Emergency Power Supply Dispatch Method and System for Base Stations Based on Endurance Prediction Residual

By constructing multi-dimensional operating condition feature vectors and prediction models, the lower limit of the predicted battery life of base stations is calculated, and the scheduling of power supply equipment is optimized. This solves the problem of low feasibility of emergency power supply scheduling for base stations in existing technologies and achieves more efficient emergency power supply resource management.

CN122823462BActive Publication Date: 2026-11-17CHINA TOWER CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202611307393.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2026-08-27
Publication Date
2026-11-17
Estimated Expiration
2046-08-27

AI Technical Summary

Technical Problem

In the existing emergency power supply scheduling of base stations, the deterministic processing method of battery life leads to deviations in scheduling decisions, which may result in premature resource occupation or scheduling delays, affecting the feasibility and efficiency of emergency power supply.

Method used

By constructing a multi-dimensional operating condition feature vector and combining a one-dimensional causal convolutional network and a multilayer perceptron model, the predicted battery life of the base station is predicted. Based on the sample prediction residual sequence and the guarantee level, the lower limit of the predicted battery life is calculated as a constraint on the arrival time of the power supply equipment. Multi-objective optimization is then performed to determine the target power supply equipment.

Benefits of technology

This reduces scheduling delays or premature resource occupancy caused by forecasting errors, improves the feasibility of emergency power supply scheduling and resource utilization efficiency, and lowers overall operating costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122823462B_ABST
    Figure CN122823462B_ABST
Patent Text Reader

Abstract

This invention discloses a method and system for emergency power supply scheduling of base stations based on endurance prediction residuals, relating to the field of power supply and distribution technology. Its main purpose is to address the problem of low feasibility in existing base station emergency power supply scheduling. The method primarily includes: constructing a multi-dimensional operating condition feature vector based on feature vector extraction results; predicting the endurance prediction duration of the power-outage base station based on the multi-dimensional operating condition feature vector and a time-series monitoring data matrix within a preset time period after a power outage, and determining the target operating condition based on the multi-dimensional operating condition feature vector; calculating the lower limit of the endurance prediction duration based on the sample prediction residual sequence under the target operating condition, the protection level of the power-outage base station, and the endurance prediction duration; using the lower limit of the endurance prediction duration as a constraint on the arrival time of the power supply equipment, performing multi-objective optimization with the goal of minimizing the overall operating cost of the power supply equipment, determining the target power supply equipment from the schedulable power supply equipment, and scheduling the target power supply equipment to provide emergency power to the power-outage base station.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power supply and distribution technology, and in particular to a method and system for emergency power supply scheduling of base stations based on endurance prediction residuals. Background Technology

[0002] As core nodes of mobile network infrastructure, the operational stability of communication base stations directly affects the quality of public communication and emergency response capabilities. Currently, base stations typically use on-site battery packs as the first-level backup power source after a mains power outage to cope with short-term power outages. However, the backup power duration of batteries is limited by factors such as battery capacity, service life, ambient temperature, and load size, and the actual supporting time is often quite limited, generally only a few hours. Therefore, in the event of a prolonged power outage, operators need to activate emergency power supply plans and quickly dispatch mobile power generation equipment to the site to take over the power supply.

[0003] In existing emergency power supply dispatch schemes, the battery life of base stations is typically treated as a deterministic, fixed value in the decision-making calculations for power equipment dispatch. However, this deterministic approach ignores the complexity and uncertainty of batteries under actual operating conditions. The battery life used as the basis for decisions can be too high or too low. If the battery life is too high, the actual arrival time of the equipment may be later than the actual time when the battery is depleted, causing the base station to go out of service prematurely, rendering dispatch meaningless. If the battery life is too low, it will result in the premature allocation of emergency resources at a large number of sites, not only unnecessarily occupying limited generator equipment and transportation capacity, but also potentially crowding out the dispatch window for truly emergency sites, thus leading to numerous difficulties in the actual execution of power supply dispatch. Summary of the Invention

[0004] In view of this, the present invention provides a method and system for emergency power supply scheduling of base stations based on endurance prediction residuals, the main purpose of which is to solve the problem of low feasibility of existing emergency power supply scheduling for base stations.

