Intelligent supervision early warning model optimization method based on streaming data processing

Abstract

The invention relates to the field of data processing and artificial intelligence, in particular to an intelligent supervision early warning model optimization method based on streaming data processing, which comprises the following steps of: capturing streaming data at intervals, extracting basic attribute characteristics of the streaming data, calculating a stable representation value of the streaming data based on characteristic difference, and setting a virtual label for a supervision period. Based on the label, calling a preset supervision model only for a part of image frames in a stable period of passing verification to output a judgment result, and endowing the result to the remaining frames; and in the hopping time domain section, calling a preset supervision model for all the image frames to output a judgment result, and endowing the result to the remaining frames. Lightweight assessment of streaming data stability is realized, a processing strategy can be adjusted according to data fluctuation characteristics, result replacement analysis is performed on a stable period and a non-hopping time domain while the accuracy of hopping time domain analysis is guaranteed, the number of image frames needing to call a preset supervision model is reduced, and the accuracy of the streaming data stability is improved. And the calculation efficiency and the resource utilization rate of the model are optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing and artificial intelligence, and in particular to an intelligent supervision and early warning model optimization method based on stream data processing. BACKGROUND

[0002] With the rapid development of Internet of Things, edge computing and artificial intelligence technology, supervision systems based on stream data processing are widely used in industrial supervision, urban security, facility operation and maintenance and other fields. Such systems usually continuously collect stream data such as video and sensor readings, and use pre-set supervision models to analyze and warn in real time.

[0003] For example, Chinese Patent Publication No. CN107240957B, a method for power grid monitoring and early warning based on big data stream computing, the steps are: 1) real-time access to monitoring data of each channel and each real-time monitoring system; 2) mapping the accessed data with the monitoring and early warning instances; 3) updating the data set of the monitoring instance cache; 4) early warning calculation of the monitoring instance; 5) cleaning the early warning instance cache data; 6) determining the early warning level; 7) pushing the early warning information. The early warning method for real-time monitoring of different scenarios based on the power grid of the present application solves the problems of large amount of monitoring information data, inconsistent monitoring time of multiple channels, and inconsistent sampling of monitoring terminals, and ensures the accuracy of the early warning information.

[0004] However, the prior art still has the following problems: 1. In the prior art, the stability characteristics of stream data in the data flow dimension and the time sequence dimension are not considered, the stable state of the data stream cannot be evaluated in a lightweight manner without decoding the data content, and the processing strategy cannot be dynamically adjusted according to the characteristics of the data stream itself; 2. In the prior art, the characteristics of the stream data itself, i.e. the fluctuation of data volume and flow rate, are not considered, and a differentiated processing mechanism cannot be implemented according to this fluctuation characteristic, the same processing intensity is used for stable periods with small data volume and sudden periods with large data volume, resulting in waste of computing resources or insufficient computing resources in critical periods. SUMMARY

[0005] Therefore, the present application provides an intelligent supervision and early warning model optimization method based on stream data processing, to overcome the problems in the prior art that the stability characteristics of stream data in the data flow dimension and the time sequence dimension are not considered, the stable state of the data stream cannot be evaluated in a lightweight manner without decoding the data content, the characteristics of the stream data itself, i.e. the fluctuation of data volume and flow rate, are not considered, and a differentiated processing mechanism cannot be implemented according to this fluctuation characteristic, resulting in waste of computing resources or insufficient computing resources in critical periods.

[0006] To achieve the above object, the application provides an intelligent supervision and early warning model optimization method based on streaming data processing, comprising: Every predetermined supervision period, the streaming data is captured at intervals within the supervision period to read the basic attribute features of the streaming data at the capture time; The streaming data stability representation value is calculated based on the differences between the basic attribute features to set a virtual label for the streaming data within the supervision period; The streaming data is processed based on the virtual label, comprising, The streaming data is partially decoded to obtain a plurality of image frames, the image attribute features in the image frames are extracted, the stability of the supervision period is verified based on the differences of the image attribute features, and the result replacement analysis is performed for the verification; Or, the streaming data is globally decoded to obtain a plurality of continuous image frames, the image attribute features in the image frames are extracted, the jump time domain segment is determined based on the changes of the image attribute features, the judgment result for all the image frames is output based on the preset supervision model for the image frames in the jump time domain segment, and the result replacement analysis is performed for the non-jump time domain segment; Wherein, the result replacement analysis comprises: only obtaining the image frames at part of the time in the supervision period or / and the non-jump time domain segment, outputting the judgment result for each of the image frames based on the preset supervision model, and assigning the judgment result to the corresponding remaining image frames in the supervision period or / and the non-jump time domain segment.

