Intelligent early warning discrimination method and device based on data characteristics, equipment and medium

Abstract

The invention relates to the technical field of intelligent early warning, in particular to an intelligent early warning judgment method and device based on data characteristics, equipment and a medium. The method comprises the following steps: acquiring Internet of Things monitoring data streams in real time, and dynamically calculating a multi-dimensional characteristic vector corresponding to each data stream based on a historical sequence and a real-time state of each data stream; for each data stream, based on the historical sequence of each data stream and the corresponding multi-dimensional characteristic vector, generating a dynamic prediction baseline corresponding to each data stream through a trained time sequence prediction model; comparing the real-time data of each data stream with the corresponding dynamic prediction baseline to obtain a real-time comparison result; determining a discrimination mode based on the real-time comparison result and a dynamic index in the multi-dimensional characteristic vector; and carrying out monitoring and early warning according to the judgment mode determined by each data stream. Different early warning judgment strategies are adopted. And the early warning triggering condition is dynamically adjusted according to the multi-dimensional characteristic vector, so that the data early warning accuracy is improved.

Description

Technical Field

[0001] This application relates to the field of intelligent early warning technology, and in particular to an intelligent early warning discrimination method, device, equipment and medium based on data characteristics. Background Technology

[0002] Currently, with the widespread application of IoT technology, various IoT platforms have become core data hubs in fields such as environmental monitoring, industrial control, and smart agriculture. These platforms integrate heterogeneous monitoring data streams from multiple sources, including drones, ground sensors, weather stations, and industrial equipment, and visualize them through large data dashboards. Users can then intuitively see the actual status of various data on the IoT platform. Among these, the early warning function, as a key module for ensuring business security and timely response, has the core task of accurately and promptly identifying abnormal states and triggering alarms from massive, real-time data inflows.

[0003] In existing IoT early warning scenarios, fixed threshold-based warning schemes are commonly used. For example, administrators manually set monitoring parameters and thresholds for different data, triggering an alert when the monitored data exceeds the preset threshold. This could involve configuring a uniform detection frequency (e.g., scanning every 5 minutes) and a fixed sensitivity threshold for the monitored data items. However, such methods lack precision in data monitoring and early warning. For instance, forest fire smoke concentration data is dynamic, spatially correlated, highly volatile, and time-sensitive, requiring rapid capture at the second or even millisecond level. Soil moisture data, on the other hand, exhibits static, time-series, moderate volatility, and moderate time sensitivity, and does not suddenly change significantly except through human intervention. Therefore, using low-frequency scanning for the former can lead to disasters due to monitoring delays, while using high-frequency scanning for the latter generates numerous invalid alarms, wasting computing resources and interfering with judgment. Furthermore, monitoring thresholds for these two types of data should not be set solely based on experience.

[0004] Therefore, current data early warning methods do not take into account the different characteristics of data in different regions, scenarios, and environments, resulting in poor accuracy of data early warning in the event of data anomalies, and high false alarm and false alarm rates. Summary of the Invention

[0005] In view of the above-mentioned shortcomings and deficiencies of the prior art, this application provides an intelligent early warning discrimination method, device, equipment and medium based on data characteristics. The main purpose is to solve the problems of poor accuracy of data early warning and high false alarm rate and false alarm rate when data is abnormal.

[0006] To achieve the above objectives, the main technical solutions adopted in this application include:

[0007] In a first aspect, embodiments of this application provide an intelligent early warning and discrimination method based on data characteristics, including:

[0008] The system acquires IoT monitoring data streams in real time and dynamically calculates a multi-dimensional characteristic vector for each data stream based on its historical sequence and real-time status. The indicators of the dimensional characteristics include at least a dynamic index that characterizes the rate of data change and a fluctuation entropy value that characterizes the randomness of the data.

[0009] For each data stream, based on the historical sequence of each data stream and the corresponding multi-dimensional characteristic vector, a dynamic prediction baseline is generated for each data stream through a pre-trained time series prediction model; the dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period.

[0010] The real-time data of each data stream is compared with the corresponding dynamic prediction baseline to obtain the real-time comparison result;

[0011] Based on the real-time comparison results and the dynamic index in the multi-dimensional characteristic vector, a discrimination mode is determined; wherein, the discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation, and the warning triggering conditions in the first discrimination mode and the second discrimination mode are determined based on the fluctuation entropy value;

[0012] Monitoring and early warning are performed based on the discrimination pattern determined for each data stream.

[0013] Optionally, determining the discrimination mode based on the real-time comparison result and the dynamic index in the multi-dimensional feature vector includes: determining a deviation factor based on the deviation magnitude of the real-time data from the corresponding dynamic prediction baseline in the real-time comparison result; calculating an adjustment factor based on the dynamic index; calculating a mode discrimination threshold based on the deviation factor and the adjustment factor; and determining the mode as a first discrimination mode if the dynamic index is greater than the mode discrimination threshold, otherwise determining it as a second discrimination mode.