[0005] According to one aspect of the present invention, a base station emergency power supply scheduling method based on endurance prediction residuals is provided, comprising: In response to a power outage event of any base station, feature vectors are extracted from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and a multi-dimensional operating condition feature vector is constructed based on the feature vector extraction results. Based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, the predicted battery life of the power outage base station is predicted, and the target operating condition is determined based on the multi-dimensional operating condition feature vector. Based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station, and the predicted battery life duration, the lower limit of the predicted battery life duration of the power outage base station is calculated. The sample prediction residual sequence is constructed based on the positive prediction residual of the battery life duration of each sample base station under the target operating condition. The predicted lower limit of the battery life is used as the arrival time constraint of the power supply equipment. Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the power supply equipment. The target power supply equipment is determined from the schedulable power supply equipment, and the target power supply equipment is scheduled to provide emergency power to the power outage base station.

[0006] According to another aspect of the present invention, a base station emergency power supply scheduling system based on endurance prediction residuals is provided, comprising: The data processing module is used to respond to any power outage event of the base station, extract feature vectors from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and construct a multi-dimensional operating condition feature vector based on the feature vector extraction results. The operating condition determination module is used to predict the predicted battery life of the power outage base station based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, and to determine the target operating condition based on the multi-dimensional operating condition feature vector. The lower limit duration calculation module is used to calculate the lower limit duration of the power outage base station based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station and the duration prediction. The sample prediction residual sequence is constructed based on the duration prediction residual of each sample base station under the target operating condition. The power supply equipment screening module is used to use the predicted lower limit of the battery life as the arrival time constraint of the power supply equipment, and to perform multi-objective optimization with the goal of minimizing the overall operating cost of the power supply equipment. The module determines the target power supply equipment from the schedulable power supply equipment and schedules the target power supply equipment to provide emergency power to the power outage base station.

[0007] By employing the above-described technical solutions, the technical solutions provided by the embodiments of the present invention have at least the following advantages: This invention provides a base station emergency power supply scheduling method and system based on endurance prediction residuals. In response to a power outage event at any base station, the invention extracts feature vectors from the base station's historical operational monitoring data, battery monitoring data, and basic information. A multi-dimensional operating condition feature vector is constructed based on the feature vector extraction results. The endurance prediction duration of the base station is predicted based on the multi-dimensional operating condition feature vector and a time-series monitoring data matrix within a preset period after the power outage. The target operating condition is determined based on the multi-dimensional operating condition feature vector. Based on the sample prediction residual sequence under the target operating condition, the guarantee level of the power outage base station, and the predicted endurance duration, the lower limit of the predicted endurance duration of the power outage base station is calculated. The sample prediction residual sequence is constructed based on the positive prediction residuals of the endurance duration of each sample base station under the target operating condition. The lower limit of the predicted endurance duration is used as a constraint on the arrival time of the power supply equipment. Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the power supply equipment. A target power supply equipment is determined from the schedulable power supply equipment, and this target power supply equipment is scheduled to provide emergency power to the power outage base station. This significantly reduces scheduling delays or premature resource occupation caused by prediction errors. Simultaneously, by jointly optimizing the lower limit of the predicted endurance duration and the overall operating cost, the resource idle rate and overall operating cost in emergency power supply scheduling are effectively reduced, thereby improving the feasibility of power supply equipment scheduling.

[0008] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention, it can be implemented according to the contents of the specification. Furthermore, in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0009] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings: Figure 1 The following is a flowchart of an emergency power supply scheduling method for base stations based on endurance prediction residuals provided by an embodiment of the present invention; Figure 2 This invention provides a flowchart for calculating the lower limit of battery range prediction according to an embodiment of the present invention. Figure 3 A flowchart of an adaptive update method for residual quantiles provided by an embodiment of the present invention is shown; Figure 4 This invention provides a flowchart of a multi-objective optimization process according to an embodiment of the invention. Figure 5The diagram shows a block diagram of a base station emergency power supply scheduling system based on endurance prediction residuals provided by an embodiment of the present invention. Detailed Implementation

[0010] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0011] To address the issue of low feasibility in existing emergency power supply scheduling for base stations, this invention provides a base station emergency power supply scheduling method based on endurance prediction residuals, such as... Figure 1 As shown, the method includes steps 101-104: 101. In response to a power outage event of any base station, feature vectors are extracted from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and a multi-dimensional operating condition feature vector is constructed based on the feature vector extraction results.

