A video acquisition and storage method based on a smart energy unit

Abstract

This invention discloses a video acquisition and storage method based on a smart energy unit, comprising: receiving raw video streams sent by network video surveillance equipment and acquiring multi-dimensional power telemetry parameters in real time; calculating the comprehensive event severity and determining the event level corresponding to the comprehensive event severity; dynamically adjusting the video processing parameters of the raw video stream based on the comprehensive event severity and the operating resource budget; converting the format of the raw video stream according to the adjusted video processing parameters to obtain target streaming media data; segmenting the target streaming media data into video segments, calculating an importance score based on the event level corresponding to the video segment, and performing storage eviction management on the video segments according to the importance score. This invention achieves on-demand allocation of edge computing power and secure isolation from main business operations, improving the retention rate of high-value abnormal video recording evidence within limited storage space.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things video surveillance and power edge computing technology, and in particular to a video acquisition and storage method based on a smart energy unit. Background Technology

[0002] With the deepening of the construction of the power Internet of Things (IoT), edge data acquisition terminals have been widely deployed in power transmission and distribution nodes such as substations and distribution rooms. Introducing video monitoring systems into these nodes can combine telemetry data on the operating status of physical equipment with real-time visual images, providing multi-dimensional data support for the operation and maintenance of power equipment, fault classification, and accident tracing. Multimodal fusion sensing at the edge has significant technical value for improving the level of automated monitoring of the power grid, reducing on-site operation and maintenance costs, and increasing fault response rates.

[0003] Current video surveillance solutions in substations and similar scenarios typically employ an architecture where dedicated hard disk recorders and independent power monitoring terminals operate in parallel. These independently operating monitoring systems often continuously acquire and cyclically overwrite video streams according to a pre-set fixed schedule or constant video quality parameters. In actual operation, when sudden electrical anomalies occur, such as equipment overheating or short circuits, the fixed-quality video acquisition often cannot provide sufficient high frame rate or high-definition video evidence instantly. Furthermore, existing storage management strategies are mostly based on simple time-series first-in-first-out (FIFO) elimination. When the storage space of edge devices is limited, recordings of electrical anomalies with extremely high post-analysis value are easily overwritten by a large amount of subsequent low-value routine inspection footage, leading to the loss of critical maintenance evidence. In addition, when high-intensity video stream parsing and low-level communication tasks are simultaneously overlaid on single-board edge devices, the high-load media encoding and decoding process often disorderly preempts microprocessor resources, causing delays in core control commands or even system crashes.

[0004] In summary, existing multimodal monitoring systems struggle to simultaneously acquire high-quality images of critical faults and retain long-term evidence when faced with complex field conditions and limited edge computing and storage resources. Therefore, it is necessary to research a method that can improve the overall performance of data processing and storage in concurrent edge multitasking scenarios, ensuring reliable retention of core data and mitigating underlying computing power conflicts. Summary of the Invention

[0005] Purpose of the invention: To provide a video acquisition and storage method based on a smart energy unit, in order to solve the above-mentioned problems existing in the prior art.

[0006] Technical solution: A video acquisition and storage method based on a smart energy unit, comprising:

[0007] It receives raw video streams sent by network video surveillance equipment and collects multi-dimensional power telemetry parameters in real time;

[0008] The severity of a comprehensive event is calculated based on multidimensional power telemetry parameters, and the event level corresponding to the severity of the comprehensive event is determined.

[0009] Obtain the current system operating resource budget, and dynamically adjust the video processing parameters of the original video stream based on the overall event severity and the operating resource budget;

[0010] The original video stream is converted into a format according to the adjusted video processing parameters to obtain the target streaming media data.

[0011] The target streaming media data is segmented into video segments, an importance score is calculated based on the event level corresponding to the video segment, and storage eviction management is performed on the video segments according to the importance score.

[0012] Beneficial effects: This invention enables on-demand allocation of edge computing power and secure isolation from main business operations, thereby improving the retention rate of high-value abnormal video evidence within limited storage space. Attached Figure Description

[0013] Figure 1 This is the overall flowchart of the present invention.

[0014] Figure 2 This is a flowchart of the present invention for calculating the comprehensive event severity based on multi-dimensional power telemetry parameters.

[0015] Figure 3 This is a flowchart of the calculation of the overall event severity according to the present invention.

[0016] Figure 4 This is a flowchart illustrating the dynamic adjustment of video processing parameters of the original video stream according to the present invention.

[0017] Figure 5 This is a flowchart illustrating the differentiated allocation of video processing parameters for each camera channel according to the present invention. Detailed Implementation

[0018] like Figure 1 As shown in Example 1: This example provides a video acquisition and storage method based on a smart energy unit, which can be applied to smart energy units deployed in substations or power distribution rooms.

[0019] The smart energy unit is implemented using an IoT data acquisition device equipped with a general-purpose microprocessor and an embedded operating system. At the operating system level, a streaming media service process is pre-deployed via cross-compilation, and corresponding network socket creation permissions and storage directory read / write permissions are configured to support the continuous acquisition and streaming of upper-layer video data. This method includes the following steps:

[0020] Step 101: Receive the raw video stream sent by the network video surveillance equipment and collect multi-dimensional power telemetry parameters in real time.

[0021] The smart energy unit establishes a connection with the front-end network video monitoring equipment through the underlying communication interface.

[0022] Specifically, the signaling service process running internally listens on a preset network port, receives registration requests from network video surveillance devices, and obtains the online status of channels through device directory query commands. When establishing a media session, it sends a session invitation signaling to the network video surveillance device to negotiate media transmission parameters. After negotiation, it receives real-time transmission protocol data packets encapsulated with audio and video payloads from the other party, and extracts the original video stream after unpacking and reassembling them.

[0023] While acquiring video data, multi-dimensional power telemetry parameters of the monitored equipment, such as transformers and switchgear, are continuously collected via the local data bus. These multi-dimensional power telemetry parameters can include physical quantities such as operating voltage, load current, equipment temperature, or ambient gas pressure.

[0024] Step 102: Calculate the comprehensive event severity based on multi-dimensional power telemetry parameters, and determine the event level corresponding to the comprehensive event severity.

[0025] Multidimensional power telemetry parameters objectively reflect the real-time operating conditions of underlying power equipment. The system quantifies and evaluates the collected parameters, calculating a comprehensive event severity that characterizes the overall equipment operational risk level. The comprehensive event severity is a numerical indicator reflecting the degree to which the power system deviates from its normal operating state; a higher value indicates a higher risk of potential equipment failure or operational anomalies.

[0026] Furthermore, the severity of comprehensive events is categorized based on preset numerical range boundaries, and mapped to specific event levels. Event levels can be divided into different grades such as normal, attention, warning, and alert. Using event levels to abstract the underlying data provides clear control input for subsequent adaptive adjustments to video acquisition strategies and allocation of internal system resources.

[0027] Step 103: Obtain the current system operating resource budget, and dynamically adjust the video processing parameters of the original video stream based on the comprehensive event severity and the operating resource budget.