[0014] Optionally, the monitoring and early warning according to the discrimination mode determined for each data stream includes: when executing the first discrimination mode, calculating the instantaneous deviation value between the real-time data of the data stream and the dynamic prediction baseline; triggering an early warning when the instantaneous deviation value is greater than an instantaneous deviation threshold dynamically determined based on the fluctuation entropy value; wherein the instantaneous deviation threshold is positively correlated with the fluctuation entropy value; when executing the second discrimination mode, performing an integral calculation on the deviation of the real-time data from the stated magnitude within a preset time window to obtain a trend accumulation amount; triggering an early warning when the trend accumulation amount is greater than a trend threshold dynamically determined based on the fluctuation entropy value; wherein the trend threshold is negatively correlated with the fluctuation entropy value.

[0015] Optionally, the indicators of the dimensional characteristics also include a spatiotemporal index characterizing the temporal and spatial characteristics of the data and a timeliness index characterizing the maximum time to process data after an anomaly.

[0016] After dynamically calculating the multi-dimensional characteristic vector corresponding to each data stream, the method further includes: determining the monitoring parameters corresponding to each data stream based on the multi-dimensional characteristic vector corresponding to each data stream.

[0017] Optionally, the indicators of the dimensional characteristics may also include a correlation index, which characterizes whether the current data stream is correlated with other data streams;

[0018] The monitoring and early warning according to the discrimination mode determined for each data stream also includes: when the current data stream is executing the second discrimination mode, increasing the monitoring parameters of other data streams that are related to the current data stream, so as to increase the monitoring frequency of other related data streams.

[0019] Optionally, the monitoring and early warning according to the discrimination pattern determined for each data stream further includes:

[0020] When multiple alerts are triggered simultaneously, a basic priority score is calculated based on the multi-dimensional characteristic vector corresponding to the data stream that triggered the alert; the priority of the multiple data streams that triggered the alert is determined based on the priority score, the business scenario score of the data stream that triggered the alert, and the real-time impact factor; and the data streams that triggered the alert are sorted and displayed according to the priority.

[0021] Optionally, after monitoring and issuing early warnings according to the discrimination mode determined for each data stream, the method further includes: obtaining the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback of the data stream that triggers the early warning; and optimizing the monitoring parameters of each data stream based on the gradient boosting tree optimization model, using the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback as input.

[0022] Secondly, embodiments of this application provide an intelligent early warning and discrimination device based on data characteristics, comprising:

[0023] The acquisition unit is configured to acquire IoT monitoring data streams in real time and dynamically calculate the multi-dimensional characteristic vector corresponding to each data stream based on the historical sequence and real-time status of each data stream; the indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data.

[0024] The generation unit is configured to generate a dynamic prediction baseline for each data stream based on the historical sequence of each data stream and the corresponding multi-dimensional feature vector, using a pre-trained time series prediction model; the dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period.

[0025] The comparison unit is configured to compare the real-time data of each data stream with the corresponding dynamic prediction baseline to obtain a real-time comparison result;

[0026] The determining unit is configured to determine a discrimination mode based on the real-time comparison results and the dynamic index in the multi-dimensional characteristic vector; wherein, the discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation, and the warning triggering conditions in the first discrimination mode and the second discrimination mode are determined based on the fluctuation entropy value;

[0027] The processing unit is configured to monitor and issue early warnings according to the discrimination pattern determined for each data stream.

[0028] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the intelligent early warning and discrimination method based on data characteristics described in the first aspect.

[0029] Fourthly, this application provides an electronic device, including a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, wherein the processor executes the computer program to implement the intelligent early warning and discrimination method based on data characteristics described in the first aspect.

[0030] Using the above technical solution, this application provides an intelligent early warning discrimination method based on data characteristics. First, it acquires IoT monitoring data streams in real time, and dynamically calculates a multi-dimensional characteristic vector corresponding to each data stream based on its historical sequence and real-time status. The indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data. For each data stream, based on its historical sequence and corresponding multi-dimensional characteristic vector, a dynamic prediction baseline is generated for each data stream using a pre-trained time-series prediction model. The dynamic prediction baseline characterizes the expected fluctuation range of the data stream in the next time period. The real-time data of each data stream is compared with the corresponding dynamic prediction baseline to obtain a real-time comparison result. Based on the real-time comparison result and the dynamic index in the multi-dimensional characteristic vector, a discrimination mode is determined. The discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation. The early warning triggering conditions in the first and second discrimination modes are determined based on the fluctuation entropy value. Monitoring and early warning are performed according to the discrimination mode determined for each data stream. Compared to related technologies, this approach first uses multi-dimensional feature vectors to reflect the characteristics of different data. Then, a time-series prediction model generates a dynamic prediction baseline for each data stream. Based on the real-time comparison between the data and the dynamic prediction baseline, a warning discrimination mode is determined, enabling the application of different warning discrimination strategies to different types of data. Furthermore, the warning triggering conditions in different discrimination modes can be dynamically and adaptively adjusted based on the multi-dimensional feature vectors, replacing rigid, manually fixed thresholds, thereby improving the accuracy of data-driven warnings. Attached Figure Description

[0031] Figure 1 A flowchart illustrating an intelligent early warning and discrimination method based on data characteristics provided in an embodiment of this application;

[0032] Figure 2 A flowchart illustrating another intelligent early warning and discrimination method based on data characteristics provided in this application embodiment;

[0033] Figure 3 This is a schematic diagram of an intelligent early warning and discrimination device based on data characteristics, provided in an embodiment of this application. Detailed Implementation

[0034] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0035] To address the current issues of poor accuracy and high false alarm rates in data anomaly alerts, this application proposes an intelligent early warning discrimination method based on data characteristics. This method can be applied to monitoring systems in fields such as agricultural management, forestry resource assessment, carbon resource assessment, data visualization, and early warning management. During operation, it can execute any of the intelligent early warning discrimination methods based on data characteristics mentioned below. Figure 1 As shown, the method includes:

[0036] S101 acquires IoT monitoring data streams in real time and dynamically calculates the multi-dimensional characteristic vector corresponding to each data stream based on the historical sequence and real-time status of each data stream.