[0012] In this embodiment of the invention, the executing entity is the server of the emergency power supply dispatch system for communication base stations, which can be a local server or a cloud server; this embodiment of the invention is not limited to any particular server. When any communication base station within the jurisdiction issues a power outage warning, the current executing entity responds to the power outage event of this base station by acquiring the operational monitoring data, battery monitoring data, and basic information of the base station itself for a historical period prior to the time of power outage (including the time of power outage). The operational monitoring data and battery monitoring data are both monitoring data collected in real time and stored sequentially by the data acquisition module in the system. The operational monitoring data is data collected during the operation of the base station itself, such as communication load, energy consumption data, operating voltage, operating current, and ambient temperature; the battery monitoring data is data collected on the working status of the batteries equipped in the base station, such as voltage dips and discharge currents at the moment of power outage, and historical charge and discharge records. The basic information includes information characterizing the attributes of the base station, such as battery type, nominal capacity, years of operation, site identification, geographical location, and protection level. After acquiring the above data, it is also necessary to preprocess the data, for example, by removing obviously invalid records, unifying the sampling time, and marking missing fields. In real-world applications, multiple base stations may experience simultaneous power outages due to area-wide power failures. In other words, there can be multiple base stations experiencing power outages. During the scheduling process, the scheduling decision-making process is the same for each base station experiencing power outages. Therefore, the scheduling process will be explained using only one base station experiencing power outages as an example.

[0013] To transform the aforementioned monitoring data and static attribute information into directly usable numerical representations for subsequent prediction and operational condition identification, feature extraction was performed on each of the data. Specifically, for operational monitoring data, time-series statistical features over historical periods were extracted, including average communication load and average energy consumption. For battery monitoring data, key indicators such as voltage drop amplitude and voltage recovery slope during the initial discharge phase were extracted, and combined with the number of cycles in historical charge-discharge records, to comprehensively reflect the battery's actual discharge capacity and degree of health degradation under current operating conditions. For basic information, categorical attributes such as battery type and warranty level, and numerical attributes such as nominal capacity and service life were individually hot-coded and normalized using an encoding mapping method to form static feature vectors representing the base station's baseline conditions. Subsequently, the above three types of feature vectors were concatenated and aligned along the feature dimensions, and the dimensional differences between the dimensions were eliminated through standardization, ultimately constructing a unified multi-dimensional operational condition feature vector. This feature vector collaboratively characterizes the complete operating environment of the base station at the time of power outage from three levels: load timing characteristics, real-time energy storage status, and inherent site attributes. It provides a comprehensive, structured, and computable data foundation for subsequent evaluation and prediction of battery life and operating condition identification.

[0014] In one embodiment of the present invention, for further explanation and limitation, feature vectors are extracted from the operation monitoring data, battery monitoring data, and basic information of the power outage base station during historical periods. A multi-dimensional operating condition feature vector is constructed based on the feature vector extraction results, including: Extract the operation monitoring data of the power outage base station within a series of preset historical days before the current power outage time, and within the historical sliding window corresponding to the current power outage time each day. Calculate the average operating energy consumption and standard deviation of operating energy consumption corresponding to each historical sliding window. Based on the average operating energy consumption, standard deviation of operating energy consumption corresponding to different historical sliding windows, and the instantaneous load at the time of the power outage, calculate the tidal load deviation characteristics. Extract charge and discharge data for a preset number of times from the battery monitoring data, and calculate the average voltage drop slope and the estimated transient internal resistance based on the charge and discharge data. Use the average voltage drop slope and the estimated transient internal resistance as dynamic characteristics of battery degradation. The nominal capacity and years of operation are extracted from the aforementioned basic information as basic features; A multi-dimensional operating condition feature vector is constructed based on the tidal load deviation characteristics, the battery degradation dynamic characteristics, and the basic characteristics.