[0028] The smart energy unit is an edge computing device that needs to simultaneously handle the dual tasks of basic power data reporting and video encoding / decoding processing.

[0029] To prevent multimedia processing processes from excessively consuming underlying computing resources, the system periodically polls the microprocessor's load status and memory usage to extract the available runtime resource budget for video processing tasks. Subsequently, based on the comprehensive event severity obtained from the aforementioned assessment, dual-driven parameter adjustments are applied to the raw video stream.

[0030] When the overall event severity is high, the control logic tends to increase video processing parameters to obtain clearer on-site images; when operating resource budgets are tight, the control logic will forcibly suppress video processing parameters to ensure the core power data acquisition task. Through this adjustment mechanism that balances resource and business constraints, the quality of critical anomaly images is maximized while ensuring the stable operation of edge devices.

[0031] The video processing parameters may include variables such as output frame rate, video encoding bitrate, or image resolution.

[0032] Step 104: Convert the original video stream according to the adjusted video processing parameters to obtain the target streaming media data.

[0033] After parameter adjustment and control, the streaming media service process performs decapsulation and protocol conversion operations on the original video stream.

[0034] Specifically, the system converts the original video encoding format into target streaming media data that supports low-latency network distribution or direct playback on front-end web pages, based on the network requirements and decoding capabilities of the target playback terminal.

[0035] In actual system deployments, there are multiple ways to implement data handover between the signaling service process and the streaming media service process. In one implementation, the signaling service process forwards the extracted network data packets directly to the listening port bound to the streaming media service process via the loopback address of the local network stack. In another optional implementation, the streaming media service process actively retrieves the underlying stream data from the system memory buffer and performs format conversion by calling the application programming interface representing the dynamic streaming proxy.

[0036] Step 105: Divide the target streaming media data into video segments, calculate the importance score based on the event level corresponding to the video segment, and perform storage eviction management on the video segment according to the importance score.

[0037] To facilitate long-term management of video data and subsequent accident tracing, the continuously output target streaming media data is truncated according to a fixed time span, divided into multiple independent video segments, and stored in the device's local storage medium.

[0038] The system extracts the highest event level associated with each video segment within its generation time period and calculates the segment's importance score based on this. This importance score directly reflects the substantive evidentiary value of the current video content for power operation and maintenance analysis.

[0039] When the remaining available space on the local storage medium falls below a set safety threshold, video clips with the lowest importance scores are deleted first, in ascending order, to free up storage space. This mechanism allows high-scoring video clips recording severe power anomalies to resist conventional overwriting strategies and be retained for a long time, increasing the density of critical video evidence within limited storage capacity.

[0040] Example 2: Based on Example 1 above, as follows Figure 2 and Figure 3 As shown in the figure, this embodiment further details the specific steps for calculating the comprehensive event severity based on multi-dimensional power telemetry parameters.

[0041] In one possible implementation, the following steps are included:

[0042] Step 201: Obtain the preset normal operating range boundary and preset alarm threshold corresponding to each power telemetry parameter.

[0043] The preset normal operating range boundaries represent the parameter fluctuation range of the corresponding physical equipment in a healthy state. The preset alarm thresholds indicate the critical points where the equipment enters a clearly dangerous state. These boundaries and thresholds are pre-extracted from equipment operation and maintenance procedures or statistical distributions of data from historical stable operating cycles and are stored in the system's configuration storage module. The preset normal operating range boundaries are specifically configured with upper and lower limits, and the preset alarm thresholds are correspondingly configured with upper and lower alarm values. At the start of each evaluation cycle, the system loads these static parameters into memory for comparison.

[0044] Step 202: Based on the preset normal operating range boundary and the preset alarm threshold, calculate the normalized anomaly degree corresponding to each power telemetry parameter.

[0045] Normalized anomaly rate is an indicator that eliminates differences in physical dimensions and is used to measure the severity of a specific power telemetry parameter deviating from its healthy state. When a parameter is within the preset normal operating range boundary, the normalized anomaly rate is equal to 0. When a parameter crosses the preset normal operating range boundary but has not yet exceeded the preset alarm threshold, the system performs a linear mapping operation.

[0046] Specifically, for operating conditions where parameters exceed the upper limit, the formula for calculating the normalized outlier is as follows:

[0047] a_i(t)=(x_i(t)-H_i) / (H_i_alert-H_i);

[0048] Where a_i(t) is the normalized anomaly of the i-th power telemetry parameter at time t, x_i(t) is the real-time sampled value of the i-th power telemetry parameter at time t, H_i is the upper limit of the preset normal operating range boundary, and H_i_alert is the upper limit alarm value of the preset alarm threshold.

[0049] For operating conditions where parameters are below the lower limit, a similar reverse interpolation ratio is used for calculation. When the real-time sampled value exceeds the upper alarm value or falls below the lower alarm value, the corresponding normalized outlier is forcibly truncated to 1. After the above processing, all underlying physical parameters are converted into dimensionless feature quantities that are comparable to each other and located within closed intervals.

[0050] Step 203: Calculate the weighted sum of individual anomalies based on the normalized anomaly degree and the pre-configured parameter weights.

[0051] The weighted sum of individual anomalies is used to characterize the overall risk level of the system when the physical parameters are assumed to be independent. The pre-configured parameter weights are pre-assigned according to the importance of the equipment components to which each parameter belongs, and the sum of the weights of all monitored parameters equals 1.

[0052] By using a loop to iterate through the operations, the normalized outlier of each parameter is multiplied by its specific weight, and then all the product results are summed.

[0053] Specifically, E_ind(t)=∑(w_i*a_i(t));

[0054] Where E_ind(t) is the weighted sum of individual anomalies, w_i is the pre-configured parameter weight of the i-th power telemetry parameter, a_i(t) is the normalized anomaly of the i-th power telemetry parameter at time t, and ∑ represents the summation operation on the corresponding terms of all monitored parameters.

[0055] Step 204: Obtain the pre-configured set of key parameter coupling pairs and the corresponding preset coupling amplification coefficients.

[0056] Conventional weighted summation models cannot identify the superposition of nonlinear risks caused by concurrent parameter degradation.

[0057] To this end, the system pre-constructs a set of key parameter coupling pairs based on the physical fault mechanism model of electrical equipment. Taking a transformer thermal runaway scenario as an example, this set can contain tuples consisting of load current and transformer oil temperature. Each parameter pair is bound to a preset coupling amplification factor, which must be strictly greater than 1. The preset coupling amplification factor is used to amplify the alarm response level triggered when a specific combination of parameters deteriorates simultaneously at the control logic level.

[0058] Specifically, the construction of the key parameter coupling pair set is not a random combination of parameters, but rather a strict mapping of the physical fault evolution mechanism of the underlying power equipment. The system binds physical quantities that cannot be accurately represented by a single parameter, but which can clearly indicate significant safety hazards when multiple parameters deteriorate simultaneously, into coupling pairs.