[0037] In S101, the IoT platform can be a monitoring platform widely used in meteorological and ecological monitoring, industrial equipment management, environmental governance, and other fields. It uses a large data dashboard as the core visualization carrier to display various monitoring data streams, such as carbon footprint / forest fire prevention data collected by drones, regional water, soil, and atmospheric data, and data from monitoring stations. Specifically, this data can include soil moisture and water quality data collected by ground stations, operating temperature and vibration parameters of various industrial equipment (such as drones and data collection stations), and real-time data such as carbon dioxide concentration and soil moisture. Furthermore, it can identify data anomalies and trigger warnings to remind users to take appropriate action.

[0038] Furthermore, traditional early warning methods often employ uniform monitoring parameters and fixed thresholds, failing to consider the inherent characteristics of the data. Therefore, this embodiment first constructs multi-dimensional characteristic vectors to characterize the inherent change patterns of different data streams before making early warning judgments. The dimensional characteristics aim to represent the inherent changes of the data stream in a specific form, and the indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data.

[0039] The dynamic index is used to quantify the severity and speed of data changes. It can be based on parameters such as the mean square deviation or absolute mean of the data sequence within a preset time window (e.g., the most recent hour). The specific method used depends on the data type. If the dynamic index of the data stream is high, such as the current of industrial equipment, the system needs to have a faster response capability for this data; conversely, if the dynamic index of the data stream is low, such as the organic matter content of soil, the system should pay more attention to its long-term trend. In the subsequent steps of this embodiment, different early warning discrimination modes are set for different data, and the dynamic index is used as the basis for determining the early warning discrimination mode, thus taking into account the characteristics of different data during the early warning process and improving the accuracy of the early warning. The fluctuation entropy value is used to quantify the randomness of the data. Specifically, permutation entropy or approximate entropy can be used as the basis. The fluctuation entropy value reflects the fluctuation range of the data stream, that is, the amplitude of the vibration. For data streams with large fluctuation entropy values, a greater tolerance should be given during the early warning to avoid false alarms. Therefore, using the fluctuation entropy value as the basis for the early warning trigger condition reduces the false alarm rate.

[0040] In some embodiments, dimensional characteristics may also include spatiotemporal indices, correlation indices, and timeliness indices, thereby addressing problems in the corresponding aspects.

[0041] S102, for each data stream, based on the historical sequence of each data stream and the corresponding multi-dimensional feature vector, a dynamic prediction baseline is generated for each data stream through a pre-trained time series prediction model.

[0042] The dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period, which is a kind of prediction and foresight. Traditionally, a fixed threshold is used; if the data exceeds the fixed threshold, an alert is triggered. This embodiment uses the dynamic prediction baseline as a "basic threshold," which can be considered as roughly the range the data should fall within under normal circumstances. Then, the real-time data of each data stream is compared with the corresponding dynamic prediction baseline. If the comparison results show a significant difference, subsequent alert determination is performed.

[0043] The pre-trained time-series prediction model can utilize existing mature models, as long as they can capture complex nonlinear temporal dependencies. Examples include Long Short-Term Memory (LSTM) networks or temporal convolutional network models. By inputting the historical sequence of the data stream over a given period and the corresponding multi-dimensional feature vectors into the model, a dynamic prediction baseline can be obtained. Furthermore, in addition to historical sequences and feature vectors, contextual information, such as the current values ​​of related data streams, equipment operating conditions, and weather forecasts, can also be input into the model. This allows the dynamic prediction baseline to be more aware of the current system state and environment, making it more context-sensitive.

[0044] S103, compare the real-time data of each data stream with the corresponding dynamic prediction baseline to obtain the real-time comparison result.

[0045] S104, based on real-time comparison results and the dynamic index in the multi-dimensional characteristic vector, determines the discrimination mode.

[0046] In S104, the discrimination mode, also called the early warning discrimination mode, includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation. This embodiment does not use a single discrimination logic for all data. Firstly, the inherent characteristics and real-time performance of the data determine which early warning discrimination mode is used for data with different characteristics. As mentioned above, the dynamic index is specifically used to quantify the severity and speed of data changes. That is, data streams with higher dynamic indices require faster system response capabilities, and therefore are determined to use the first discrimination mode based on instantaneous deviation; conversely, those with lower dynamic indices are more likely to be determined to use the second discrimination mode based on trend accumulation.

[0047] Furthermore, the warning triggering conditions in both the first and second discrimination modes are determined based on the fluctuation entropy value in the characteristic vector. Specifically, the warning triggering condition is the warning triggering threshold. Although this embodiment also uses a "threshold-triggered warning" method in the first or second discrimination mode, the set threshold is dynamic, not static as in traditional solutions. This allows for dynamic determination of the warning threshold for different scenarios and data types, thereby improving the accuracy of warning discrimination.