[0007] Further, the streaming data is captured at intervals within the supervision period to read the basic attribute features of the streaming data at the capture time, comprising, Within the supervision period, the streaming data packets are extracted at a preset time interval to obtain the number of data packets, the average size of data packets, the average arrival interval of data packets, and the arrival interval jitter of data packets.

[0008] Further, the streaming data stability representation value is calculated based on the differences between the basic attribute features, comprising, The flow dimension fluctuation index is calculated based on the number of data packets and the average size of data packets; The time domain dimension fluctuation index is calculated based on the average arrival interval of data packets and the arrival interval jitter of data packets; The weighted sum result of the flow dimension fluctuation index and the time domain dimension fluctuation index is used to determine the streaming data stability representation value.

[0009] Further, the virtual label is set for the streaming data within the supervision period, comprising, If the streaming data stability representation value corresponding to the streaming data within the supervision period is greater than a preset stability threshold, a stable state label is set for the streaming data; if a stable data feature value corresponding to the stream data in the supervision period is less than or equal to a preset stable threshold, setting an unstable state label for the stream data; The virtual label includes a stable state label and an unstable state label.

[0010] Further, the partial decoding of the stream data includes, If the virtual label of the stream data is the stable state label, decoding partial data segments of all stream data corresponding to the supervision period according to a preset rule to obtain a plurality of image frames.

[0011] Further, verifying the stability of the supervision period based on the difference of the image attribute features includes, extracting the average brightness value and the color histogram variance value of each of the decoded image frames; determining the image attribute features based on the average brightness value and the color histogram variance value; verifying the stability of the supervision period based on the difference of the image attribute features.

[0012] Further, verifying the stability of the supervision period based on the difference of the image attribute features includes, analyzing the change of the image attribute features corresponding to each of the image frames in a time sequence order; if the change of the image attribute features corresponding to each of the image frames meets a preset stable condition, confirming that the supervision period stability verification is passed; if the change of the image attribute features corresponding to each of the image frames does not meet the preset stable condition, confirming that the supervision period stability verification is not passed.

[0013] Further, the preset stable condition includes, satisfying a first determination criterion and a second determination criterion at the same time; The first determination criterion includes that the absolute difference of the image attribute features between any two adjacent image frames in the supervision period is less than a first set threshold; The second determination criterion includes that the standard deviation of the image attribute features in a sliding analysis window formed by a preset number of continuous image frames in the supervision period is less than a second set threshold.

[0014] Further, determining the jump time domain segment based on the change of the image attribute features includes, calculating the change amount of the image attribute features between the continuous image frames, and marking the corresponding time as a jump starting point when the change amount exceeds a jump threshold; Starting from the jump starting point, the feature change amount of the subsequent image frames is continuously monitored, and when the feature change amount falls below the stable threshold and remains for a preset time length, the corresponding time is marked as the jump ending point; The period between the jump starting point and the jump ending point is determined as the jump time domain segment.

[0015] Further, the intelligent supervision and early warning model optimization method based on streaming data processing further comprises, Record the determination results of all image frames with abnormalities and send them to the demand side for subsequent analysis.

[0016] Compared with the prior art, the present application captures streaming data at intervals and extracts its basic attribute features, calculates the stable representation value of streaming data based on feature difference, and then sets a virtual label for the supervision period. Based on the label, in the stable period verified, only part of the image frames call the preset supervision model to output the determination result, and the result is assigned to the remaining frames; in the jump time domain segment, all image frames call the preset supervision model to output the determination result, and the result is assigned to the remaining frames. The present application realizes lightweight evaluation of the stability of streaming data, can adjust the processing strategy according to the data fluctuation characteristics, ensures the accuracy of analysis in the jump time domain segment, and reduces the number of image frames that need to call the preset supervision model by implementing result replacement analysis in the stable period and non-jump time domain segment, thereby optimizing the calculation efficiency and resource utilization of the model.