[0059] For example, the system is pre-configured to include a current-temperature coupling pair between load current and transformer oil temperature. From an electrophysical perspective, an abnormal increase in load current will lead to a sharp increase in transformer copper and iron losses. When the cooling system fails, the transformer oil temperature will inevitably rise synchronously. The simultaneous deterioration of both constitutes a typical physical precursor to transformer thermal runaway. As another example, the system is pre-configured to include a pressure-temperature coupling pair between ambient sulfur hexafluoride gas pressure and switchgear internal temperature. Synchronous anomalies in these two parameters indicate accelerated aging of the insulating medium caused by gas chamber sealing deterioration or partial discharge. Through this setting that strongly binds abstract algorithm inputs to specific equipment physical fault models, the system can accurately distinguish between independent minor parameter disturbances and complex fault conditions with highly intrinsic physical correlations.

[0060] Step 205: For each parameter pair in the key parameter coupling pair set, extract the smaller normalized anomaly degree among the two power telemetry parameters constituting the parameter pair as the coupling base value.

[0061] The logic for extracting the minimum value constitutes the mathematical boundary of the coupling judgment. When only one of the two power telemetry parameters constituting the parameter pair is abnormal while the other is completely normal, the normalized anomaly degree of the normal parameter is equal to 0, and the extracted minimum value is also equal to 0, thus blocking the path of single-source anomaly triggering coupled amplification. Only when both parameters show a deterioration trend and their respective normalized anomalies are both greater than 0, is the extracted coupling baseline value a positive real number, providing a reference value for subsequent nonlinear amplification.

[0062] Step 206: Multiply the basic coupling value by the corresponding preset coupling amplification factor to obtain the coupling anomaly degree of each parameter.

[0063] Performing scalar multiplication amplifies coupling risks. To avoid overflow of the amplified values, the multiplication result is forcibly truncated.

[0064] Specifically, e_ij(t)=min(1,c_ij*e_base_ij(t));

[0065] Where e_ij(t) is the coupling anomaly degree corresponding to the parameter pair consisting of the i-th parameter and the j-th parameter, c_ij is the preset coupling amplification coefficient corresponding to the parameter pair, e_base_ij(t) is the coupling base value obtained by the extraction operation, and min is the minimum value operation, which is used to limit the final coupling anomaly degree to no more than 1.

[0066] Step 207: Select the maximum value among all parameter pairs in terms of coupling anomaly degree as the comprehensive coupling anomaly term.

[0067] In monitoring scenarios with multiple key parameter coupling pairs, the system may simultaneously detect various complex degradation indicators. The system iterates through all calculated coupling anomalies within the set and extracts the single item with the highest value. This extraction operation ensures that the system's control resources prioritize responding to the most critical complex fault condition. The extracted extreme value is assigned as the comprehensive coupling anomaly item.

[0068] Step 208: Compare the weighted sum of individual anomalies with the comprehensive coupled anomalies, and extract the larger value as the comprehensive event severity.

[0069] Specifically, E(t) = max(E_ind(t), E_coup(t));

[0070] Where E(t) is the overall event severity after fusion processing, E_ind(t) is the weighted sum of individual anomalies, E_coup(t) is the overall coupled anomaly, and max is the operation instruction to extract the larger value of the two.

[0071] For example, suppose that at a certain moment, the normalized anomaly of the load current is calculated to be 0.35, the normalized anomaly of the transformer oil temperature is calculated to be 0.30, and the preset coupling amplification factor is set to 2.5. The weighted sum of individual anomalies might yield a value of 0.16, indicating a minor risk; however, after extracting the smaller value of 0.30 and multiplying it by the amplification factor of 2.5, the generated comprehensive coupling anomaly term reaches 0.75. After comparison, the final comprehensive event severity is set to 0.75, the system's judgment level jumps, and the problem of missed composite faults caused by independent weighting is reduced.

[0072] Furthermore, in another optional implementation, to capture hidden fault conditions where parameters undergo drastic and sudden changes, after obtaining the comprehensive event severity, the system also simultaneously calculates the weighted sum of the rates of change of each parameter within a preset time window. Specifically, the system extracts the absolute value of the difference between the sampled value at the current moment and the sampled value at a previous historical moment, compares it with the span of the parameter alarm interval, multiplies it by the parameter weight, and sums the results. To avoid drastic disturbances causing the indicators to exceed limits, a truncation constraint is applied to the weighted sum of the rates of change, and then a second extreme value extraction is performed with the comprehensive event severity.

[0073] The calculation formula is as follows:

[0074] E_final(t)=max(E(t),min(1,sum(w_i*|x_i(t)-x_i(t-dt)| / (H_i_alert-L_i_alert))));

[0075] Where E_final(t) is the final corrected comprehensive event severity, E(t) is the comprehensive event severity output from the aforementioned comparison operation, dt is the set calculation time window, x_i(t) and x_i(t-dt) are the corresponding real-time sampled value and historical sampled value, respectively, and L_i_alert is the lower limit alarm value of the preset alarm threshold. The physical meanings of the remaining variables are consistent with those described above.

[0076] like Figure 4 and Figure 5 As shown in Example 3, the specific execution steps for dynamically adjusting the video processing parameters of the original video stream based on the comprehensive event severity and runtime resource budget are further explained.

[0077] In one possible implementation, the network video surveillance device is connected to multiple camera channels, including the following steps:

[0078] Step 301: Obtain the spatial correlation matrix between the pre-configured power telemetry parameters and the camera channel.

[0079] Conventional global monitoring strategies typically implement uniform image quality enhancement across all camera channels upon receiving an alarm, leading to an instantaneous depletion of computing resources on individual boards. To achieve precise allocation of processing power, the system establishes a spatial dimension mapping model during the deployment phase. The spatial correlation matrix is ​​structurally represented as a two-dimensional floating-point array. Its row dimension corresponds to the various power telemetry parameters generated by the monitored underlying physical devices, and its column dimension corresponds to the field of view of each camera channel deployed on-site. Each element in this spatial correlation matrix represents the correlation coefficient between the monitoring area of ​​a specific physical quantity and the field of view of a specific monitoring image, with values ​​strictly limited to a closed interval between 0 and 1. The closer the value is to 1, the stronger the physical correlation between the camera channel capturing the abnormal state of that power telemetry parameter.

[0080] The specific numerical settings of the correlation coefficient elements in the spatial correlation matrix depend on the geometric overlap between the deployment location of the power sensor and the actual physical field of view of the camera channel.

[0081] In the system initialization deployment scenario, technicians extract the three-dimensional physical coordinates of each sensor that generates power telemetry parameters, and at the same time extract the lens installation position, focal length, and field of view parameters of each camera channel.

[0082] When the physical coordinates of a sensor are completely located within the core field of view of a camera channel, the correlation coefficient between the corresponding parameter of the sensor and the camera channel is calibrated to a maximum value approaching 1.