[0048] S105 monitors and issues warnings according to the discrimination pattern determined for each data stream.

[0049] In this embodiment, IoT monitoring data streams are first acquired in real time, and a multi-dimensional characteristic vector corresponding to each data stream is dynamically calculated based on the historical sequence and real-time status of each data stream. The indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data. For each data stream, a dynamic prediction baseline is generated based on the historical sequence and the corresponding multi-dimensional characteristic vector of each data stream through a pre-trained time series prediction model. The dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period. The real-time data of each data stream is compared with the corresponding dynamic prediction baseline to obtain a real-time comparison result. Based on the real-time comparison result and the dynamic index in the multi-dimensional characteristic vector, a discrimination mode is determined. The discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation. The warning triggering conditions in the first and second discrimination modes are determined based on the fluctuation entropy value. Monitoring and warning are performed according to the discrimination mode determined for each data stream. Compared to related technologies, this approach first uses multi-dimensional feature vectors to reflect the characteristics of different data. Then, a time-series prediction model generates a dynamic prediction baseline for each data stream. Based on the real-time comparison between the data and the dynamic prediction baseline, a warning discrimination mode is determined, enabling the application of different warning discrimination strategies to different types of data. Furthermore, the warning triggering conditions in different discrimination modes can be dynamically and adaptively adjusted based on the multi-dimensional feature vectors, replacing rigid, manually fixed thresholds, thereby improving the accuracy of data-driven warnings.

[0050] Optionally, the discrimination mode is determined based on the real-time comparison results and the dynamic index in the multi-dimensional feature vector, including: determining the deviation factor based on the deviation of the real-time data from the corresponding dynamic prediction baseline in the real-time comparison results; calculating the adjustment factor based on the dynamic index; calculating the mode discrimination threshold based on the deviation factor and the adjustment factor; if the dynamic index is greater than the mode discrimination threshold, it is determined as the first discrimination mode, otherwise it is the second discrimination mode.

[0051] In this embodiment, the deviation of the real-time data from the corresponding dynamic prediction baseline in the real-time comparison results represents the difference between the currently monitored real-time data and the baseline value of the data under normal conditions predicted by the dynamic prediction baseline. If the difference is large, an anomaly may occur. Therefore, the deviation factor reflects the urgency of the sudden anomaly in the current data. Its design principle is: the larger the deviation, the larger the value of the deviation factor, so that the system is more inclined to select the discrimination mode that can respond quickly.

[0052] The dynamic index reflects the inherent rate of change and activity of the data itself. A data stream with a high dynamic index (such as device current) typically has a wide normal fluctuation range. However, if its dynamic index fluctuates drastically in the near future (i.e., poor stability), it means that the data stream is currently in an "unstable" period. Using an overly sensitive discrimination model in this situation can easily lead to false alarms. Therefore, the adjustment factor is used to characterize the stability of the data's dynamics. Its design principle is: the greater the recent fluctuation of the dynamic index (the more unstable), the larger the value of the adjustment factor should be, thus raising the threshold for model selection and prompting the system to adopt a more robust and interference-resistant discrimination model. A reference calculation formula for the deviation factor and adjustment factor is given below. It should be noted that the deviation factor and adjustment factor only need to be designed according to the above logic; the following calculation formula is only an example:

[0053] (Formula 1)

[0054] In Formula 1, DF is the deviation factor, and A is the instantaneous deviation magnitude. For reference deviation range, This is a preset upper limit constant.

[0055] (Formula 2)

[0056] In Formula 2, RF is the adjustment factor. A coefficient greater than 0 is used to control the strength of the influence of volatility on the adjustment factor. This is the standard deviation of the dynamic index over the recent sliding time window, used to measure its volatility or instability. For reference fluctuation values, they can be set based on historical typical values ​​of the data stream or human experience.

[0057] Specifically, the pattern discrimination threshold Please refer to Formula 3:

[0058] (Formula 3)

[0059] In Formula 3, For pattern discrimination threshold, Set a threshold based on the base. The coefficient is greater than 0. If the current dynamic index Dt is greater than the pattern discrimination threshold, the first discrimination mode is selected; otherwise, the second discrimination mode is selected. In this way, different early warning discrimination strategies are applied to different types of data.

[0060] Optionally, monitoring and early warning are performed according to a discrimination mode determined for each data stream, including: when executing the first discrimination mode, calculating the instantaneous deviation value between the real-time data of the data stream and the dynamic prediction baseline; triggering an early warning if the instantaneous deviation value is greater than an instantaneous deviation threshold dynamically determined based on the fluctuation entropy value; wherein, the instantaneous deviation threshold is positively correlated with the fluctuation entropy value; when executing the second discrimination mode, integrating the magnitude of the real-time data deviation within a preset time window to obtain the trend accumulation; triggering an early warning if the trend accumulation is greater than a trend threshold dynamically determined based on the fluctuation entropy value; wherein, the trend threshold is negatively correlated with the fluctuation entropy value.

[0061] In this embodiment, in either the first or second discrimination mode, the warning triggering condition can be dynamically and adaptively adjusted based on the multi-dimensional characteristic vector, replacing the rigid, manually fixed threshold, thereby improving the accuracy of data warnings. Specifically, the warning triggering condition, i.e., the threshold for triggering the warning, is the instantaneous deviation threshold in the first discrimination mode and the trend threshold in the second discrimination mode, both of which are dynamically set by the fluctuation entropy value in the multi-dimensional characteristic vector.

