Low-carbon signal base station energy consumption optimization control system based on edge computing

Abstract

This invention relates to the field of base station energy-saving control technology, specifically providing a low-carbon signal base station energy consumption optimization control system based on edge computing. The system includes a historical operating curve construction module, a power step event identification module, a load trajectory sample extraction module, a step trigger condition construction module, and a real-time energy consumption control module. This invention can determine the low-power switching trigger timing based on the target base station's own historical power decline pattern, reducing false triggering and persistent high power consumption under low load caused by fixed threshold control, thereby improving the accuracy of base station energy consumption optimization control and the low-carbon operation effect.

Description

Technical Field

[0001] This invention relates to the field of base station energy-saving control technology, and in particular to a low-carbon signal base station energy consumption optimization control system based on edge computing. Background Technology

[0002] With the continuous expansion of mobile communication networks, the number and operating time of signal base stations are constantly increasing, and the proportion of base station equipment in the total energy consumption of communication networks is gradually rising. Especially during low-traffic periods, a large number of base stations still need to remain in continuous operation to ensure the continuity of user access, service carrying, and network coverage. Therefore, reducing the energy consumption of base station operation while ensuring the quality of communication services has become an important technical direction in the low-carbon construction of communication networks.

[0003] Existing base station energy-saving control methods typically determine whether to implement power reduction, hibernation, resource shutdown, or low-power switching based on operational information such as service load, equipment power, and number of users. With the deployment of edge computing nodes in communication networks, some base station operational data can be collected, processed, and responded to closer to the base station, thus providing a lower-latency data processing foundation for base station energy consumption control.

[0004] However, the relationship between base station power consumption and current service load is not always instantaneous. In some low-service-load scenarios, base station power may still remain at a high level; while in other similar-load scenarios, the base station can enter a lower-power state. Some methods control based on the current load or fixed thresholds, making it difficult to accurately identify the correspondence between service load changes and actual power reduction from historical operating data. This can easily lead to problems such as inaccurate energy-saving trigger timing, low-power handover failure, or the continued existence of a low-load, high-power state. Summary of the Invention

[0005] To address the aforementioned technical issues, this invention proposes a low-carbon signal base station energy consumption optimization control system based on edge computing. By utilizing historical operating data from the edge side to analyze the power consumption variation patterns of the base station, the accuracy of base station energy consumption control can be improved.

[0006] To achieve the above objectives, the present invention provides the following technical solution: The low-carbon signal base station energy consumption optimization control system based on edge computing includes: The historical operating curve construction module is used to acquire historical service load data and historical power data of the target base station collected by edge computing nodes, and to construct the historical load curve and historical power curve of the target base station. The power step event identification module is used to extract multiple stable power platforms from the historical power curve and identify multiple power step drop events in the historical power data based on the multiple stable power platforms. The load trajectory sample extraction module is used to determine the transition time and transition type of each power step drop event, and extract the pre-step load trajectory and non-triggered control trajectory of each power step drop event from the historical load curve based on the transition time. The step trigger condition construction module is used to extract candidate step condition data from multiple pre-step load trajectories corresponding to each transition type, and to perform comparative verification based on multiple non-triggered control trajectories corresponding to each transition type to construct a step trigger condition sequence for each transition type. The real-time energy consumption control module is used to match the current service load trajectory of the target base station with the step trigger condition sequence corresponding to the current transition type during the real-time operation of the target base station, and to perform real-time energy consumption optimization control of the target base station based on the matching result.

[0007] Preferably, multiple stable power plateaus are extracted from the historical power curve, and multiple power step drop events in the historical power data are identified based on these multiple stable power plateaus, including: The historical power curve is truncate with a preset window length. The power difference of each sliding window is calculated based on the maximum and minimum power values ​​within the sliding window. Candidate power platforms in multiple sliding windows are identified based on the power difference. Traverse multiple candidate power platforms in the historical power curve, calculate the candidate center power of each candidate power platform, and perform neighbor merging on multiple candidate power platforms based on the candidate center power to determine multiple stable power platforms in the historical power curve. Extract the platform start time, platform end time, and platform power center value for each stable power platform. Based on the platform start time, platform end time, and platform power center value, calculate the transition duration and power center value difference between any two adjacent stable power platforms. Based on the transition duration and power center value difference, identify multiple power step drop events in historical power data.