[0083] When the sensor's physical coordinates are at the edge of the camera channel's field of view, this correlation coefficient is calibrated as an intermediate transition value;

[0084] When the sensor's physical coordinates are completely within the blind zone of the camera channel's field of view, the correlation coefficient is calibrated to 0.

[0085] Through the data input setting logic based on the above three-dimensional physical space mapping, the spatial correlation matrix can characterize the inherent cross-modal correlation between invisible electrical telemetry data and visible monitoring images.

[0086] Step 302: Use the spatial correlation matrix to perform spatial mapping processing on the normalized anomaly degree corresponding to each power telemetry parameter to obtain the channel-level event correlation degree corresponding to each camera channel.

[0087] The dimensionless state of each parameter is continuously collected during the current scheduling cycle, and then projected onto the visual acquisition terminal through matrix multiplication and addition operations.

[0088] R_k(t) = ∑(m_ik*a_i(t));

[0089] Where R_k(t) is the channel-level event correlation degree calculated by the k-th camera channel at time t, m_ik is the correlation coefficient element of the i-th power telemetry parameter and the k-th camera channel in the spatial correlation matrix, a_i(t) is the normalized anomaly degree of the i-th power telemetry parameter at time t, and ∑ represents the summation operation of the corresponding product terms of all monitored parameters.

[0090] Step 303: Based on the channel-level event correlation, differentiate the video processing parameters corresponding to each camera channel. Adjust the resource constraints of the differentiated video processing parameters based on the runtime resource budget to obtain the final video processing parameters for each camera channel.

[0091] By extracting channel-level event correlation, discrete Boolean alarms are transformed into continuous quality allocation weights. When a local electrical anomaly occurs in a certain area, only specific camera channels whose physical field of view covers that area can accumulate a high channel-level event correlation, while the values ​​obtained by camera channels in unrelated areas tend to be close to 0.

[0092] Step 304: Obtain the preset baseline frame rate and the original frame rate.

[0093] The preset baseline frame rate corresponds to the low-power monitoring parameters of the device when no abnormal events occur, and is set to a low value to save system bus transmission bandwidth. The raw frame rate corresponds to the full-load parameters of the complete video stream natively output by the front-end camera hardware. The system extracts these two baseline values ​​and uses them as the lower and upper limits of the physical range for subsequent adaptive adjustment.

[0094] Step 305: Based on the channel-level event correlation, perform a linear mapping between the baseline frame rate and the original frame rate to obtain the independent target output frame rate for each camera channel.

[0095] The channel-level event correlation is converted into a dimensionless quality parameter, and then the specific frame rate metric is calculated by interpolation within the extreme value range of this quality parameter. The lower quality limit baseline value Qmin is set to 0, and the upper quality limit full load value Qmax is set to 1, so that the quality parameter Q_k(t) varies within a closed interval of 0 to 1, ensuring that the target output frame rate fully covers the baseline frame rate and the original frame rate.

[0096] The calculation formula is as follows:

[0097] Q_k(t)=Q_min+(Q_max-Q_min)*min(1,R_k(t) / R_th);

[0098] Where Q_k(t) is the quality parameter obtained by independently mapping the k-th camera channel, Q_min is the system-set lower limit benchmark value for quality, Q_max is the system-set upper limit full load value for quality, R_k(t) is the corresponding channel-level event correlation degree, R_th is the set channel quality full load threshold parameter, and min is the minimum value extraction operation. The channel quality full load threshold parameter R_th is used to determine at what level the channel-level event correlation degree should reach for the corresponding channel's video quality to be improved to full load. Its specific value can be determined based on the number of camera channels at the deployment site and the column normalization characteristics of the spatial correlation matrix.

[0099] After obtaining the quality parameters, a linear transformation of the physical quantity of frame rate is further performed.

[0100] fps_out_k(t)=fps_min+(fps_max-fps_min)*Q_k(t);

[0101] Where fps_out_k(t) is the target output frame rate calculated for the k-th camera channel, fps_min is the obtained baseline frame rate, and fps_max is the obtained original frame rate.

[0102] Step 306: Use the target output frame rate corresponding to each camera channel as the video processing parameter.

[0103] The output frame rate values ​​calculated independently for each channel are assigned to the underlying video stream processing pipeline, thus establishing the execution target for each data channel in the next cycle.

[0104] Step 307: When using the target output frame rate corresponding to each camera channel as a video processing parameter and performing adjustment, the inter-frame temporal layering structure of the original video stream is parsed, and the non-reference frames of the enhancement layer with the highest temporal layer are discarded first, while the reference frames of the base layer are retained, until the actual output frame rate of each camera channel matches the corresponding target output frame rate.

[0105] Specifically, when using high-efficiency video coding standards to process video streams, forcibly reducing data packets may cause screen tearing in the client player due to the loss of critical reference information.

[0106] The decapsulation module deployed within the system identifies the current frame's level in the temporal hierarchy by reading the temporal identifier in the network abstraction layer unit header identifier field. When executing a frame rate reduction command, the system selectively discards enhancement layer image frames with the highest temporal identifier. Since the underlying base layer image frames remain intact, the client's video decoder can still reconstruct the image using the preserved reference frames, thus smoothly reducing the pressure on the communication bus while ensuring the continuity of the visual monitoring footage.

[0107] Step 308: Monitor the time series changes of event levels in real time.

[0108] The system's state machine continuously tracks the comprehensive event severity level assessed within each scheduling cycle. This step aims to capture moments of abrupt changes in system operating conditions and prevent response lag issues caused by slow, periodic recording mechanisms.

[0109] Step 309: When an upward jump in event level is detected, a keyframe request command is generated and sent to the network video surveillance device to obtain the keyframe of the current frame. In other words, the network video surveillance device immediately inserts the keyframe of the current frame into the original video stream.

[0110] When an event level escalates from normal to warning or from warning to alarm, it indicates that a dangerous action may be occurring on-site. At this point, passively waiting for the next critical image frame in the next natural cycle, as per standard protocols, would miss the optimal opportunity to record the moment the incident originates. The system triggers protocol signaling interaction, using the session-initiated protocol's control message channel to forcibly issue a command to the underlying monitoring hardware. This causes the hardware encoder to immediately interrupt the existing image sequence and forcibly insert a complete spatial reference frame, thereby ensuring that the monitoring image quality reaches its highest level instantaneously at the initial stage of the emergency.

[0111] Furthermore, as an offline adaptive improvement mechanism for the system, the following steps are also included during the long-term operation of receiving the original video stream:

[0112] Step 310: Periodically calculate the average severity of comprehensive events within the set historical time period, and calculate the false trigger rate of events for each camera channel within the set historical time period;

[0113] The system uses the past day or week as a fixed backtracking time window to calculate the average of the severity records of all comprehensive events stored in the database. This average objectively reflects the basic risk characteristics and equipment aging status of a specific monitoring site during long-term operation.