[0062] As mentioned above, the first discrimination mode is based on instantaneous deviation. The system needs to respond quickly, so it first acquires the instantaneous deviation value between the real-time data of the data stream and the dynamic prediction baseline. This deviation is then compared with the instantaneous deviation threshold determined by the fluctuation entropy value. For data streams with high inherent noise and drastic fluctuations (high fluctuation entropy value), the system automatically raises the instantaneous alarm threshold. That is, only a larger instantaneous anomaly will trigger an alarm, effectively avoiding frequent false alarms caused by inherent data noise or normal large fluctuations. Conversely, for very stable data streams (low fluctuation entropy value), the system sets a lower instantaneous alarm threshold, maintaining high sensitivity to even minor sudden anomalies. Therefore, the instantaneous deviation threshold is positively correlated with the fluctuation entropy value.

[0063] Similarly, the second discrimination mode focuses on capturing slowly developing, persistent trend anomalies. It first continuously monitors the magnitude of data deviations from its dynamic prediction baseline within a preset time window (e.g., the past 30 minutes, 24 hours), integrating or accumulating these deviations to obtain a trend accumulation. This is the complete opposite of the first discrimination mode. For data streams with high volatility entropy values, because their normal fluctuations are already large, the system lowers the trend alarm threshold, making the system more sensitive to persistent anomalies. For stable data with low volatility entropy values, any small, persistent deviation could be a sign of a significant problem; therefore, the system raises the trend threshold, thus setting the trend threshold negatively correlated with the volatility entropy value.

[0064] Optionally, the indicators of dimensional characteristics also include the spatiotemporal index, which characterizes the temporal and spatial characteristics of the data, and the timeliness index, which characterizes the maximum time to process data after an anomaly.

[0065] After dynamically calculating the multi-dimensional characteristic vector corresponding to each data stream, the method also includes: determining the monitoring parameters corresponding to each data stream based on the multi-dimensional characteristic vector corresponding to each data stream.

[0066] In this embodiment, the spatiotemporal index specifically characterizes the temporal and spatial characteristics of data, that is, it marks the spatiotemporal characteristics of the data stream. For example, historical sequence data such as device temperature logs, which are purely composed of timestamps, has a high temporal and low spatial index [1,0]. UAV inspection image data or ground sensor grid data carrying latitude and longitude information have both a high temporal index and a high spatial index [1,1]. The timeliness index identifies the maximum waiting time from the occurrence of the data anomaly to the point where action must be taken. The maximum waiting time can usually be determined from business scenario presets, historical handling records, or textbooks in the field. For example, a maximum waiting time of less than 10 minutes is considered high timeliness, and a maximum waiting time of more than 24 hours is considered low timeliness. Monitoring parameters refer to the control parameters of sensors and other devices that collect data from each data stream.

[0067] It should be noted that the multi-dimensional characteristic vector proposed in this embodiment includes spatiotemporal index, timeliness index, dynamic index, fluctuation entropy value, and correlation index. Based on these five different dimensions, the data can be divided into 32 types. In practical applications, not all combinations are meaningful; approximately 16 common and effective classifications exist. Different monitoring parameters can be set for each type. For example, forest fire prevention drone smoke concentration data exhibits high spatiotemporal index, extremely high timeliness index, high dynamic index, high fluctuation entropy value, and high correlation index, thus classifying it as a "high spatiotemporal-high timeliness-high dynamic-medium-high fluctuation-high correlation" type. The monitoring parameters are set to a detection frequency of 1 second / time, high warning sensitivity, low filtering threshold, and automatic linkage with wind speed and temperature data. When the wind speed is ≥5m / s, the sensitivity automatically increases by 20%. This ensures timely response to critical warnings while also improving the overall level of intelligence and automation.

[0068] Optionally, the metrics for dimensional characteristics may also include a correlation index, which characterizes whether the current data stream is correlated with other data streams;

[0069] Monitoring and early warning according to the discrimination mode determined for each data stream also includes: when the current data stream is executing the second discrimination mode, increasing the monitoring parameters of other data streams that are related to the current data stream, so as to increase the monitoring frequency of other related data streams.

[0070] In this embodiment, the dimensional characteristic indicators also include a correlation index, used to quantify the strength of the intrinsic connection between the current data stream and one or more other data streams in the platform. Such correlations are complex in practical application scenarios. In this embodiment, the correlations include physical causal relationships, such as soil moisture affecting vegetation growth, and logical dependencies, such as the state of upstream equipment affecting downstream equipment in a production line. This embodiment limits the monitoring parameters of other data streams correlated with the current data stream to the second discrimination mode. This is because in the second discrimination mode, the system detects a potential but not yet critical, persistent abnormal trend. For example, soil moisture in area A begins to slowly decrease, and equipment temperature shows a slight but continuous upward trend. Therefore, in this case, the monitoring parameters of correlated data, such as the NDVI index of vegetation in area A, the temperature of irrigation or other monitoring equipment in area A, and the surface temperature of area A, can be adjusted to improve the early warning capability.