[0015] In this embodiment of the invention, using the current power outage time as a benchmark, operational monitoring data of the power outage base station is extracted from the historical sliding windows corresponding to the current power outage time for each day within a preset number of historical days (e.g., 7 calendar days) prior to the current power outage time. The average operating energy consumption and standard deviation of operating energy consumption corresponding to each historical sliding window are then calculated. Based on the average operating energy consumption, standard deviation of operating energy consumption corresponding to different historical sliding windows, and the instantaneous load at the time of the power outage, tidal load deviation characteristics are calculated to characterize the degree of fluctuation and deviation of the base station load at the same time on different days. The tidal load deviation of base station i is then described. The calculation formula is expressed as: ;in, Indicates the current instantaneous load. Indicates the standard deviation of operating energy consumption. This represents the average operating energy consumption. This feature, expressed in numerical form, precisely indicates whether the power outage occurred during a peak period of 5G service activity.

[0016] Secondly, charge-discharge data for a preset number of cycles is extracted from the battery monitoring data. Based on this data, the average voltage drop slope and estimated transient internal resistance are calculated, and these two values ​​are used as dynamic characteristics of battery degradation to reflect the actual aging degree and discharge capacity decline of the battery at the current stage. The dynamic characteristics of battery degradation include the average voltage drop slope and the estimated transient internal resistance. The average voltage drop slope... The calculation formula is expressed as: Where N represents the number of complete charge-discharge cycles included in this calculation. This indicates that during the nth complete charge and discharge process, the discharge begins at... voltage value, This indicates that during the nth complete charge and discharge process, the time at which the discharge ends... The voltage value. Estimated transient internal resistance value. The calculation formula is expressed as: ;in, This indicates the momentary voltage drop after the current power outage occurs. This represents the discharge current instantaneously after a power outage. The mean voltage drop slope and the estimated transient internal resistance mathematically constitute a dimension-reduced mapping of the battery's true health, which can more sensitively capture the degree of battery aging under micro-circulation compared to relying solely on the number of years in operation.

[0017] At the same time, the nominal capacity is extracted from the basic information. and years of operation As fundamental characteristics, these characterize the inherent configuration conditions of the base station energy storage system. Finally, the aforementioned tidal load deviation characteristics, battery degradation dynamic characteristics, and fundamental characteristics, along with ambient temperature, are analyzed. By splicing and fusing the data, a multi-dimensional feature vector of the power outage base station is obtained, i.e. This feature vector comprehensively depicts the complete operating environment of the base station at the moment of power outage from three dimensions: load fluctuation characteristics, battery aging status, and base station performance configuration, providing a data foundation for subsequent operating condition clustering and battery life prediction.

[0018] 102. Based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, predict the predicted battery life of the power outage base station, and determine the target operating condition based on the multi-dimensional operating condition feature vector.

[0019] In this embodiment of the invention, the multi-dimensional operating condition feature vector serves two purposes: first, it acts as a static input feature for predicting battery life, using a pre-trained model to predict the expected battery life of the current power outage base station, i.e., the predicted battery life; second, it serves as the basis for identifying the operating condition of the current power outage base station, identifying the target operating condition that best matches the current base station's operating status from a pre-constructed set of operating conditions through feature matching. The time-series monitoring data matrix refers to a two-dimensional matrix formed by extracting continuous time-series monitoring data within a preset time period after the power outage, starting from the moment of the power outage, and aligning them according to the time step. The time-series monitoring data includes at least two dimensions: operational monitoring data and battery monitoring data within that time period, used to characterize the transient response process of the base station load and battery in the initial stage of discharge.

[0020] In one embodiment of the present invention, for further explanation and limitation, determining the target working condition based on the multi-dimensional working condition feature vector includes: Retrieve pre-built clusters of operating conditions; Calculate the distance between the multidimensional working condition feature vector and the center of each working condition cluster, and take the working condition corresponding to the working condition cluster with the smallest distance as the target working condition of the power outage base station; The method further includes: For any operating condition cluster, the historical multidimensional operating condition feature vector of each sample base station in the cluster is updated at a preset time interval, and the offset distance between the updated historical multidimensional operating condition feature vector and the center of the cluster is calculated. If the offset distance of any sample base station is greater than the preset operating condition offset distance threshold, the sample base station is reassigned to the cluster with the smallest distance to realize operating condition migration.