[0114] Step 311: Based on the average severity of the comprehensive events, dynamically configure the duration of the loop pre-recording buffer for video recording and update the spatial correlation matrix synchronously.

[0115] Specifically, based on the average severity of comprehensive events, the duration of the loop pre-recording buffer for video recording is dynamically configured so that the pre-recording depth of the original video stream is cached according to the duration of the loop pre-recording buffer; and the correlation coefficient of the corresponding camera channel in the spatial correlation matrix is ​​updated synchronously based on the event false trigger rate.

[0116] For sites that experience prolonged parameter fluctuations and have high average values, the system automatically extends the memory residency time allocated to the video stream pipeline, expanding the capture time window for event precursors. For sites with stable operation, the buffer depth is reduced to reclaim available memory.

[0117] T_pre(t)=T_base+T_ext*E_mean;

[0118] Where T_pre(t) is the calculated duration of the dynamic cyclic pre-recording buffer, T_base is the set basic pre-recording safety time, T_ext is the maximum allowed time margin for expansion, and E_mean is the calculated average comprehensive event severity over a specific historical period. Furthermore, based on the comparison results of long-term video tags, the system performs weight decay and dynamic updates on the spatial correlation matrix elements corresponding to channels that generate frequent false triggers, thereby achieving adaptive optimization in response to complex on-site conditions.

[0119] By introducing a nonlinear severity assessment model coupled with multidimensional physical parameters and a spatial mapping matrix, discrete electrical anomalies were successfully converted into continuous quality allocation weights for the visual acquisition channel. This mechanism breaks the deadlock of blind global linkage, allowing limited audio and video processing computing power to be targeted to the core spatial area where the anomaly occurred. As a result, while maintaining low overall power consumption of the equipment, the frame rate and visual continuity of the image at the moment of critical failure are significantly improved.

[0120] Example 4: In one possible implementation, obtaining the system's current operating resource budget and video stream format conversion includes the following steps:

[0121] Step 401: Within the set scheduling period, collect the resource utilization rate of the device's main services in real time.

[0122] The primary responsibility of edge computing devices is to ensure the data integrity of core power telemetry services. During operation, the system reads the process status statistics interface file at the operating system level to extract the CPU time cycles occupied by the main business process responsible for power acquisition and control, thereby calculating the actual processing load ratio of the main business process within the current time slice.

[0123] Step 402: Obtain the pre-configured system underlying operating overhead and reserved safety margin.

[0124] The system's underlying operating overhead represents the proportion of basic computing power necessary to maintain the normal operation of the operating system kernel and basic resident services. The reserved safety margin is used to cope with potential surges in computing power demand during batch data uploads or the execution of emergency remote control commands in the main power grid operations. Both parameters are fixed and provided by the equipment's factory configuration or the issued operation and maintenance configuration items.

[0125] Step 403: Subtract the resource occupancy rate of the main business, the underlying operating overhead of the system, and the reserved safety margin from the total processing resources of the system to obtain the video resource budget.

[0126] Specifically, the calculated video resource budget is as follows:

[0127] C_video(t)=max(0,C_total-C_power(t)-C_os-C_reserve);

[0128] Wherein, C_total is the normalized total processing resource capacity of the edge device, expressed as a percentage and always equal to 1, C_power(t) is the resource utilization rate of the main business of the device collected in real time, C_os is the acquired system underlying operating overhead, C_reserve is the set reserved safety margin, and max is the maximum value extraction operation, used to limit the lower limit of the deduction result to 0.

[0129] Step 404: Use the video resource budget as the runtime resource budget.

[0130] After assigning values ​​to the variables, the system inputs the calculated available computing power into the control logic of video processing.

[0131] Step 405: Calculate the ratio of the video resource budget to the maximum available resources under system idle conditions to generate a resource adjustment factor. This ratio reflects the performance limit that the current video pipeline can be allowed to operate.

[0132] R(t)=min(1,C_video(t) / C_ref);

[0133] Where R(t) is the generated resource adjustment factor, C_video(t) is the video resource budget, C_ref is the maximum available resource baseline value under system no-load conditions, and min is the minimum value extraction operation, used to constrain the generated ratio to a range not greater than 1.

[0134] Step 406: Obtain the target execution parameters that are independently determined based on the overall event severity.

[0135] The target execution parameters correspond to the expected processing indicators calculated solely based on visual quality requirements in the preceding process, and can be specifically represented as the target output frame rate obtained by independently mapping each camera channel.

[0136] Step 407: Based on the comprehensive event severity and the preset event-parameter mapping rules, determine the target execution parameters; use the resource adjustment factor to proportionally reduce the target execution parameters to obtain the final video processing parameters.

[0137] The system enforces safety constraints here to forcibly intervene in visual requirements.

[0138] fps_actual_k(t)=fps_out_k(t)*R(t);

[0139] Where fps_actual_k(t) is the final video processing parameter issued for a specific camera channel, fps_out_k(t) is the target execution parameter obtained, and R(t) is the resource adjustment factor.

[0140] In a complex operating condition, if a severe electrical fault occurs on-site, increasing the target output frame rate requirement, but simultaneously triggering a large number of concurrent protection messages, leading to increased occupancy of main service resources, the calculated resource adjustment factor will drop significantly. Through the aforementioned multiplication operation, the lower-order resource adjustment factor will unconditionally suppress the higher-order frame rate demand. This processing mechanism establishes the absolute priority of resource safety constraints over video quality requirements, reducing the risk of edge devices crashing due to full load of video components.

[0141] Step 408: Obtain the preset multi-level protection threshold, which includes a first-level threshold and a second-level threshold.

[0142] The system loads and sets multi-level hard circuit breaker boundary parameters to execute low-level forced blocking when multiplication constraints cannot smooth out peak computing power.

[0143] Step 409: When the resource utilization rate of the main business exceeds the first-level threshold, bypass the dynamic adjustment process of the resource budget and forcibly switch the video processing parameters to the preset lowest quality mode.

[0144] When the core load exceeds the first line of defense, the system stops calculating the smooth degradation logic of the resource adjustment factor and activates the hard power switch. At this time, the streaming media service module only retains the basic keyframe extraction function, discarding all enhanced frames to minimize the power consumption of the video pipeline.

[0145] Step 410: When the resource utilization rate of the main business continues to rise and exceeds the secondary threshold, a video processing pause command is triggered to release all video processing resources until the resource utilization rate of the main business falls back below the secondary threshold.

[0146] If the load continues to worsen and breaches the second line of defense, the system will directly sever the video process's scheduling privileges over the microprocessor. The primary service will monopolize all microprocessor cycles until the dangerous situation is resolved.

[0147] Step 411: While collecting the resource utilization rate of the device's main business in real time, the available memory of the operating system is also monitored simultaneously.

[0148] In addition to competing for computation cycles, streaming media services also consume a large amount of random access memory space when caching data. To prevent triggering the operating system's low-memory killer process, the system periodically reads the memory allocation table status at the kernel level.