[0008] Preferably, the transition time and transition type of each power step drop event are determined, and the pre-step load trajectory and non-triggered control trajectory of each power step drop event are extracted from the historical load curve based on the transition time, including: The step drop node of the power step drop event is determined based on the platform power center value, and the timestamp corresponding to the step drop node in the historical power curve is recorded as the transition time of the power step drop event. Based on the multiple platform power center values ​​corresponding to the front and back ends of multiple power step drop events, the front-end power range and back-end power range of multiple power step drop events are constructed, and the front-end power range and back-end power range are divided into multiple front-end power intervals and back-end power intervals. Multiple front-end power ranges and back-end power ranges are combined to construct multiple transition types. Based on the front-end power range and back-end power range to which the power step drop event belongs, the transition type to which each power step drop event belongs is identified. In the historical load curve, a load curve segment within a preset backtracking time is extracted from the transition time of the power step drop event to obtain the pre-step load trajectory corresponding to each power step drop event. The sliding control window is determined based on the preset backtracking duration. The historical load curves are traversed based on the sliding control window to extract the untriggered control trajectory of each pre-step load trajectory.

[0009] Preferably, candidate step condition data are extracted from multiple pre-step load trajectories corresponding to each transition type, and a comparative test is performed based on multiple non-triggered control trajectories corresponding to each transition type. The step triggering condition sequence for each transition type is constructed as follows: Extract individual load upper bound values ​​from each pre-step load trajectory corresponding to the transition type, and determine the candidate low load judgment upper bound based on multiple individual load upper bound values; Determine the individual effective low load duration for each pre-step load trajectory with respect to the upper bound of the candidate low load judgment, and determine the candidate low load holding duration based on the effective low load duration of multiple individuals; Extract individual load pulse segments of each pre-step load trajectory with respect to the upper bound of candidate low load determination, and determine the candidate low load pulse shape based on multiple individual load pulse segments; Multiple sets of candidate step condition data are constructed based on the upper bound of the candidate low load judgment, the duration of the candidate low load, and the pulse shape of the candidate low load. Based on the multiple pre-step load trajectories and non-triggered control trajectories corresponding to each transition type, calculate the successful sample matching ratio and control sample rejection ratio of each group of candidate step condition data. The matching parameters for each group of candidate step condition data are calculated based on the proportion of successful sample matching and the proportion of control sample rejection. The candidate step condition data with the largest matching parameters are selected according to the matching parameters to construct the step triggering condition sequence of the transition type.

[0010] Preferably, matching the current service load trajectory of the target base station with the step triggering condition sequence corresponding to the current transition type, and performing real-time energy consumption optimization control of the target base station based on the matching result includes: Acquire the current service load data and current power data of the target base station, identify the current stable power platform corresponding to the current power data, match the power center value of the current stable power platform with the front-end power range of each transition type, and determine the successfully matched transition type as the step type to be triggered. Based on the step triggering condition sequence corresponding to the step type to be triggered, identify the step critical point in the current service load trajectory, and extract the critical load duration and critical pulse shape based on the step critical point. Determine whether the critical load duration and critical pulse pattern of the current service load trajectory meet the step triggering condition sequence. If they do, generate a low-power state switching instruction, which is used to trigger the target base station to switch from the front-end power range of the current step type to the back-end power range.

[0011] Preferably, the matching parameters for each group of candidate step condition data are calculated based on the proportion of successful sample matching and the proportion of control sample rejection, including: Calculate the product of the successful sample matching rate and the control sample rejection rate, and the sum of the successful sample matching rate and the control sample rejection rate. Use the ratio between the product of the successful sample matching rate and the control sample rejection rate and the sum of the successful sample matching rate and the control sample rejection rate as the matching parameter for the candidate step conditional data.

[0012] Preferably, the method of extracting the untriggered control trajectory for each pre-step load trajectory by traversing historical load curves using a sliding control window includes: Extract candidate load curve segments from multiple sliding control windows, calculate the trajectory distance between each candidate load curve segment and the pre-step load trajectory, and use the candidate load curve segments whose trajectory distance is less than a preset similarity distance threshold and whose corresponding power step drop event does not occur within a preset response time as the untriggered control trajectory of the pre-step load trajectory.