[0017] Especially, the present application considers the virtual label prediction mechanism. In actual situations, the streaming data flow and timing fluctuate dramatically, but the prior art cannot predict its stable state without decoding the content. The virtual label mechanism realizes lightweight pre-evaluation of the stable state of data flow based on basic attribute features, providing a key decision basis for subsequent differentiated processing.

[0018] Especially, the present application considers the result replacement analysis mechanism. In actual situations, the data changes slowly in the stable period, and uniform processing causes a lot of waste of computing resources. The result replacement analysis mechanism realizes the reduction of computing load in the stable period under the premise of ensuring the continuity of supervision by assigning the determination results of part of the frames to the remaining frames.

[0019] Especially, the present application considers the identification of jump time domain and full decoding analysis. In actual situations, there is a risk that key mutation events are ignored or missed due to insufficient computing resources. This mechanism ensures the integrity and accuracy of key period supervision by performing supervision model analysis on all image frames in the identified jump time domain segment, effectively avoiding the loss of important early warning information. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1A flow data processing-based intelligent supervision early warning model optimization method according to an embodiment of the application; Figure 2 A logic block diagram of setting a virtual label according to an embodiment of the application; Figure 3 A logic block diagram of decoding flow data of a stable state label according to an embodiment of the application; Figure 4 A logic block diagram of performing difference verification according to an embodiment of the application. DETAILED DESCRIPTION

[0021] In order to make the objects and advantages of the present application clearer, the present application will be further described below in conjunction with embodiments. It should be understood that the specific embodiments described herein merely serve the purpose of explaining the present application and are not intended to limit the present application.

[0022] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. It should be understood by those skilled in the art that the embodiments merely serve the purpose of explaining the technical principles of the present application and are not intended to limit the protection scope of the present application.

[0023] It should be noted that, in the description of the present application, the terms indicating the direction or positional relationship such as "upper", "lower", "left", "right", "inner", "outer" and the like are based on the direction or positional relationship shown in the drawings, which is merely for the purpose of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application.

[0024] In addition, it should also be noted that, in the description of the present application, unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connecting" should be understood in a broad sense, for example, can be fixed connection, can also be detachable connection, or integral connection; can be mechanical connection, can also be electrical connection; can be direct connection, can also be indirect connection through an intermediate medium, can be the internal communication of two elements. Those skilled in the art can understand the specific meaning of the above terms in the present application according to the specific circumstances.

[0025] Please refer to Figure 1 shown, Figure 1 A flow data processing-based intelligent supervision early warning model optimization method according to an embodiment of the application, the flow data processing-based intelligent supervision early warning model optimization method of the present application comprises: Step S1, every predetermined supervision period, flow data is captured at intervals in the supervision period to read the basic attribute features of the flow data at the capture time; Step S2, calculating a stable representation value of the streaming data based on the differences between the basic attribute features, to set a virtual label for the streaming data in the supervision period; Step S3, processing the streaming data based on the virtual label, including, performing partial decoding processing on the streaming data to obtain a plurality of image frames, extracting image attribute features in the image frames, verifying the stability of the supervision period based on the differences between the image attribute features, and performing result substitution analysis for the verification passing; Or, performing global decoding on the streaming data to obtain a plurality of continuous image frames, extracting image attribute features in the image frames, determining a jump time domain segment based on the changes of the image attribute features, outputting a judgment result for all the image frames based on a preset supervision model for the image frames in the jump time domain segment, and performing result substitution analysis for the non-jump time domain segment; Wherein, the result substitution analysis includes, only obtaining image frames at part of time in the supervision period or / and non-jump time domain segment, outputting a judgment result for each of the image frames based on a preset supervision model, and assigning the judgment result to the corresponding remaining image frames in the supervision period or / and non-jump time domain segment.