[0149] Step 412: When it is determined that the available memory is lower than the preset memory alarm threshold, the depth of the loop pre-recording buffer for streaming media data is forcibly reduced to the length of a single image group (GOP) to release the memory space occupied by the loop pre-recording buffer.

[0150] This step quickly releases heap memory by discarding historical buffer blocks, ensuring that the system does not run out of memory and crash.

[0151] Step 413: Obtain the client's video decoding capability declaration.

[0152] When a video stream is ready to be distributed externally, the streaming media service process extracts the supported encoding type information fed back by the receiving end during the session description protocol negotiation phase of signaling interaction, or during the parameter query phase when the transmission connection is established.

[0153] Step 414: When the video decoding capability declaration indicates support for the encoding format of the original video stream, the original video stream is de-encapsulated and the target stream container format is re-encapsulated according to the adjusted video processing parameters to obtain the target streaming media data.

[0154] When the client has the decoding capability of a new, efficient encoding standard, the system executes the default first conversion path. At this point, the microprocessor only needs to strip the outer protocol header of the original video data and directly package the original video elementary stream into the container format of the target transport protocol. This pure transcoding process does not change the pixel array structure of the internal video frames, resulting in extremely low CPU power consumption.

[0155] Step 415: When the video decoding capability declaration indicates that the encoding format of the original video stream is not supported, the preset hardware encoding and decoding unit is invoked to decode the original video stream and re-encode it into a compatible format according to the adjusted video processing parameters, and then encapsulate it into the target streaming media data.

[0156] When the client is an older device that only supports traditional basic encoding standards, the system switches to the second conversion path. The system feeds the extracted high-dimensional video elementary stream directly to the independent hardware multimedia processing engine built into the system-on-chip via the memory-mapped interface. The hardware decoder restores it to lossless pixel matrix data, which is then transmitted to the hardware encoder via the internal data bus. The encoded stream is compressed into a basic compatible format, and finally, the streaming media process completes the protocol encapsulation.

[0157] Furthermore, as an alternative, when the system is in the operation mode of calling the hardware encoding and decoding unit for transcoding, the aforementioned process of proportionally reducing using resource adjustment factors can be replaced by constraining the video frame rate instead of constraining the target output bit rate of the hardware encoder.

[0158] bitrate_actual(t)=bitrate_max*R(t);

[0159] Where bitrate_actual(t) is the upper limit of the actual operating bitrate of the constrained hardware encoder, bitrate_max is the configured baseline bitrate upper limit, and R(t) is the generated resource adjustment factor. Using bitrate compression as an alternative can maintain the smoothness of the visual stream, but it will increase compression artifacts in a single frame.

[0160] A dual control defense system is adopted, consisting of hard stripping of core business resources and multiplication constraints on adjustment factors. When the core telemetry load of the power grid surges, the control logic can forcibly suppress the resource consumption of the underlying multimedia encoding and decoding process through unconditional multiplication degradation and even multi-level hardware circuit breaker mechanisms. This mechanism establishes the absolute scheduling priority of core power data and reduces the security risks of monitoring components being overloaded and dragging down the underlying operating system.

[0161] Example 5 provides a dynamic anti-jitter processing mechanism for weak network communication environments, which is suitable for unreliable transmission deployment scenarios where edge devices access remote video surveillance hardware via wireless mobile networks.

[0162] In one possible implementation, the following steps are included:

[0163] Step 501: Extract the data packets carrying the original video stream in real time and calculate the network packet loss rate within the sliding time window.

[0164] In wireless communication links, the underlying data transmission protocol typically does not provide absolute reliability guarantees. The system's receiving and processing module continuously parses the underlying transmission protocol header, extracting identification fields such as sequence numbers to detect discontinuities in the data stream.

[0165] The system maintains a sliding monitoring window with a fixed time span in memory. By comparing the total number of data packets expected to be received within this window with the actual number of missing data packets, the real-time network packet loss rate is calculated. When the network packet loss rate exceeds a set tolerance threshold, the system sends a control message to the front-end monitoring equipment via the control signaling channel to reduce the video output bitrate or frame rate.

[0166] Furthermore, if the duration of the data stream interruption exceeds the set disconnection threshold, the system determines that the current connection is invalid, actively releases the current media session to the lower end, and re-initiates the session establishment request according to the preset exponential backoff delay strategy.

[0167] Step 502: Calculate the average arrival interval deviation between consecutively arriving data packets, and extract the corresponding standard deviation based on the average arrival interval deviation.

[0168] Due to the combined effects of wireless channel fading and communication base station scheduling, the arrival time of data packets at the receiving end often exhibits random delay fluctuations. The system uses the received timestamp to record the actual time interval between the arrival of two adjacent data packets at the local network physical interface, and combines this with the initial timestamp from the sending end to extract the arrival interval deviation. Subsequently, based on conventional statistical methods, the system calculates the distribution of arrival interval deviations for a set number of historical consecutive data packets, extracting their corresponding standard deviation to reflect the current dispersion of transmission delay in the communication link.

[0169] Step 503: Based on the standard deviation and network packet loss rate, adaptively calculate and adjust the depth of the network jitter buffer, and use the network jitter buffer to perform transmission smoothing processing on the original video stream.

[0170] The system allocates a dynamically adjusted network jitter buffer at the application layer to temporarily store out-of-order or delayed data packets. When determining the specific time span of this buffer, the system uses the calculated standard deviation as a positively correlated adjustment criterion.

[0171] Specifically, when the network environment deteriorates and the standard deviation increases significantly, the system proportionally increases the depth of the network jitter buffer to provide a more tolerant reassembly waiting window; when the network environment returns to stability and the standard deviation shrinks to the baseline level, the system proportionally reduces the depth of the buffer to reduce the overall end-to-end latency of the video stream.

[0172] Through an adaptive scaling mechanism, the system uses memory cache space to exchange for the continuity of data transmission time, effectively absorbing timing jitter in the wireless transmission process, and outputting smooth raw video stream data to the backend streaming media decapsulation module.

[0173] Example 6 further details the specific steps involved in calculating the importance score based on the event level corresponding to the video clip. Furthermore, based on the actual needs of business deployment, this example also expands upon auxiliary functions such as storage eviction management, cloud data synchronization, and multi-protocol interface services. By introducing an event persistence dimension into the scoring model, this implementation method effectively addresses the pain point of instantaneous harmless pulses consuming storage space, thereby improving the local retention rate of high-value videos.

[0174] In one possible implementation, the following steps are included:

[0175] Step 601: Obtain the preset event activity determination threshold.

[0176] The system configuration module stores a numerical boundary for filtering underlying noise. This preset event activity judgment threshold is usually consistent with the lower limit of the warning level. When the overall event severity is below this threshold, the system determines that the equipment is in a normal fluctuation state; only when the overall event severity is equal to or greater than this threshold does the system determine that the power equipment is currently in a true event activity period.

[0177] Step 602: Calculate the proportion of active time periods within the time span of the video segment where the overall event severity exceeds the event activity threshold.