[0071] Optionally, monitoring and early warning according to the discrimination pattern determined for each data stream also includes:

[0072] When multiple alerts are triggered simultaneously, a basic priority score is calculated based on the multi-dimensional characteristic vector corresponding to the data stream that triggered the alert; the priority of the multiple data streams that triggered the alert is determined based on the priority score, the business scenario score of the data stream that triggered the alert, and the real-time impact factor; and the data streams that triggered the alert are sorted and displayed according to their priority.

[0073] In this embodiment, considering that the IoT platform may simultaneously receive multiple abnormal alarms from different data sources during actual operation—for example, a soil moisture warning for region A and an equipment failure warning for irrigation pump station B may occur simultaneously—related technologies simply present these alarms in chronological or random order. This prevents users from quickly focusing on the most urgent and impactful issues, thus delaying critical handling and reducing overall operational efficiency. Therefore, this embodiment proposes a dynamic priority ranking mechanism based on data characteristics and real-time context to ensure that alarm information is presented intelligently and systematically. Specifically, each dimension is pre-set with a corresponding weight coefficient. Then, based on the weight coefficients of different dimensions and the multi-dimensional characteristic vectors corresponding to the data streams triggering the alarms, a basic priority score is calculated. This is then combined with the business scenario scores and real-time impact factors of the data streams triggering the alarms to determine the priority of multiple data streams triggering alarms. Finally, these data streams are sorted and displayed according to their priority.

[0074] Because real-world applications involve a wide variety of data types and allow for flexible data type settings, only a few simple examples are provided below. For instance, the base weight for "Dynamic-High Timeliness-Strong Correlation" data is 0.4, while the base weight for "Static-Low Timeliness-Weak Correlation" data is 0.1. Next, a weight of 0.3 is added based on the user-defined business importance. Finally, real-time impact factors, including the scope of influence, duration, and degree of anomaly in the associated data, are incorporated, with a weight of 0.3. Priority is then calculated through weighted averages. Specifically, in industrial scenarios, equipment temperature is a Dynamic-High Timeliness-Medium Correlation data type with a base weight of 0.35, while environmental humidity is a Static-Medium Timeliness-Weak Correlation data type with a base weight of 0.15. If both the equipment temperature warning and the ambient humidity warning have a business importance weight of 0.3, and the equipment temperature warning affects 10 devices for 10 minutes (real-time impact factor determined to be 0.25), and the ambient humidity warning affects a single area for 2 minutes (real-time impact factor determined to be 0.05), then the equipment temperature warning score = 0.35 + 0.3 + 0.25 = 0.9, and the ambient humidity warning score = 0.15 + 0.3 + 0.05 = 0.5. The system automatically prioritizes the equipment temperature warning to ensure that core data anomalies are handled first. In this case, both the real-time impact factor and the basic weight can be manually set based on historical experience, so we will not elaborate further.

[0075] Optionally, after monitoring and issuing early warnings according to the discrimination mode determined for each data stream, the method further includes: obtaining the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback of the data stream that triggers the early warning; and optimizing the monitoring parameters of each data stream based on the gradient boosting tree optimization model, using the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback as input.

[0076] In this embodiment, as Figure 2 As shown in S201-S205, after determining the multi-dimensional characteristic vector, classifying the data type through the multi-dimensional characteristic vector, determining the monitoring parameters, and performing monitoring and early warning according to the monitoring parameters, the process also includes the step of optimizing the monitoring parameters based on the gradient boosting tree optimization model.

[0077] Specifically, the system first performs closed-loop data collection. It automatically collects data from the entire process, including alert trigger data (such as the trigger value, time, environmental conditions, and data characteristic classification), handling process data (such as handling steps, execution time, and resource input), handling effect data (such as alert cancellation time, whether the anomaly was resolved, and the amount of loss), and user feedback data (such as user-marked misjudgments / missed judgments and rule optimization suggestions). All data is stored in a rule optimization database according to a unified standard to ensure data integrity and availability. Then, a rule optimization model based on Gradient Boosting Tree (XGBoost) is constructed and trained periodically (e.g., daily at midnight) on the closed-loop data. The model deeply integrates data characteristic classification features.

[0078] Feature engineering: Extracting key features from the data, including early warning level, data characteristic classification such as dynamic / static, high / low timeliness, trigger frequency, handling effect score, environmental parameters, etc.

[0079] Model training: With the goals of improving early warning accuracy (false positive rate, false negative rate) and handling efficiency (handling time, resource consumption), the model is trained to learn the optimal rule parameters for different data characteristic classifications and different scenarios (such as the optimal detection frequency for "dynamic-high timeliness" data).

[0080] Model evaluation: Five-fold cross-validation is used to evaluate model performance and ensure the reliability of optimization suggestions. For example, the threshold adjustment scheme predicted by the model must improve the early warning accuracy by ≥5% to be considered effective.

[0081] After training, rule optimization suggestions are automatically generated, including threshold range adjustment, detection frequency optimization, early warning level adaptation, and priority weight adjustment. These optimization suggestions are deeply integrated with data characteristic classification. This achieves a complete optimization loop, ensuring the stable implementation of the overall early warning strategy.