[0026] Specifically, the length of the predetermined supervision period needs to balance between real-time and processing overhead. If the period is too long, it may cause a delay in responding to changes in data flow state, and cannot capture sudden events in time. If the period is too short, it will introduce excessive computing overhead due to frequent execution of data capture and stability evaluation. The length of the predetermined supervision period in the embodiment of the present application is in the range of [30 seconds, 60 seconds].

[0027] Specifically, the streaming data is captured at intervals within the supervision period to read the basic attribute features of the streaming data at the capture time, including, Within the supervision period, the data packets are extracted at a preset time interval to obtain the number of data packets, the average size of data packets, the average arrival interval of data packets, and the arrival interval jitter of data packets.

[0028] Specifically, the size of the preset time interval is related to the predetermined supervision period. In the embodiment of the present application, the interval is set to one sixth of the length of the current "predetermined supervision period". For example, when the supervision period is 30 seconds, the sampling interval is 5 seconds, which ensures that enough data points can be obtained within the period to calculate the basic attribute features, while avoiding excessive performance overhead due to too frequent sampling.

[0029] Specifically, calculating a stable representation value of the streaming data based on the differences between the basic attribute features includes, calculating a flow dimension fluctuation index based on the number of data packets and the average size of data packets; The time-domain dimension fluctuation index is calculated based on a data packet average arrival interval and a data packet arrival interval jitter; A flow data stability characteristic value is determined according to a weighted sum result of the flow dimension fluctuation index and the time-domain dimension fluctuation index.

[0030] Specifically, the calculation of the flow dimension fluctuation index is based on a weighted sum of a data packet arrival number change rate and a data packet average size change rate; and the calculation of the time-domain dimension fluctuation index is based on a weighted sum of a data packet average arrival interval change rate and a data packet arrival interval jitter change rate.

[0031] The data packet arrival number change rate is determined by calculating an absolute value of a data packet arrival number difference between a current monitoring period and a previous monitoring period, and then dividing the absolute value by the data packet arrival number of the previous monitoring period.

[0032] The data packet average size change rate is determined by calculating an absolute value of a data packet average size difference between a current monitoring period and a previous monitoring period, and then dividing the absolute value by the data packet average size of the previous monitoring period.

[0033] The data packet average arrival interval change rate is determined by calculating an absolute value of a data packet average arrival interval difference between a current monitoring period and a previous monitoring period, and then dividing the absolute value by the data packet average arrival interval of the previous monitoring period.

[0034] The data packet arrival interval jitter change rate is determined by calculating an absolute value of a data packet arrival interval jitter difference between a current monitoring period and a previous monitoring period, and then dividing the absolute value by the data packet arrival interval jitter of the previous monitoring period.

[0035] Specifically, in the calculation of the flow dimension fluctuation index, different weighting coefficients are respectively assigned to the data packet arrival number change rate and the data packet average size change rate. The change of the data packet arrival number can more directly reflect the abnormal fluctuation of the flow level than the change of the data packet size. Therefore, the weighting coefficient of the data packet arrival number change rate is preferably set to 0.7, and the weighting coefficient of the data packet average size change rate is preferably set to 0.3, and the weighted sum of the two is the flow dimension fluctuation index.

[0036] Specifically, in calculating the time-domain dimension fluctuation index, different weighting coefficients are respectively assigned to the data packet average arrival interval change rate and the data packet arrival interval jitter change rate. Considering that the jitter of the data packet arrival interval is more sensitive to reveal the instability at the time sequence level than the average interval itself, the embodiment of the application preferably sets the weighting coefficient of the data packet average arrival interval change rate to 0.4, sets the weighting coefficient of the data packet arrival interval jitter change rate to 0.6, and obtains the time-domain dimension fluctuation index by weighting and summing the two.

[0037] Referring to Figure 2 shown in FIG. 1, Figure 2 a logic block diagram of setting a virtual label according to an embodiment of the application, setting a virtual label for the streaming data in the supervision period comprises, if the stable state label is set for the streaming data, decoding part of the data segments of the streaming data in the supervision period according to a preset rule to obtain a plurality of image frames. if the stable state label is set for the streaming data, decoding part of the data segments of the streaming data in the supervision period according to a preset rule to obtain a plurality of image frames. wherein the virtual label comprises the stable state label and the unstable state label.