[0013] The present invention has the following beneficial effects: This invention uses edge computing nodes to perform correlation analysis on the historical service load data and historical power data of the target base station. First, it identifies the actual power step drop events that have occurred in the target base station from the historical power curve. Then, it constructs step trigger condition sequences for different transition types by combining the corresponding pre-step load trajectory and the non-triggered control trajectory. This allows energy consumption control to be based on the historical power drop patterns of the target base station itself, rather than simply relying on fixed load thresholds or preset time strategies. During the real-time operation of the target base station, the type of transition to be triggered is determined according to the current power state. Based on the matching result of the current service load trajectory and the corresponding step trigger condition sequence, a low-power state switch is triggered, thereby improving the accuracy of low-power switching timing, reducing the situation where power remains at a high level under low service load, reducing the probability of ineffective energy-saving triggers and frequent switching, and improving the stability and effectiveness of low-carbon energy consumption optimization control of the signal base station. Attached Figure Description

[0014] Figure 1This is a schematic diagram of the low-carbon signal base station energy consumption optimization control system based on edge computing in an embodiment of the present invention. Detailed Implementation

[0015] To enable those skilled in the art to better understand the technical solutions of this invention, the technical solutions of the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of a low-carbon signal base station energy consumption optimization control system based on edge computing, provided in one embodiment of the present invention. The system includes a historical operating curve construction module, a power step event identification module, a load trajectory sample extraction module, a step trigger condition construction module, and a real-time energy consumption control module. These modules can be deployed in edge computing nodes on the target base station side, or they can be implemented collaboratively by the edge computing node and the base station energy consumption management unit, thereby completing historical data analysis, trigger condition construction, and real-time energy consumption control at a location close to the base station's operating site.

[0017] In one embodiment, the historical operating curve construction module is used to obtain historical service load data and historical power data of the target base station collected by the edge computing node, and construct the historical load curve and historical power curve of the target base station.

[0018] Specifically, historical service load data is used to characterize the changes in service load of the target base station during its historical operation. This data may include one or more of the following: physical resource block utilization, uplink and downlink throughput, number of connected users, number of active users, and number of service requests. In this embodiment, historical service load data can be uniformly converted into service load values ​​that characterize the service load intensity of the target base station, enabling service operation data from different sources to form a continuous service load sequence on the same time scale.

[0019] Historical power data is used to characterize the actual power changes of the target base station during its historical operation. It can include one or more of the following: the overall input power of the target base station, the power of the radio frequency unit, the power of the baseband processing unit, and the output power of the power module. In this embodiment, the historical power data can use the overall power of the target base station as the analysis object, or corresponding power curves can be constructed for the radio frequency unit or the baseband processing unit separately, in order to subsequently identify the power degradation process that occurs in the actual operation of the target base station.

[0020] Edge computing nodes collect the aforementioned historical service load data and historical power data according to a preset sampling period, and record the corresponding sampling time for each sampled data. For service load data and power data with different sampling periods, time alignment processing can be performed according to a unified time base to establish a correspondence between service load values ​​and power values ​​within the same sampling time or the same sampling period. For missing data, abnormally abrupt data, or data with discontinuous sampling times, data can be removed, interpolated, or smoothed to obtain historical data that meets the requirements of subsequent analysis.

[0021] After time alignment is completed, the historical service load sequence and historical power sequence are arranged in the order of sampling time to generate the historical load curve and historical power curve of the target base station, which are used to reflect the change process of the service carrying intensity of the target base station over time, as well as the change process of the actual power consumption of the target base station over time.

[0022] In one embodiment, the power step event identification module is used to construct the historical power curve output by the historical operation curve construction module, extract multiple stable power platforms formed by the target base station during historical operation, determine the actual power drop process that has occurred from the historical power change process of the target base station itself, and identify multiple power step drop events in the historical power data based on the multiple stable power platforms.

[0023] Specifically, the historical power curve is truncated using a preset window length to obtain multiple sliding windows. For each sliding window, the maximum and minimum power values ​​within that window are extracted, and the difference between the maximum and minimum power values ​​is calculated as the power difference for that sliding window. When the power difference is less than a preset platform fluctuation threshold, it indicates that the power change within that sliding window is small, reflecting that the target base station is in a relatively stable power operating state within that time range. Therefore, this sliding window is identified as a candidate power platform.

[0024] After obtaining multiple candidate power platforms, the process iterates through these platforms in the historical power curve and calculates the candidate center power for each platform. The candidate center power can be the average, median, or other statistical value that characterizes the power level of the candidate platform. Based on the candidate center power, the multiple candidate power platforms are merged to identify multiple stable power platforms in the historical power curve.