[0082] Furthermore, as Figure 1 and Figure 2 The specific implementation of the method shown in this embodiment provides an intelligent early warning and discrimination device based on data characteristics, such as... Figure 3 As shown, the device includes:

[0083] The acquisition unit 301 is configured to acquire IoT monitoring data streams in real time and dynamically calculate the multi-dimensional characteristic vector corresponding to each data stream based on the historical sequence and real-time status of each data stream; the indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data.

[0084] The generation unit 302 is configured to generate a dynamic prediction baseline for each data stream based on the historical sequence of each data stream and the corresponding multi-dimensional characteristic vector, using a pre-trained time series prediction model; the dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period.

[0085] The comparison unit 303 is configured to compare the real-time data of each data stream with the corresponding dynamic prediction baseline to obtain a real-time comparison result;

[0086] The determining unit 304 is configured to determine a discrimination mode based on the real-time comparison result and the dynamic index in the multi-dimensional characteristic vector; wherein, the discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation, and the warning triggering conditions in the first discrimination mode and the second discrimination mode are determined based on the fluctuation entropy value;

[0087] Processing unit 305 is configured to monitor and issue warnings according to a discrimination pattern determined for each data stream.

[0088] In a specific application scenario, the determining unit 304 is further configured to: determine a deviation factor based on the deviation magnitude between the real-time data and the corresponding dynamic prediction baseline in the real-time comparison result; calculate an adjustment factor based on the dynamic index; calculate a mode discrimination threshold based on the deviation factor and the adjustment factor; and determine the first discrimination mode if the dynamic index is greater than the mode discrimination threshold, otherwise determine the second discrimination mode.

[0089] In a specific application scenario, the processing unit 305 is further configured to, when executing the first discrimination mode, calculate the instantaneous deviation value between the real-time data of the data stream and the dynamic prediction baseline; if the instantaneous deviation value is greater than an instantaneous deviation threshold dynamically determined based on the fluctuation entropy value, then trigger an early warning; wherein the instantaneous deviation threshold is positively correlated with the fluctuation entropy value; when executing the second discrimination mode, perform integral calculation on the deviation of the real-time data from the stated magnitude within a preset time window to obtain a trend accumulation amount; if the trend accumulation amount is greater than a trend threshold dynamically determined based on the fluctuation entropy value, then trigger an early warning; wherein the trend threshold is negatively correlated with the fluctuation entropy value.

[0090] In specific application scenarios, the acquisition unit 301 is further configured to determine the monitoring parameters corresponding to each data stream based on the multi-dimensional characteristic vector corresponding to each data stream.

[0091] In specific application scenarios, the processing unit 305 is further configured to increase the monitoring parameters of other data streams that are related to the current data stream when the second discrimination mode is executed in the current data stream, so as to increase the monitoring frequency of other related data streams.

[0092] In specific application scenarios, the processing unit 305 is further configured to, when multiple warnings are triggered simultaneously, calculate a basic priority score based on the multi-dimensional characteristic vector corresponding to the data stream that triggered the warning; determine the priority of the multiple data streams that triggered the warning based on the priority score, the business scenario score of the data stream that triggered the warning, and the real-time impact factor; and sort and display the data streams that triggered the warning according to the priority.

[0093] In specific application scenarios, the processing unit 305 is further configured to acquire the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback of the data stream that triggers the early warning; and optimize the monitoring parameters of each data stream based on the gradient boosting tree optimization model, using the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback as input.

[0094] It should be noted that other corresponding descriptions of the functional units involved in the intelligent early warning and discrimination device based on data characteristics provided in this embodiment can be found in [reference]. Figure 1 and Figure 2 The corresponding descriptions in [the document] will not be repeated here.

[0095] Based on the above, Figure 1 and Figure 2 Accordingly, this embodiment also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-described method. Figure 1 and Figure 2 The method shown.

[0096] Based on this understanding, the technical solution of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as CD-ROM, USB flash drive, mobile hard drive, etc.) and includes several instructions to cause a computer device (such as personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of this application.

[0097] Based on the above, Figure 1 and Figure 2 The method shown, and Figure 3 To achieve the above objectives, this application also provides an electronic device, which can be configured on a computer side, etc. The device includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to achieve the above-described objectives. Figure 1 and Figure 2 The method shown.

[0098] Optionally, the aforementioned physical devices may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0099] Those skilled in the art will understand that the physical device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0100] The storage medium may also include an operating system and a network communication module. The operating system is a program that manages the hardware and software resources of the aforementioned physical device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used to enable communication between the various components within the storage medium, as well as communication with other hardware and software in the information processing physical device.

[0101] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented using software plus necessary general-purpose hardware platforms, or it can be implemented in hardware. By applying the solution of this embodiment, compared with related technologies, it first reflects the characteristics of different data through multi-dimensional characteristic vectors, then generates a dynamic prediction baseline corresponding to each data stream through a time-series prediction model, and determines the early warning discrimination mode of the data based on the real-time comparison results of the data and the dynamic prediction baseline, thereby enabling the application of different early warning discrimination strategies for different types of data. Furthermore, the early warning triggering conditions in different discrimination modes can be dynamically and adaptively adjusted according to the multi-dimensional characteristic vectors, replacing rigid manually fixed thresholds, thereby improving the accuracy of data early warning.