[0038] Specifically, the preset stable threshold value can be determined by analyzing the historical streaming data in the normal operation state of the system. The specific method is to collect the streaming data of several supervision periods in a historical period, calculate the stable state label of each period, and obtain the average value of the stable state label. The average value of the stable state label is used as the stable threshold value.

[0039] Referring to Figure 3 shown in FIG. 1, Figure 3 a logic block diagram of decoding the streaming data with the stable state label according to an embodiment of the application, the partial decoding process of the streaming data comprises, if the stable state label is set for the streaming data, decoding part of the data segments of the streaming data in the supervision period according to a preset rule to obtain a plurality of image frames.

[0040] Specifically, the partial decoding process is to decode part of the data segments of the streaming data in the supervision period according to a preset rule to obtain a plurality of image frames. The preset rule adopts an equal interval sampling decoding strategy, and the specific implementation is to divide all the streaming data in the supervision period into a plurality of continuous data segments in time sequence, and cyclically extract according to the fixed mode of "decoding one data segment and skipping five data segments".

[0041] Specifically, the image attribute features in the image frames are extracted, and the stability of the supervision period is verified based on the differences in the image attribute features, including, For each of the image frames obtained after decoding, an average brightness value and a color histogram variance value are extracted respectively; The image attribute features are determined based on the average brightness value and the color histogram variance value; The stability of the supervision period is verified based on the differences in the image attribute features.

[0042] Specifically, the ratio of the average brightness value of the current image frame to the preset brightness standard value is calculated, and the ratio is determined as a brightness factor; the ratio of the color histogram variance value of the current image frame to the preset color variance standard value is calculated, and the ratio is determined as a color factor; the brightness factor and the color factor are weighted and summed according to a preset weight, and the result is the image attribute feature value of the current image frame.

[0043] Specifically, the brightness standard value is obtained by calculating the average of the average brightness values of a plurality of image frames in a historical normal scene, which provides a unified brightness reference for subsequent calculation of the brightness factor.

[0044] Specifically, the color variance standard value is obtained by calculating the average of the color histogram variance values of a plurality of image frames in a historical normal scene, which provides a stable color distribution reference for subsequent calculation of the color factor.

[0045] By converting the absolute feature value into a ratio factor relative to the standard reference and fusing it, this method can effectively eliminate the influence of the difference in absolute feature values under different environmental conditions, thereby more accurately representing the relative change characteristics of the image content and providing a reliable basis for subsequent verification of the stability of the supervision period.

[0046] Please refer to Figure 4 as shown, Figure 4 the logic block diagram for the difference verification of the invention embodiment, the stability of the supervision period is verified based on the differences in the image attribute features, including, The change of the image attribute features corresponding to each of the image frames is analyzed in time sequence; If the change of the image attribute features corresponding to each of the image frames meets the preset stable condition, it is confirmed that the supervision period stability verification is passed; If the change of the image attribute features corresponding to each of the image frames does not meet the preset stable condition, it is confirmed that the supervision period stability verification is not passed.

[0047] Specifically, the preset stable condition includes, satisfy the first determination criterion and the second determination criterion at the same time; The first determination criterion comprises that, in the supervision period, an absolute difference value of image attribute feature values between any two adjacent image frames needs to be less than a first set threshold value. The second determination criterion comprises that, in the supervision period, a standard deviation of image attribute feature values in a sliding analysis window composed of a preset number of continuous image frames needs to be less than a second set threshold value.

[0048] Specifically, the number of continuous image frames in the sliding analysis window ranges from 24 frames to 60 frames. In the embodiment of the present application, 25 frames are taken.

[0049] When and only when the image frame sequence in the supervision period simultaneously satisfies the first determination criterion and the second determination criterion, it is determined that the preset stability condition is met, and it is confirmed that the stability verification of the supervision period is passed; otherwise, it is determined that the verification is not passed. Meeting the preset stability condition represents that the image content in the supervision scene remains stable in the supervision period, and there is no significant picture change caused by factors such as sudden change of light, rapid movement of objects or camera shaking.