[0025] The aforementioned proximity merging refers to merging temporally adjacent candidate power platforms. When two candidate power platforms satisfy the temporal proximity relationship (i.e., belong to two adjacent time windows) and the difference between their candidate center powers is less than a preset platform merging threshold, the two candidate power platforms are merged into the same stable power platform. If a subsequent candidate power platform continues to be temporally adjacent to the stable power platform, and the difference between its candidate center power and the current platform power center value of the stable power platform is less than the platform merging threshold, it is then merged into the stable power platform. In this way, multiple candidate power platforms resulting from sliding truncation can be merged into a single temporally continuous stable power platform.

[0026] For each stable power platform, its start time, end time, and power center value are extracted. The start time is the earliest sampling time within that stable power platform, the end time is the latest sampling time, and the power center value characterizes the power level corresponding to that platform. This power center value can be calculated from multiple power sample values ​​within the platform; in this embodiment, the median of these sample values ​​is used as the power center value. Therefore, the historical power curve can be represented as multiple stable power platforms and unstable transition intervals between adjacent stable power platforms.

[0027] When identifying power step drop events, multiple stable power platforms are arranged in chronological order of their start times, and any two adjacent stable power platforms are analyzed. For the first and second stable power platforms that are temporally adjacent, the transition time between them is calculated based on the end time of the first stable power platform and the start time of the second stable power platform. The difference in power center values ​​is then calculated based on the power center values ​​of the first and second stable power platforms.

[0028] When the power center value of the first stable power platform is higher than that of the second stable power platform, and the difference between the two power center values ​​is greater than a preset step drop threshold, and the transition time between the two is less than a preset transition time, the power drop process from the first stable power platform to the second stable power platform is determined as a power step drop event.

[0029] The power step drop events identified using the above method can characterize the actual transition from a higher power stable state to a lower power stable state during the historical operation of the target base station. Compared to directly inferring the timing of low-power handover based on service load, analyzing the historical power curves themselves allows for the extraction of actual power drop events that have occurred in the target base station. This enables subsequent extraction of pre-step load trajectories and construction of step trigger conditions to be based on the actual power consumption variation patterns of the target base station.

[0030] In one embodiment, the load trajectory sample extraction module is used to receive multiple power step drop events output by the power step event identification module, determine the transition time and transition type corresponding to each power step drop event, and extract the pre-step load trajectory corresponding to each power step drop event from the historical load curve output by the historical operating curve construction module based on the transition time, as well as extract the non-triggered control trajectory corresponding to the pre-step load trajectory.

[0031] Specifically, for any power step drop event, the step drop node is determined based on the center values ​​of the platform power of the preceding and following stable power platforms corresponding to the power step drop event. In one specific implementation, the sampling point in the historical power curve where the power value first falls below the intermediate power value between the center values ​​of the preceding and following stable power platforms can be identified as the step drop node, and the timestamp corresponding to this step drop node in the historical power curve is recorded as the transition time of the power step drop event.

[0032] After determining the transition times of each power step drop event, multiple platform power center values ​​corresponding to the front and back ends of the multiple power step drop events are further extracted. The front-end platform power center value is the platform power center value of the stable power platform before the power step drop event, and the back-end platform power center value is the platform power center value of the stable power platform after the power step drop event. Based on the multiple front-end and back-end platform power center values, front-end power ranges and back-end power ranges are constructed, respectively.

[0033] In a specific implementation, the front-end power range and back-end power range can be divided into multiple front-end power intervals and back-end power intervals based on the distribution of power center values ​​of multiple front-end platforms and back-end platforms. Each power interval is used to characterize the power level before and after a power step drop event. Based on this, multiple front-end power intervals and multiple back-end power intervals are combined to construct multiple transition types. Each transition type corresponds to a combination of a front-end power interval and a back-end power interval, used to represent the historical power drop process of the target base station from a certain front-end power level to a certain back-end power level.

[0034] For each power step drop event, the transition type can be identified based on the front-end power range to which the front-end platform power center value belongs and the back-end power range to which the back-end platform power center value belongs. Thus, multiple power step drop events can be classified according to their power levels before and after the power drop, ensuring that power step drop events of the same transition type have similar front-end and back-end power levels.

[0035] Based on the transition time and transition type of the aforementioned power step drop event, the pre-step load trajectory is extracted from the historical load curve.

[0036] Specifically, for any power step drop event, using the transition time of the power step drop event as the endpoint, a historical load curve segment within a preset backtracking time is extracted, and this historical load curve segment is determined as the pre-step load trajectory corresponding to the power step drop event, representing the service load change process of the target base station before the actual power step drop occurs. Multiple power step drop events under the same transition type correspond to multiple pre-step load trajectories, thus forming a set of pre-step load trajectories corresponding to that transition type.