[0178] The system divides continuously generated video streams into video segments with fixed-length time spans. Within these time spans, the system accumulates the total number of time segments where the overall event severity output by the internal state machine exceeds a preset event activity threshold. Subsequently, the calculated sum of event activity times is compared to the total time span of the video segment to output the event activity time ratio. This ratio reflects the persistence of a specific power anomaly event in the current video segment, accurately distinguishing between transient interference pulses and continuously deteriorating hardware faults.

[0179] Step 603: Extract the maximum value of the overall event severity within the video segment, and amplify and adjust the maximum value using the event active time ratio to obtain the effective severity.

[0180] Since the event level is determined by classifying the comprehensive event severity through a preset numerical range boundary, there is a monotonic correspondence between the two. Therefore, calculation based on the comprehensive event severity is equivalent to calculation based on the event level, and using continuous values ​​can obtain a more refined scoring discrimination.

[0181] Read all the comprehensive event severity values ​​generated within the video segment and extract the maximum value as the peak feature. Multiply the event activity time ratio by the pre-configured persistence enhancement coefficient, add a constant of 1 as the base value, and multiply this base value by the extracted maximum value to output the effective severity after persistence compensation.

[0182] This calculation mechanism ensures that persistent anomalous states can achieve higher fundamental feature scores than instantaneous spikes.

[0183] E_j^eff=E_j^max*(1+β*D_j);

[0184] Where E_j^eff is the effective severity calculated for the j-th video segment, E_j^max is the maximum value of the comprehensive event severity extracted within the j-th video segment, β is the system-preconfigured persistence enhancement coefficient, and D_j is the statistically obtained event active time ratio of the j-th video segment.

[0185] For example, suppose a certain event segment has a peak value of 0.5, a statistically active ratio of 0.8, and a configured persistence enhancement operator of 0.5. Substituting into the above calculation logic, the effective severity of this segment is 0.5*(1+0.5*0.8)=0.7. Through this amplification adjustment mechanism, the feature score of an oil temperature deterioration event with a low peak value but persistent occurrence has the potential to surpass that of a lightning voltage spike event with an extremely high peak value that disappears instantly.

[0186] Step 604: Calculate the importance score for the video segment based on the effective severity. Specifically, the effective severity is used as the dominant factor, and a weighted calculation is performed in combination with pre-configured scoring weights to obtain the importance score for the video segment.

[0187] The system extracts the effective severity of the above calculation output into a multi-dimensional comprehensive scoring module, which serves as the core dominant factor characterizing the correlation of underlying power events.

[0188] Step 605: Obtain the access frequency of the video segment and calculate the time decay factor based on the time span. Specifically, obtain the access frequency of the video segment, calculate the time difference between the generation time of the video segment and the current system time, and calculate the time decay factor based on the time difference;

[0189] The access frequency record shows the total number of times the video clip has been accessed and downloaded by upper-layer application systems or operations and maintenance personnel through the network interface since its generation.

[0190] It also calculates the time span that has elapsed from the generation of the video clip to the current moment.

[0191] Specifically, the system extracts the time difference between the current system time and the time the video segment was generated. This time difference is multiplied by a preset decay rate coefficient to obtain a negative exponential term. Then, the negative exponential term is raised to the power of the base of the natural logarithm to output the time decay factor. This factor characterizes the exponential decline in the value of video evidence over time.

[0192] Step 606: Combine the effective severity, access frequency, and time decay factor to calculate the importance score.

[0193] Assign corresponding importance weights to effective severity, access frequency, and time decay factor. After normalizing the access frequency to a baseline value, multiply the above three parameters by their corresponding weights and sum them to output the final comprehensive score of the video segment.

[0194] S_j=w_E*E_j^eff+w_A*(A_j / A_ref)+w_D*e^(-λ*(t_now-t_j));

[0195] Where S_j is the output importance score for the j-th video segment, w_E is the weight assigned to the effective severity, E_j^eff is the effective severity, w_A is the weight assigned to the access frequency, A_j is the obtained access frequency, A_ref is the preset normalized access reference value, w_D is the weight assigned to the time decay factor, λ is the set exponential decay rate coefficient, t_now is the current system time, t_j is the generation time of the j-th video segment, and e is the natural constant.

[0196] In one optional implementation, the aforementioned weight parameters wE, wA, and wD can be configured to emphasize different aspects of the three dimensions—event severity, access frequency, and time freshness—according to the actual operational scenario. Those skilled in the art can determine a suitable weight combination through comparative experiments on a verification dataset. The exponential decay rate coefficient λ can be adaptively adjusted according to the actual business requirements for the timeliness of video evidence; those skilled in the art can determine a suitable value without creative effort. The normalized access reference benchmark value Aref is used to eliminate the influence of the dimension of access frequency and can be set as the statistical value of the average number of accesses for all video segments within the system's operating cycle.

[0197] For example, suppose the exponent value calculated after time decay is 0.315, the effective severity is 0.7, and the normalized value of the access frequency index is 0.4. Combining the assigned weights, the system sums up the product terms to obtain the specific scalar value used for global ranking.

[0198] Step 607: Obtain the current storage space utilization rate. When the storage space utilization rate exceeds the preset utilization rate threshold, perform deletion operations on the video segments in order of importance score from low to high.

[0199] The system periodically scans the percentage of used capacity in local storage. When this percentage exceeds a set cleanup threshold, the storage management process performs an ascending-order sorting algorithm on the importance scores of all locally stored video clips and instructs the file system to sequentially erase media files with the lowest scores from the front of the queue until available storage space is restored to a safe level.

[0200] Step 608: Determine the event level corresponding to each video segment and query the cloud backup status corresponding to each video segment.

[0201] Before executing the file deletion command, the system introduces an asset protection interception mechanism. The system extracts the highest event level from the metadata of the video segment and queries the remote transmission confirmation flag of the file returned by the synchronization communication service component.

[0202] Step 609: When the event level corresponding to the video segment reaches the preset highest alarm level, or when the cloud backup status of the video segment indicates that the backup to the cloud server has not been completed, a write protection lock is marked on the video segment so that the corresponding video segment is not involved in the deletion operation.

[0203] This step establishes the legal boundaries for protecting critical evidence. By imposing a read-only lock attribute on the file descriptor, the risk of high-risk incident recordings or unarchived data being accidentally overwritten by routine eviction mechanisms is eliminated, ensuring the data integrity for incident tracing.

[0204] Furthermore, regarding the peripheral data synchronization and multi-protocol distribution mechanism in commercial deployments, this embodiment also includes the following system-level processing steps:

[0205] Step 610: Establish a synchronous communication link between the device and the cloud management platform locally.

[0206] The system connects to a remote server cluster via Transmission Control Protocol (TCP) sockets.

[0207] Step 611: When the event level corresponding to the video segment is identified as the preset highest alarm level, a high-priority uplink is triggered to push the corresponding video segment and multi-dimensional power telemetry parameters to the cloud management platform in real time.