[0050] Specifically, the first set threshold value is set based on the statistical result of the difference value of image attribute feature values between continuous image frames in the historical normal scene. By calculating the average value of the difference value of feature values between adjacent frames in the historical data, the first set threshold value is set to 1.2 times the average value. This setting can effectively identify the abnormal jump of picture content, and can also avoid misjudgment caused by normal gradual change of environmental light or slow movement of objects.

[0051] Specifically, the second set threshold value is set based on the statistical result of the standard deviation of image attribute feature values of a preset number of continuous image frame sequences in the historical normal scene. By analyzing the distribution of the historical standard deviation of image attribute feature values in the sliding analysis window composed of a preset number of continuous image frames, the second set threshold value is set to the average value of the historical standard deviation.

[0052] It can be understood that the specific values of the above two threshold values can be adjusted according to different application scenarios. In the security supervision scene with high requirements for picture stability, a relatively small threshold value can be used; in the industrial visual detection scene with relatively stable environment, the threshold value requirement can be appropriately relaxed. In actual system deployment, a small-scale experimental verification or trial operation can be performed by those skilled in the art to determine the most suitable threshold parameter for the current scene.

[0053] Specifically, the result replacement analysis for verification passing comprises, obtaining decoded image frames in the supervision period, and inputting a preset supervision model; outputting a determination result for each decoded image frame based on the preset supervision model, and assigning the determination result to the corresponding remaining image frame in the supervision period.

[0054] Specifically, determining the jump time domain segment based on the change of the image attribute feature includes, calculating the change of the image attribute feature between consecutive image frames, and marking the corresponding time as a jump starting point when the change exceeds a jump threshold; starting from the jump starting point, continuously monitoring the feature change of the subsequent image frames, and marking the corresponding time as a jump ending point when the feature change falls below a stable threshold and remains for a preset time length; determining the time period between the jump starting point and the jump ending point as the jump time domain segment.

[0055] Specifically, the intelligent supervision and early warning model optimization method based on streaming data processing further includes, record all image frames with abnormal determination results and send them to the demand side for subsequent analysis.

[0056] Specifically, the preset supervision model is not limited, and it is used to identify abnormal targets in image frames. Those skilled in the art can pre-train an image processing model capable of identifying predetermined targets in images, or use existing open-source image processing models capable of identifying predetermined targets. This is prior art and will not be repeated. When an abnormal target is identified in an image frame, it can be determined that the image frame is abnormal.

[0057] It can be understood that all image frames with abnormal determination results are recorded and sent to the demand side for subsequent analysis. By structuring the identified abnormal information, it ensures that the early warning information is traceable, verifiable, and usable. The demand side can be a local storage server, a cloud analysis platform, a supervision center alarm system, or a third-party business platform, thereby supporting real-time alarm, manual review, deep analysis, or automatic linkage of different levels of business response. This design balances resource efficiency and supervision effectiveness while reducing normal data processing overhead, ensuring that key abnormal information enters higher-level decision-making and disposal processes.

[0058] So far, the technical solutions of the present application have been described in conjunction with the preferred embodiments shown in the drawings, but those skilled in the art will readily understand that the protection scope of the present application is obviously not limited to these specific embodiments. Those skilled in the art can make equivalent changes or replacements to related technical features without deviating from the principles of the present application, and the technical solutions after these changes or replacements will fall within the protection scope of the present application.