[0037] Furthermore, the non-triggered control trajectory corresponding to each pre-step load trajectory is extracted. Specifically, the size of the sliding control window is determined according to a preset backtracking duration, and historical load curves are traversed based on the sliding control window to obtain multiple candidate load curve segments. For each candidate load curve segment, it is determined whether the candidate load curve segment has a similar load change process to the pre-step load trajectory under the corresponding transition type. If the candidate load curve segment has a similar load change process to the pre-step load trajectory, but no power step drop event of the corresponding transition type appears in the historical power curve within a preset response time after the time corresponding to the candidate load curve segment, then the candidate load curve segment is determined as a non-triggered control trajectory.

[0038] In a specific implementation, trajectory distance can be used to determine the similarity between a candidate load curve segment and the load trajectory before the step jump. Specifically, the trajectory distance between the candidate load curve segment and the load trajectory before the step jump is calculated. The trajectory distance can be one of Euclidean distance, mean absolute distance, or dynamic time warping distance. When the trajectory distance is less than a preset similarity distance threshold, it is determined that the candidate load curve segment and the load trajectory before the step jump have a similar load change process. The aforementioned identified non-triggered control trajectory can characterize a situation in which a similar service load change process to the load trajectory before the step jump exists during historical operation, but this service load change process does not cause a corresponding power step drop.

[0039] In one embodiment, the step trigger condition construction module is used to extract candidate step condition data from multiple pre-step load trajectories corresponding to each transition type, and to perform a comparison test based on multiple non-triggered control trajectories corresponding to the transition type to construct a step trigger condition sequence for each transition type. The step trigger condition sequence is used to characterize the load change conditions that the service load trajectory of the target base station needs to meet before the front-end power level drops to the back-end power level under the corresponding transition type.

[0040] Specifically, for any transition type, an individual load upper bound value is first extracted from each pre-step load trajectory. This individual load upper bound value characterizes the low load level reached by the pre-step load trajectory before the power step drop event occurs. In one embodiment, a low load statistical interval can be set at the end of each pre-step load trajectory, and a preset quantile value of the load sampled values ​​within this low load statistical interval can be determined as the individual load upper bound value of that pre-step load trajectory. By statistically processing multiple individual load upper bound values ​​under the same transition type, the candidate low load determination upper bound corresponding to that transition type is determined.

[0041] After determining the upper bound of the candidate low load criteria, the individual effective low load duration of each pre-step load trajectory with respect to the upper bound of the candidate low load criteria is further determined. Specifically, for each pre-step load trajectory, the endpoint of the pre-step load trajectory is used as the backtracking starting point, and a continuous load interval that meets the effective low load condition is searched backwards in time, and this continuous load interval is determined as the effective low load interval.

[0042] The effective low load conditions include: the load value within the continuous load interval is mainly the load state that is not higher than the upper limit of the candidate low load judgment; if there is a load segment within the continuous load interval that exceeds the upper limit of the candidate low load judgment, then the load segment is determined as a load pulse segment, and the pulse height, pulse width and pulse area of ​​each load pulse segment are statistically analyzed. When the statistical results of the load pulse segment meet the preset pulse limit conditions, the load pulse segment is not used as the interruption point of the effective low load interval.

[0043] The aforementioned preset pulse limitation conditions can be that the pulse height of the load pulse segment does not exceed a preset height limit, the pulse width of the load pulse segment does not exceed a preset width limit, and the pulse area of ​​the load pulse segment does not exceed a preset area limit. Therefore, while allowing limited short-term load fluctuations during low-service operation of the target base station, this avoids misjudging frequent or intense load fluctuations as effective low-load conditions.

[0044] Finally, the duration of the aforementioned effective low load interval is determined as the individual effective low load duration of the load trajectory before the step transition. Statistical processing is performed on multiple individual effective low load durations under the same transition type to determine the candidate low load maintenance duration corresponding to the transition type, which represents the time condition under which the target base station service load needs to be maintained at a low load level before the power step drop event of the transition type occurs.

[0045] Based on this, individual load pulse segments of each pre-step load trajectory with respect to the upper bound of the candidate low load determination are extracted. That is, multiple load pulse segments appearing in the effective low load duration of the above-mentioned individual load segments in the pre-step load trajectory are counted. The maximum pulse height, cumulative pulse width and cumulative pulse area of ​​multiple load pulse segments are calculated to obtain the individual load pulse segments of the pre-step load trajectory.