[0021] In this embodiment of the invention, the operating condition clusters are obtained by clustering the historical multi-dimensional operating condition feature vectors of all sample base stations within the target scheduling area. That is, cluster analysis is performed in advance based on the historical multi-dimensional operating condition feature vectors of all sample base stations within the target scheduling area to construct several operating condition clusters. Each cluster represents a typical operating condition with similar load characteristics, energy storage status, and basic attribute conditions. Specifically, density clustering algorithms can be used to perform cluster analysis on the historical multi-dimensional operating condition feature vectors to obtain... individual working condition clusters Each cluster corresponds to a central vector. and covariance matrix .

[0022] When it is necessary to determine the operating condition of the current power outage base station, the pre-constructed operating condition clusters are retrieved, the distance between the multi-dimensional operating condition feature vector of the base station and the center of each cluster is calculated, and the operating condition corresponding to the cluster with the smallest distance is determined as the target operating condition of the power outage base station, so that the subsequent prediction residual statistical inference based on historical samples under the target operating condition has the same comparability.

[0023] To adapt to long-term operating condition drift phenomena such as seasonal fluctuations in base station load, battery aging processes, and capacity expansion upgrades, for each operating condition cluster, the latest historical multidimensional operating condition feature vectors of each sample base station within the cluster are reacquired at preset time intervals, such as monthly or quarterly, and the offset distance between the vector and the center of the cluster is calculated. Mahalanobis distance is used to calculate the current multidimensional operating condition feature vector of the base station. Distance to the center of each operating condition cluster : .

[0024] Define operating condition drift threshold If the overall offset distance of samples within any cluster exceeds the preset operating condition offset distance threshold, that is, the current vector is at a distance from its original cluster. distance Then the base station Reassigned to the target cluster with the minimum Mahalanobis distance : And assign new dynamic operating condition labels. This mechanism ensures the consistency of residual distribution statistics from the underlying mathematical logic, thus providing a consistent operating condition reference benchmark for subsequent range prediction and scheduling decisions.

[0025] In one embodiment of the present invention, for further explanation and limitation, the prediction of the remaining battery life of the power outage base station based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage includes: Based on the one-dimensional causal convolutional network, feature extraction is performed on the time-series monitoring data matrix to obtain time-series discharge features. Based on the multilayer perceptron, feature extraction is performed on the multi-dimensional operating condition feature vector to obtain structured battery performance features. The time-series discharge characteristics and the structured battery performance characteristics are spliced ​​together by the fusion network, and the spliced ​​fusion characteristics are mapped by the fully connected layer to obtain the predicted driving range.

[0026] In this embodiment of the invention, the prediction of battery life is achieved using a dual-stream architecture battery life prediction model. The trained dual-stream architecture battery life prediction model includes a one-dimensional causal convolutional network, a multilayer perceptron, a fusion network, and a fully connected layer. The one-dimensional causal convolutional network and the multilayer perceptron are two parallel branches. The one-dimensional causal convolutional network is a dynamic temporal branch used to process dynamic temporal data, i.e., to extract features from the temporal monitoring data matrix to output a dynamic latent vector characterizing the temporal discharge features. The multilayer perceptron is a static feature stream used to process static structured features and output static latent vectors that characterize the performance features of the structured battery. The above vectors are concatenated through a fusion network and then mapped through a fully connected layer to output the predicted battery life. , represented as ;in, This represents vector concatenation. Both represent the weight parameter matrix obtained from model training.

