical time series data set into an input sequence and an output sequence; Constructing an initial long short-term memory network based on a deep learning framework, training the initial long short-term memory network according to the input sequence and the output sequence, and performing performance monitoring based on the validation set to obtain a trained long short-term memory network; The trained long short-term memory network is evaluated based on the test set to obtain a long short-term memory network, so as to predict the initial historical time series data according to the long short-term memory network.

4. The time series data playback method of an Internet of Things device according to claim 1, characterized in that: Based on the project requirement information, the device attribute joint list is iteratively screened to obtain the historical time series data set to be analyzed, including: Filter the device attribute joint list according to the analysis object information to obtain a data playback analysis object list; Perform secondary screening on the historical time series data set corresponding to the data playback object list according to the time range information to obtain an initial historical time series data set to be analyzed; Constructing index retrieval information corresponding to the initial historical time series data set to be analyzed based on the index data corresponding to the initial historical time series data set to be analyzed and the time range information; According to the index retrieval information, the playback video and analysis results of the preset data storage module are retrieved to determine whether the initial historical time series data set to be analyzed exists; If yes, the existing initial historical time series data set to be analyzed is deleted to obtain the historical time series data set to be analyzed; If not, the initial historical time series data set to be analyzed is used as the historical time series data set to be analyzed.

5. The time series data playback method of an Internet of Things device according to claim 1, characterized in that: The historical time series data set to be analyzed is traversed according to the preset sliding time window to determine the continuous time area according to the number of key points in each sliding time window and the average number of key points in each sliding time window, specifically including: Determine a complete analysis time range corresponding to the historical time series data set to be analyzed, so as to traverse the historical time series data set to be analyzed within the complete analysis time range according to a preset sliding time window; The number of key points corresponding to each key point type in each sliding time window during the traversal process is accumulated to obtain the number of key points in each sliding time window; Determine the total number of the preset sliding time windows according to the complete analysis time range and the time length of the preset sliding time window, and determine the average number of key points in each of the preset sliding time windows according to the number of key points in each sliding time window and the total number; According to the relationship between the number of current preset sliding time windows and the total number of preset sliding time windows, and the relationship between the number of key points in the preset sliding time windows and the average number, the continuous time region corresponding to the historical time series data set to be analyzed is determined.

6. The time series data playback method of an Internet of Things device according to claim 1, characterized in that: Sorting based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine the longest continuous time region as the display region specifically includes: Sorting is performed based on the number of preset sliding time windows corresponding to each of the continuous time regions to determine a preset number of continuous time regions with the longest length, and using the preset number of continuous time regions as display regions; If the number of the continuous time areas is less than the preset number of segments, all the continuous time areas are used as display areas.

7. The time series data playback method of an Internet of Things device according to claim 1, characterized in that: Generating visualization content of the historical time series data set to be analyzed in the display area for playback, specifically including: Inputting the historical time series data set to be analyzed in the display area into a preset data playback engine, so as to convert the historical time series data set to be analyzed in the display area into visualization content according to the preset data playback engine; wherein the visualization content includes: static charts and dynamic videos; The static charts and dynamic videos of the historical time series data set to be analyzed are played back based on the preset playback script.

8. A time series data playback device for an Internet of Things device, characterized in that: The device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can: execute any of the methods described in claims 1-7.

Images