[0208] This step ensures cross-site emergency backup of data before local equipment is damaged under extreme operating conditions.

[0209] Step 612: The original video stream is simultaneously encapsulated into a first streaming media protocol format that supports plugin-free playback on web pages and a second streaming media protocol format that supports low-latency playback on mobile devices.

[0210] The streaming media service unit uses dual output pipelines to convert the raw video into protocol stream data carrying extended encoding fields to be compatible with browsers, and mobile data based on real-time streaming protocols.

[0211] Step 613: Using a unified application programming interface (API), the target streaming media data is made available for retrieval and retrieval services, so that the upper-level power automation system can extract the corresponding video segments.

[0212] In the first embodiment, it can also be:

[0213] According to the adjusted video processing parameters, the original video stream is simultaneously encapsulated into a first streaming media protocol format data that supports plugin-free playback on web pages and a second streaming media protocol format data that supports low-latency playback on mobile devices. The first streaming media protocol format data and the second streaming media protocol format data together constitute the target streaming media data.

[0214] After the target streaming media data is segmented into video clips, a unified application programming interface (API) is used to provide retrieval and retrieval services for the video clips, enabling the upper-level power automation system to extract the corresponding video clips.

[0215] The system exposes a standardized Hypertext Transfer Protocol interface to support the data retrieval needs of upper-layer business systems.

[0216] This solution constructs a dynamic scoring and elimination system that integrates event persistence and time-related degradation indicators. By assigning retention weights to persistent electrical degradation exceeding instantaneous peak signals at the computing layer, and applying write protection locks to files in non-cloud states with the highest alarm level, it improves the asset survival density and full lifecycle retention rate of high-value incident tracing evidence within limited local storage media.

Claims

1. A video acquisition and storage method based on a smart energy unit, characterized in that, include: It receives raw video streams sent by network video surveillance equipment and collects multi-dimensional power telemetry parameters in real time; The severity of a comprehensive event is calculated based on multidimensional power telemetry parameters, and the event level corresponding to the severity of the comprehensive event is determined. Obtain the current system operating resource budget, and dynamically adjust the video processing parameters of the original video stream based on the overall event severity and the operating resource budget; The original video stream is converted into a format according to the adjusted video processing parameters to obtain the target streaming media data. The target streaming media data is segmented into video segments, an importance score is calculated based on the event level corresponding to the video segment, and storage eviction management is performed on the video segments according to the importance score.

2. The method according to claim 1, characterized in that, The comprehensive event severity is calculated based on multi-dimensional power telemetry parameters, including: Obtain the preset normal operating range boundaries and preset alarm thresholds corresponding to each power telemetry parameter; Based on the preset normal operating range boundary and preset alarm threshold, calculate the normalized anomaly degree corresponding to each power telemetry parameter; The overall event severity is calculated based on the normalized anomaly degree corresponding to each power telemetry parameter.

3. The method according to claim 2, characterized in that, Based on the normalized anomaly degree corresponding to each power telemetry parameter, the comprehensive event severity is calculated, including: The weighted sum of individual anomalies is calculated based on the normalized anomaly score and the pre-configured parameter weights. Obtain the pre-configured set of key parameter coupling pairs and their corresponding preset coupling amplification coefficients; For each parameter pair in the key parameter coupling pair set, the smaller normalized anomaly between the two power telemetry parameters constituting the parameter pair is extracted as the coupling base value. Multiply the basic coupling value by the corresponding preset coupling amplification factor to obtain the coupling anomaly degree of each parameter. The maximum value among all parameter pairs in terms of coupling anomaly degree is selected as the comprehensive coupling anomaly term; The weighted sum of individual anomalies is compared with the comprehensive coupled anomaly item, and the larger value is extracted as the comprehensive event severity.

4. The method according to claim 1, characterized in that, The network video surveillance equipment is connected to multiple camera channels; The video processing parameters of the raw video stream are dynamically adjusted based on the overall event severity and runtime resource budget, including: Obtain the spatial correlation matrix between pre-configured power telemetry parameters and camera channels; The normalized anomaly degree corresponding to each power telemetry parameter is spatially mapped using the spatial correlation matrix to obtain the channel-level event correlation degree corresponding to each camera channel. Based on the channel-level event correlation, the video processing parameters corresponding to each camera channel are allocated differently.

5. The method according to claim 4, characterized in that, Based on channel-level event correlation, video processing parameters corresponding to each camera channel are differentiated and allocated, including: Obtain the preset baseline frame rate and the original frame rate; Based on the channel-level event correlation, a linear mapping is performed between the baseline frame rate and the original frame rate to obtain the independent target output frame rate for each camera channel; The target output frame rate corresponding to each camera channel is used as the video processing parameter.

6. The method according to claim 1, characterized in that, After determining the event level corresponding to the overall event severity, the following is also included: Real-time monitoring of time-series changes in event severity; When an upward jump in event level is detected, a keyframe request command is generated and sent to the network video surveillance equipment to obtain the keyframe of the current scene.

7. The method according to claim 1, characterized in that, The original video stream is format-converted according to the adjusted video processing parameters to obtain the target streaming media data, including: Obtain the client's video decoding capability declaration; When the video decoding capability declaration indicates support for the encoding format of the original video stream, the original video stream is decapsulated and the target stream container format is recapsulated to obtain the target streaming media data. When the video decoding capability declaration indicates that the encoding format of the original video stream is not supported, the preset hardware codec unit is invoked to decode the original video stream and re-encode it into a compatible format, and then encapsulate it into the target streaming media data.

8. The method according to claim 1, characterized in that, Obtain the system's current runtime resource budget, including: Within the set scheduling cycle, the resource utilization rate of the device's main business is collected in real time; Obtain the pre-configured system underlying operating overhead and reserved security margin; The video resource budget is obtained by subtracting the resource utilization rate of the main business, the underlying system operating overhead, and the reserved security margin from the total processing resources of the system. The video resource budget will be used as the operating resource budget.

9. The method according to claim 8, characterized in that, The video processing parameters of the raw video stream are dynamically adjusted based on the overall event severity and runtime resource budget, including: The ratio of the video resource budget to the maximum available resources under system idle conditions is calculated to generate a resource adjustment factor. Obtain the target execution parameters independently determined based on the overall event severity; The target execution parameters are proportionally reduced using a resource adjustment factor to obtain the final video processing parameters.

10. The method according to claim 8, characterized in that, After collecting the resource utilization rate of the main business of the acquisition device, it also includes: Obtain preset multi-level protection thresholds, which include primary thresholds and secondary thresholds; When the resource utilization rate of the main business exceeds the first-level threshold, bypassing the dynamic adjustment process of the resource budget, the video processing parameters are forcibly switched to the preset lowest quality mode. When the resource utilization rate of the main business continues to rise and exceeds the secondary threshold, a video processing pause command is triggered to release all video processing resources until the resource utilization rate of the main business falls back below the secondary threshold.

Images