[0027] The training process of the dual-stream architecture battery life prediction model includes: First, collecting historical power outage event data from each base station within the target scheduling area, with each historical power outage event serving as a training sample. The training sample includes a time-series monitoring data matrix for the first K minutes after the power outage, the corresponding static structured feature vector, and the actual battery life label for that power outage event. The time-series monitoring data matrix is ​​normalized, and the categorical features in the static structured feature vector are one-hot encoded, while the numerical features are standardized to eliminate dimensional differences. The processed training samples are then divided into training and validation sets according to a preset ratio. Mean squared error is used as the loss function, with the training objective being to minimize the error between the actual battery life and the model's predicted value. The backpropagation algorithm and the Adam optimizer are used to jointly optimize all weight parameters in the one-dimensional causal convolutional network, multilayer perceptron, fusion network, and fully connected layer. An early stopping strategy is employed during training; training is terminated when the validation set loss no longer decreases after several consecutive iterations to prevent overfitting. After training, the weight parameter matrices are... It is stored in a fixed location for use during the online inference phase.

[0028] 103. Based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station, and the predicted endurance duration, calculate the lower limit of the predicted endurance duration of the power outage base station.

[0029] In this embodiment of the invention, after determining the current operating condition category of the power outage base station, the target operating condition of the power outage base station is used as the retrieval condition. Several sample base stations under the same or similar operating conditions are selected from the historical sample database, along with the positive prediction residuals between the model-predicted battery life and the actual battery life of each sample base station. These individual base station positive prediction residuals are arranged in ascending order to form a sample prediction residual sequence. The lower limit of the battery life prediction is used to characterize the earliest time boundary at which the battery may be exhausted under the current prediction confidence level. Since the base station protection level is also introduced in the calculation process, using the lower limit of the battery life prediction as the scheduling basis can reserve sufficient response windows for high-priority base stations and avoid prematurely allocating resources to low-risk sites, thus ensuring that the scheduling of power supply equipment is truly based on reliable and controllable data.

[0030] In this process, the predicted battery life duration is not directly used as the basis for scheduling power supply equipment, but rather as an intermediate reference for calculating the lower limit of the predicted battery life duration in subsequent decisions. Since a single predicted battery life duration is essentially a point estimate given by the model under ideal or average conditions, it cannot reflect the width of the confidence interval of the prediction result, nor can it distinguish the differentiated safety margins that different base stations should enjoy due to differences in data quality, similarity of operating conditions, or protection levels. If this point estimate is directly used for scheduling decisions, whether the predicted value is too high or too low, it may lead to a timing mismatch between the equipment arrival time and the actual battery depletion time, making the scheduling scheme lack sufficient fault tolerance when dealing with prediction uncertainties. Therefore, by reducing the predicted battery life duration to an intermediate variable, the lower limit of the predicted battery life duration, which integrates the statistical characteristics of the prediction residuals and the base station protection level information, is used as the final scheduling criterion.

[0031] In one embodiment of the present invention, for further illustration and limitation, such as Figure 2 As shown, based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station, and the predicted endurance duration, the lower limit of the predicted endurance duration of the power outage base station is calculated, including: 1031. Based on the protection level, match the allowable power loss risk level parameter, and use the complement of the allowable power loss risk level parameter as the quantile.

[0032] 1032. Extract residual quantiles from the predicted residual sequence of the sample based on the quantiles.

[0033] 1033. Calculate the difference between the predicted range duration and the residual quantile. If the difference is greater than the preset lower limit duration threshold, the difference is used as the lower limit duration of the predicted range. If the difference is less than or equal to the preset lower limit duration threshold, an abnormal warning message for model prediction is generated and fed back.

[0034] In this embodiment of the invention, the protection level is pre-configured with the permissible level of power outage risk. The higher the protection level, The smaller the value, the more negatively correlated the protection level is with the permissible power outage risk level parameter. For example, the permissible power outage risk level parameter for core aggregation stations can be 0.01, for important macro stations 0.05, and for ordinary stations 0.10. This mapping relationship is configured by the power protection strategy and remains determined within the same scheduling cycle. This further involves using the complement of the permissible power outage risk level parameter. As a quantile, the corresponding residual quantile is extracted from the sample prediction residual sequence under the target operating condition. This quantile reflects the maximum positive deviation that the model prediction may have under the allowable risk level, that is, the degree to which the predicted value is higher than the actual battery life. Based on this, the difference between the predicted battery life duration and the residual quantile is calculated. If the difference is greater than the preset lower limit duration threshold, the difference is taken as the lower limit duration of the battery life prediction for this base station. This value defines, statistically, the earliest time boundary at which the battery may be exhausted under the current prediction credibility and guarantee requirements. If the difference is less than or equal to the preset lower limit duration threshold, it indicates that there is a significant conflict between the current prediction result and the deviation distribution of historical samples under similar operating conditions. The prediction reliability of the model is no longer sufficient to support the generation of subsequent scheduling criteria. In this case, an abnormal model prediction warning is generated and fed back to remind the operation and maintenance personnel to intervene manually or activate the backup guarantee plan, so as to ensure that the safety bottom line of emergency response is not breached when the prediction model fails.