[0046] Statistical analysis is performed on the maximum pulse height, cumulative pulse width, and cumulative pulse area of ​​multiple individual load pulse segments under the same transition type to determine the candidate low-load pulse pattern. This pattern characterizes the allowable short-term load fluctuation range in the load trajectory before the step transition without violating the power reduction condition of that transition type. In an optional implementation, the above statistical analysis can be performed by calculating the mean or median of multiple parameters. This embodiment does not specifically limit the statistical process.

[0047] After obtaining the upper bound of the candidate low load judgment, the duration of the candidate low load, and the pulse shape of the candidate low load, multiple sets of candidate step condition data are constructed based on these data.

[0048] In one implementation, the upper bound of the candidate low load determination, the duration of the candidate low load hold, and the pulse shape of the candidate low load can be used as basic condition data. The basic condition data can be adjusted by multiple ratios according to a preset tightening ratio to generate multiple sets of candidate step condition data. Specifically, tightening the upper bound of the candidate low load determination can involve lowering the upper bound; tightening the duration of the candidate low load hold can involve extending the duration of the candidate low load hold; and tightening the pulse shape of the candidate low load can involve reducing the allowable pulse height, shortening the allowable pulse width, and reducing the allowable pulse area. Through this method, multiple sets of candidate step condition data, ranging from broad to strict, can be obtained.

[0049] After generating multiple sets of candidate step condition data, a comparative test is performed on each set of candidate step condition data based on multiple pre-step load trajectories and multiple non-triggered control trajectories corresponding to that transition type. Specifically, the step triggering condition construction module matches each set of candidate step condition data with multiple pre-step load trajectories and counts the number of pre-step load trajectories that meet the candidate step condition data to calculate the successful sample matching percentage of the candidate step condition data. Simultaneously, the same set of candidate step condition data is matched with multiple non-triggered control trajectories, and the number of non-triggered control trajectories that do not meet the candidate step condition data is counted to calculate the control sample rejection percentage of the candidate step condition data. Ultimately, each set of candidate step condition data corresponds to one successful sample matching percentage and one control sample rejection percentage.

[0050] Matching parameters for each group of candidate step condition data are calculated based on the percentage of successful sample matches and the percentage of control sample rejections. In one implementation, the product of the percentage of successful sample matches and the percentage of control sample rejections, as well as the sum of the percentages of successful sample matches and the percentages of control sample rejections, can be calculated. The ratio between the product of the percentage of successful sample matches and the sum of the percentages of successful sample matches and the percentages of control sample rejections is used as the matching parameter for the candidate step condition data. When a group of candidate step condition data has a high percentage of successful sample matches and a high percentage of control sample rejections, its corresponding matching parameter is larger. The matching parameter can be used to measure the ability of candidate step condition data to distinguish between successfully triggered trajectories and non-triggered control trajectories.

[0051] Finally, based on the matching parameters corresponding to each group of candidate step condition data, the candidate step condition data with the largest matching parameters is selected from multiple groups of candidate step condition data, and a step triggering condition sequence corresponding to the transition type is constructed based on this candidate step condition data. This step triggering condition sequence includes the upper bound of the target low load judgment, the duration of the target low load, and the pulse shape of the target low load. In this way, a corresponding step triggering condition sequence can be constructed for each transition type.

[0052] In one embodiment, the real-time energy consumption control module is used to determine the current step type to be triggered and the current service load trajectory based on the current service load data and current power data of the target base station during real-time operation, and to match the current service load trajectory with the step triggering condition sequence corresponding to the step type to be triggered, and to perform real-time energy consumption optimization control on the target base station based on the matching result.

[0053] Specifically, the current service load data and current power data of the target base station can be obtained by using the same sampling period as the historical operation curve construction module. The current service load trajectory is formed based on the current service load data, and the current stable power platform of the target base station is identified based on the current power data.

[0054] When identifying the current stable power platform, the real-time power consumption control module can detect the power changes of the target base station within a preset stable detection period prior to the current moment. When the difference between the maximum and minimum power values ​​within the preset stable detection period is less than the aforementioned platform fluctuation threshold, the current power data within that time range is determined as the current stable power platform, and the power center value of the current stable power platform is calculated based on multiple power sample values ​​within the current stable power platform.

[0055] The aforementioned power center value is matched with the front-end power range of each transition type. If the power center value of the current stable power platform falls into the front-end power range of a certain transition type, it indicates that the current power state of the target base station corresponds to the power state before the historical step drop event of that transition type. Therefore, this transition type is determined as the step type to be triggered.