Claims

1. An intelligent supervision and early warning model optimization method based on stream data processing, characterized in that, The method comprises the following steps: Every predetermined regulatory period, interval capture streaming data to read the basic attribute features of the streaming data at the capture time; Based on the difference between each basic attribute feature, calculate the stable representation value of the streaming data to set a virtual label for the streaming data in the regulatory period; Based on the virtual label, process the streaming data, including, Partial decoding processing of the streaming data to obtain a plurality of image frames, extract image attribute features in the image frames, verify the stability of the regulatory period based on the difference of the image attribute features, and for the verification, perform result substitution analysis; Or, full decoding of the streaming data to obtain a plurality of continuous image frames, extract image attribute features in the image frames, determine the jump time domain segment based on the change of the image attribute features, output the judgment result for all the image frames based on the preset regulatory model for the image frames in the jump time domain segment, and perform result substitution analysis for the non-jump time domain segment; Wherein, the result substitution analysis includes, only obtaining image frames at part of the time in the regulatory period or / and non-jump time domain segment, outputting the judgment result for each of the image frames based on the preset regulatory model, and assigning the judgment result to the corresponding remaining image frames in the regulatory period or / and non-jump time domain segment. 2.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Interval capture streaming data in the regulatory period to read the basic attribute features of the streaming data at the capture time includes, In the regulatory period, extract the streaming data packet at a preset time interval, obtain the packet arrival number, packet average size, packet average arrival interval, and packet arrival interval jitter. 3.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Based on the difference between each basic attribute feature, calculate the stable representation value of the streaming data includes, Calculate the flow dimension fluctuation index based on the packet arrival number and the packet average size; Calculate the time domain dimension fluctuation index based on the packet average arrival interval and the packet arrival interval jitter; Determine the stable representation value of the streaming data according to the weighted sum result of the flow dimension fluctuation index and the time domain dimension fluctuation index.

4. The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Setting a virtual label for the streaming data in the regulatory period includes, If the stable representation value of the streaming data corresponding to the streaming data in the regulatory period is greater than the preset stable threshold, set a stable state label for the streaming data; If the stable representation value of the streaming data corresponding to the streaming data in the regulatory period is less than or equal to the preset stable threshold, set an unstable state label for the streaming data; Wherein, the virtual label includes a stable state label and an unstable state label. 5.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Partial decoding processing of the streaming data includes, If the virtual label of the streaming data is a stable state label, partial decoding processing of the streaming data is performed to obtain a plurality of image frames, extract image attribute features in the image frames, verify the stability of the regulatory period based on the difference of the image attribute features, and for the verification, perform result substitution analysis; If the virtual label of the streaming data is an unstable state label, the streaming data is globally decoded to obtain continuous image frames, image attribute features in the image frames are extracted, a jump time domain segment is determined based on changes in the image attribute features, a preset supervision model is used to output a judgment result for all the image frames based on the image frames in the jump time domain segment, and result replacement analysis is performed on a non-jump time domain segment. 6.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Extracting image attribute features in the image frames and verifying stability of the supervision period based on differences in the image attribute features include, For each of the image frames obtained after decoding, an average brightness value and a color histogram variance value are extracted; Determine image attribute features based on the average brightness value and the color histogram variance value; Verify stability of the supervision period based on differences in the image attribute features.

7. The intelligent supervision and early warning model optimization method based on stream data processing according to claim 6, characterized in that, Verifying stability of the supervision period based on differences in the image attribute features includes, Analyze changes in the image attribute features corresponding to each of the image frames in a time sequence order; If the change in the image attribute features corresponding to each of the image frames meets a preset stability condition, it is determined that the supervision period stability verification is passed; If the change in the image attribute features corresponding to each of the image frames does not meet the preset stability condition, it is determined that the supervision period stability verification is not passed. 8.The method of claim 7, wherein, The preset stability condition includes, Both the first judgment criterion and the second judgment criterion are met; The first judgment criterion includes that, within the supervision period, the absolute difference between the image attribute features of any two adjacent image frames needs to be less than a first set threshold value; The second judgment criterion includes that, within the supervision period, the standard deviation of the image attribute features in a sliding analysis window composed of a preset number of continuous image frames needs to be less than a second set threshold value. 9.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Determining the jump time domain segment based on changes in the image attribute features includes, Calculate the change amount of the image attribute features between continuous image frames. When the change amount exceeds a jump threshold value, mark the corresponding time as a jump start point; From the jump start point, continuously monitor the feature change amount of the subsequent image frames. When the feature change amount falls below a stable threshold value and remains below the stable threshold value for a preset length of time, mark the corresponding time as a jump end point; Determine the time period between the jump start point and the jump end point as the jump time domain segment. 10.The intelligent supervision and early warning model optimization method based on streaming data processing according to claim 1, characterized in that, Also includes, Record all image frames with abnormal judgment results and send them to the demand end for subsequent analysis.

Images