[0035] The preset lower limit duration threshold can be 0 or a number close to 0. When the preset lower limit duration threshold is 0, the formula for the predicted lower limit duration of battery life is expressed as: ;in, This indicates the lower limit of the predicted battery life. Indicates the predicted battery life. Indicates working condition Below, the permissible power outage risk level parameter is: The residual quantiles at that time. It should be noted that the max function is only used to implement the lower bound constraint that the difference takes the value of 0. The triggering logic for the anomaly warning exists independently of this formula. When the result of the difference calculation is less than or equal to 0, an anomaly warning information for the model prediction is generated and fed back synchronously. This warning does not participate in the formula calculation, but is used as an independent alarm mechanism to remind the operation and maintenance personnel that the current prediction result may have a serious deviation and manual intervention is required for verification.

[0036] It is important to note that the prediction residuals are positive, indicating that the predicted values ​​are overestimated, meaning the model tends to make optimistic estimates, which corresponds to scenarios with higher scheduling risks. After determining the target operating condition to which the power outage base station belongs, the residual quantiles at corresponding positions are extracted from the ascending sample prediction residual sequence corresponding to that operating condition. These quantiles statistically quantify the possible extent of overestimation under this type of operating condition, providing an empirical bias reference for subsequent calculation of the lower limit of the predicted battery life.

[0037] In one embodiment of the present invention, for further illustration and limitation, such as Figure 3 As shown, the method further includes: 201. If the number of samples in the sample prediction residual sequence under the target working condition is less than the preset sample number threshold, or the missing proportion of key fields in the historical sample data on which the sample prediction residual sequence is based is greater than the preset proportion threshold, then the sample penalty coefficient is retrieved according to the sample number or the missing proportion.

[0038] 202. The allowable power loss risk level parameter is negatively corrected based on the sample penalty coefficient, and the complement of the corrected allowable power loss risk level parameter is used as the new quantile. The process is then returned to the step of extracting residual quantiles from the sample predicted residual sequence based on the quantiles, so as to achieve adaptive updating of residual quantiles.

[0039] In this embodiment of the invention, to ensure the accuracy of residual quantiles, before extracting residual quantiles from the sample predicted residual sequence based on quantiles, the allowable power loss risk level parameter is corrected according to the sample size of the sample predicted residual sequence under the target operating condition and the field missingness of the data on which the sequence is constructed. If the sample size of the sample predicted residual sequence under the target operating condition is less than a preset sample size threshold, or the missing proportion of key fields in the historical sample data on which the sample predicted residual sequence is constructed is greater than a preset proportion threshold, it indicates that the statistical basis of the sample under the current target operating condition is insufficient or the data quality is poor. The corresponding sample penalty coefficient is retrieved according to the sample size or missing proportion. The smaller the sample size and the higher the missing proportion, the larger the penalty coefficient value. Subsequently, the allowable power loss risk level parameter is negatively corrected according to the sample penalty coefficient, that is, the allowable power loss risk level parameter is subtracted from the sample penalty coefficient, so that the corrected allowable power loss risk level parameter is lower than the original parameter. The complement of the corrected allowable power loss risk level parameter is used as the new quantile, and the step of extracting residual quantiles from the sample predicted residual sequence based on quantiles is returned to execute, so as to realize the adaptive update of residual quantiles according to sample quality. Through the above correction method, when the sample size is insufficient or the data quality is poor, the quantiles automatically shift towards the higher end of the residual distribution, thereby making the calculated lower limit of the range prediction more conservative. This compensates for the decrease in prediction reliability caused by insufficient statistical information, ensuring that sufficient scheduling safety margin can still be maintained even under conditions with weak data foundation. The preset sample size threshold can be set according to the confidence requirement of the residual statistics under the target operating condition, for example, it can be set to 30. The preset proportion threshold can be set according to the influence of key fields on the range prediction, for example, it can be set to 20%.