[0102] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0103] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0104] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for intelligent early warning and discrimination based on data characteristics, characterized in that, include: The system acquires IoT monitoring data streams in real time and dynamically calculates a multi-dimensional characteristic vector for each data stream based on its historical sequence and real-time status. The indicators of the dimensional characteristics include at least a dynamic index that characterizes the rate of data change and a fluctuation entropy value that characterizes the randomness of the data. For each data stream, based on the historical sequence of each data stream and the corresponding multi-dimensional characteristic vector, a dynamic prediction baseline is generated for each data stream through a pre-trained time series prediction model; the dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period. The real-time data of each data stream is compared with the corresponding dynamic prediction baseline to obtain the real-time comparison result; Based on the real-time comparison results and the dynamic index in the multi-dimensional characteristic vector, a discrimination mode is determined; wherein, the discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation, and the warning triggering conditions in the first discrimination mode and the second discrimination mode are determined based on the fluctuation entropy value; Monitoring and early warning are performed based on the discrimination pattern determined for each data stream.

2. The method according to claim 1, characterized in that, The determination of the discrimination mode based on the real-time comparison results and the dynamic index in the multi-dimensional feature vector includes: Based on the deviation magnitude between the real-time data and the corresponding dynamic prediction baseline in the real-time comparison results, the deviation factor is determined; Calculate the adjustment factor based on the dynamic index; Calculate the pattern discrimination threshold based on the deviation factor and the adjustment factor; If the dynamic index is greater than the pattern discrimination threshold, it is determined to be the first discrimination mode; otherwise, it is the second discrimination mode.

3. The method according to claim 1, characterized in that, The monitoring and early warning based on the discrimination pattern determined for each data stream includes: When the first discrimination mode is executed, the instantaneous deviation between the real-time data of the data stream and the dynamic prediction baseline is calculated; If the instantaneous deviation value is greater than the instantaneous deviation threshold dynamically determined based on the fluctuation entropy value, an early warning is triggered; wherein the instantaneous deviation threshold is positively correlated with the fluctuation entropy value; When the second discrimination mode is executed, the deviation of real-time data from the stated magnitude is integrally calculated within a preset time window to obtain the cumulative trend amount; If the cumulative trend amount exceeds a trend threshold dynamically determined based on the fluctuation entropy value, an early warning is triggered; wherein the trend threshold is negatively correlated with the fluctuation entropy value.

4. The method according to claim 1, characterized in that, The indicators of the dimensional characteristics also include the spatiotemporal index, which characterizes the temporal and spatial characteristics of the data, and the timeliness index, which characterizes the maximum time to process data after an anomaly. After dynamically calculating the multi-dimensional feature vector corresponding to each data stream, the method further includes: Based on the multi-dimensional characteristic vector corresponding to each data stream, the monitoring parameters corresponding to each data stream are determined.

5. The method according to claim 4, characterized in that, The indicators of the dimensional characteristics also include a correlation index, which characterizes whether the current data stream is related to other data streams; The monitoring and early warning based on the discrimination pattern determined for each data stream also includes: When the second discrimination mode is executed in the current data stream, the monitoring parameters of other data streams that are related to the current data stream are increased to increase the monitoring frequency of other related data streams.

6. The method according to claim 1, characterized in that, The monitoring and early warning based on the discrimination pattern determined for each data stream also includes: When multiple warnings are triggered simultaneously, a basic priority score is calculated based on the multi-dimensional characteristic vector corresponding to the data stream that triggered the warning. Based on the priority score, the business scenario score of the data stream that triggered the warning, and the real-time impact factor, the priority of multiple data streams that triggered the warning is determined; The data streams that trigger warnings are sorted and displayed according to the stated priority.

7. The method according to claim 1, characterized in that, After monitoring and issuing early warnings according to the discrimination pattern determined for each data stream, the method further includes: Obtain the characteristic vector, discrimination mode, manual discrimination result, and manual handling feedback of the data stream that triggered the warning; Based on the gradient boosting tree optimization model, the monitoring parameters of each data stream are optimized by taking the feature vector, discrimination mode, manual discrimination results and manual processing feedback as inputs.

8. An intelligent early warning and discrimination device based on data characteristics, characterized in that, include: The acquisition unit is configured to acquire IoT monitoring data streams in real time and dynamically calculate the multi-dimensional characteristic vector corresponding to each data stream based on the historical sequence and real-time status of each data stream; the indicators of the dimensional characteristics include at least a dynamic index characterizing the rate of data change and a fluctuation entropy value characterizing the randomness of the data. The generation unit is configured to generate a dynamic prediction baseline for each data stream based on the historical sequence of each data stream and the corresponding multi-dimensional feature vector, using a pre-trained time series prediction model; the dynamic prediction baseline is used to characterize the expected fluctuation range of the data stream in the next time period. The comparison unit is configured to compare the real-time data of each data stream with the corresponding dynamic prediction baseline to obtain a real-time comparison result; The determining unit is configured to determine a discrimination mode based on the real-time comparison results and the dynamic index in the multi-dimensional characteristic vector; wherein, the discrimination mode includes a first discrimination mode based on instantaneous deviation and a second discrimination mode based on trend accumulation, and the warning triggering conditions in the first discrimination mode and the second discrimination mode are determined based on the fluctuation entropy value; The processing unit is configured to monitor and issue early warnings according to the discrimination pattern determined for each data stream.

9. An electronic device, comprising a storage medium, a processor, and a computer program stored on the storage medium and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images