[0056] After determining the type of step to be triggered, the corresponding step triggering condition sequence is read. Based on the low load judgment upper bound in the step triggering condition sequence, the step critical point is identified from the current service load trajectory. The current service load trajectory can be traced back from the current moment to find the earliest sampling point that can form an effective low load interval, and this sampling point is determined as the step critical point. The time length between the step critical point and the current moment is determined as the critical load duration. In other words, if there are pulse segments in the current service load trajectory that continuously maintain a position not exceeding the low load judgment upper bound, or if there are pulse segments that exceed the low load judgment upper bound but meet preset pulse limit conditions, the duration corresponding to these continuous segments that meet the above conditions is recorded as the critical load duration. The critical pulse morphology of this continuous segment is then extracted, namely, the maximum pulse height, cumulative pulse width, and cumulative pulse area of ​​the multiple possible pulse segments.

[0057] Determine whether the critical load duration and critical pulse pattern of the current service load trajectory meet the step triggering condition sequence corresponding to the step type to be triggered. Specifically, when the critical load duration is greater than or equal to the low load holding duration in the step triggering condition sequence, and the maximum pulse height, cumulative pulse width, and cumulative pulse area in the critical pulse pattern do not exceed the pulse limit corresponding to the step triggering condition sequence, the current service load trajectory is determined to meet the step triggering condition sequence.

[0058] When the current service load trajectory meets the step triggering condition sequence corresponding to the step type to be triggered, a corresponding low-power state switching instruction is generated. The low-power state switching instruction is used to trigger the target base station to switch from the state corresponding to the front-end power range of the current step type to the state corresponding to the back-end power range. The low-power state switching instruction can carry the step type to be triggered, the front-end power range, the back-end power range, and the satisfied step triggering condition sequence, so that the energy management unit of the target base station can perform the corresponding low-power state refresh or low-power state switching according to the low-power state switching instruction.

[0059] Based on the above method, the triggering conditions derived from the historical power step decline process of the target base station can be used to identify suitable times for low-power state switching during the real-time operation of the target base station, reducing invalid triggers for low-power state switching and improving the accuracy of base station energy consumption optimization control. Through the cooperation between the above modules, this invention can utilize historical operating data collected by edge computing nodes to deduce the triggering basis for low-power switching from the actual power decline process of the target base station itself, and perform matching control based on the current service load trajectory during real-time operation. Compared with energy-saving control based solely on the current service load or fixed time periods, this invention can improve the accuracy of energy consumption control triggering timing, reduce situations where power remains at a high level under low service load, and thus improve the effectiveness of low-carbon operation control of the target base station.

[0060] The above are merely specific embodiments of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art. Parts not described in detail in this specification are prior art known to those skilled in the art.

Claims

1. A low-carbon signal base station energy consumption optimization control system based on edge computing, characterized in that, include: The historical operating curve construction module is used to acquire historical service load data and historical power data of the target base station collected by edge computing nodes, and to construct the historical load curve and historical power curve of the target base station. The power step event identification module is used to extract multiple stable power platforms from the historical power curve and identify multiple power step drop events in the historical power data based on the multiple stable power platforms. The load trajectory sample extraction module is used to determine the transition time and transition type of each power step drop event, and extract the pre-step load trajectory and non-triggered control trajectory of each power step drop event from the historical load curve based on the transition time. The step trigger condition construction module is used to extract candidate step condition data from multiple pre-step load trajectories corresponding to each transition type, and to perform comparative verification based on multiple non-triggered control trajectories corresponding to each transition type to construct a step trigger condition sequence for each transition type. The real-time energy consumption control module is used to match the current service load trajectory of the target base station with the step trigger condition sequence corresponding to the current transition type during the real-time operation of the target base station, and to perform real-time energy consumption optimization control of the target base station based on the matching result.

2. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 1, characterized in that, Extract multiple stable power plateaus from historical power curves, and identify multiple power step drop events in historical power data based on these stable power plateaus, including: The historical power curve is truncate with a preset window length. The power difference of each sliding window is calculated based on the maximum and minimum power values ​​within the sliding window. Candidate power platforms in multiple sliding windows are identified based on the power difference. Traverse multiple candidate power platforms in the historical power curve, calculate the candidate center power of each candidate power platform, and perform neighbor merging on multiple candidate power platforms based on the candidate center power to determine multiple stable power platforms in the historical power curve. Extract the platform start time, platform end time, and platform power center value for each stable power platform. Based on the platform start time, platform end time, and platform power center value, calculate the transition duration and power center value difference between any two adjacent stable power platforms. Based on the transition duration and power center value difference, identify multiple power step drop events in historical power data.

3. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 2, characterized in that, The transition time and transition type of each power step drop event are determined. Based on the transition time, the pre-step load trajectory and non-triggered control trajectory of each power step drop event are extracted from the historical load curve, including: The step drop node of the power step drop event is determined based on the platform power center value, and the timestamp corresponding to the step drop node in the historical power curve is recorded as the transition time of the power step drop event. Based on the multiple platform power center values ​​corresponding to the front and back ends of multiple power step drop events, the front-end power range and back-end power range of multiple power step drop events are constructed, and the front-end power range and back-end power range are divided into multiple front-end power intervals and back-end power intervals. Multiple front-end power ranges and back-end power ranges are combined to construct multiple transition types. Based on the front-end power range and back-end power range to which the power step drop event belongs, the transition type to which each power step drop event belongs is identified. In the historical load curve, a load curve segment within a preset backtracking time is extracted from the transition time of the power step drop event to obtain the pre-step load trajectory corresponding to each power step drop event. The sliding control window is determined based on the preset backtracking duration. The historical load curves are traversed based on the sliding control window to extract the untriggered control trajectory of each pre-step load trajectory.

4. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 3, characterized in that, Candidate step condition data were extracted from multiple pre-step load trajectories corresponding to each transition type, and comparative tests were performed based on multiple non-triggered control trajectories corresponding to each transition type. The step triggering condition sequence for each transition type was constructed, including: Extract individual load upper bound values ​​from each pre-step load trajectory corresponding to the transition type, and determine the candidate low load judgment upper bound based on multiple individual load upper bound values; Determine the individual effective low load duration for each pre-step load trajectory with respect to the upper bound of the candidate low load judgment, and determine the candidate low load holding duration based on the effective low load duration of multiple individuals; Extract individual load pulse segments of each pre-step load trajectory with respect to the upper bound of candidate low load determination, and determine the candidate low load pulse shape based on multiple individual load pulse segments; Multiple sets of candidate step condition data are constructed based on the upper bound of the candidate low load judgment, the duration of the candidate low load, and the pulse shape of the candidate low load. Based on the multiple pre-step load trajectories and non-triggered control trajectories corresponding to each transition type, calculate the successful sample matching ratio and control sample rejection ratio of each group of candidate step condition data. The matching parameters for each group of candidate step condition data are calculated based on the proportion of successful sample matching and the proportion of control sample rejection. The candidate step condition data with the largest matching parameters are selected according to the matching parameters to construct the step triggering condition sequence of the transition type.

5. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 4, characterized in that, Matching the current service load trajectory of the target base station with the step trigger condition sequence corresponding to the current transition type, and performing real-time energy consumption optimization control of the target base station based on the matching results includes: Acquire the current service load data and current power data of the target base station, identify the current stable power platform corresponding to the current power data, match the power center value of the current stable power platform with the front-end power range of each transition type, and determine the successfully matched transition type as the step type to be triggered. Based on the step triggering condition sequence corresponding to the step type to be triggered, identify the step critical point in the current service load trajectory, and extract the critical load duration and critical pulse shape based on the step critical point. Determine whether the critical load duration and critical pulse pattern of the current service load trajectory meet the step triggering condition sequence. If they do, generate a low-power state switching instruction, which is used to trigger the target base station to switch from the front-end power range of the current step type to the back-end power range.

6. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 4, characterized in that, The matching parameters for each group of candidate step condition data are calculated based on the successful sample matching rate and the control sample rejection rate, including: Calculate the product of the successful sample matching rate and the control sample rejection rate, and the sum of the successful sample matching rate and the control sample rejection rate. Use the ratio between the product of the successful sample matching rate and the control sample rejection rate and the sum of the successful sample matching rate and the control sample rejection rate as the matching parameter for the candidate step conditional data.

7. The low-carbon signal base station energy consumption optimization control system based on edge computing according to claim 3, characterized in that, Based on a sliding control window, historical load curves are traversed to extract the untriggered control trajectory for each pre-step load trajectory, including: Extract candidate load curve segments from multiple sliding control windows, calculate the trajectory distance between each candidate load curve segment and the pre-step load trajectory, and use the candidate load curve segments whose trajectory distance is less than a preset similarity distance threshold and whose corresponding power step drop event does not occur within a preset response time as the untriggered control trajectory of the pre-step load trajectory.

Images