[0040] In one embodiment of the present invention, for further explanation and limitation, the method further includes: Under the circumstances of normal power restoration, power supply equipment access, or battery undervoltage reaching the protection threshold, the actual battery support time is obtained, and the difference between the predicted battery life and the actual battery support time is used as the prediction residual of the power outage base station. The residuals of the power outage base stations are used as incremental samples to update the sample prediction residual sequence of the target operating condition, so that the lower limit of the battery life prediction can be calculated based on the updated sample prediction residual sequence during subsequent power supply scheduling.

[0041] In this embodiment of the invention, when a power outage occurs due to normal power restoration, successful connection of power supply equipment, or battery undervoltage reaching a protection threshold causing the base station to power down, the actual battery life during that power outage event is recorded. Incremental samples are then constructed using the predicted battery life for that power outage event and the actual battery life. The prediction residual for any sample base station... ;in, To correspond to the actual support duration of historical power outages, incremental samples are constructed and then updated into the sample prediction residual sequence for the target operating condition. The sample prediction residual sequence is maintained using a sliding window approach; when the sequence length reaches a preset upper limit, the oldest historical sample is replaced with a new sample, achieving a rolling update of the residual sequence. Through this method, the sample prediction residual sequence can continuously absorb prediction deviation information from the latest power outage events, adaptively tracking the impact of factors such as battery aging, load changes, and operating condition drift on prediction accuracy. This ensures that the quantiles and lower limit of the range prediction calculated based on this sequence always reflect the model's true prediction error distribution in the current time period, thereby guaranteeing that the statistical boundaries upon which scheduling decisions are based are continuously optimized as the system operates. Furthermore, this process updates the prediction lower limit calibration parameters, eliminating the need to retrain the range prediction model after each event.

[0042] 104. Using the predicted lower limit of the battery life as the arrival time constraint of the power supply equipment, perform multi-objective optimization with the goal of minimizing the overall operating cost of the power supply equipment, determine the target power supply equipment from the schedulable power supply equipment, and schedule the target power supply equipment to provide emergency power to the power outage base station.

[0043] In this embodiment of the invention, dispatchable power supply equipment refers to various emergency power supply resources that are currently available, can be uniformly allocated by the dispatching platform, and have the capability for mobile deployment. Specifically, these can include portable mobile generators of different power levels, towed diesel generator vehicles, and new energy storage emergency power supply vehicles. The attributes of the power supply equipment include its current location, such as storage site or parking location identification, rated output power, fuel reserves or remaining power, travel time and on-site deployment time, and its own operational status, such as whether it is on standby, under maintenance, or already deployed. These attributes collectively determine whether the equipment can reach the target base station within the specified arrival time constraint and meet its power requirements.

[0044] After calculating the predicted lower limit of the power outage base station's remaining power, the scheduling decision uses this duration as a hard constraint. That is, the cumulative time from receiving the scheduling instruction to arriving at the power outage base station and becoming capable of supplying power must not exceed this lower limit to ensure that power is supplied before the battery runs out. Based on this, multi-objective optimization is performed with the goal of minimizing the overall operating cost of the power supply equipment. The overall operating cost includes at least the equipment scheduling and transportation costs such as fuel consumption, task switching costs, and unallocated penalties. The optimization process involves traversing all schedulable power supply equipment. Under the premise of satisfying the arrival time constraint and ensuring that the equipment's output power is not lower than the base station's load requirements, the power supply equipment with the optimal overall operating cost is selected as the target power supply equipment through multi-objective trade-offs and ranking. A scheduling instruction is then issued to the target power supply equipment to execute the emergency power supply task.

[0045] By using the lower limit of the predicted battery life as a hard time boundary for scheduling triggering, and combining it with multi-objective cost optimization, the power supply scheduling can simultaneously meet the dual objectives of reliable arrival before battery depletion and economic rationality in resource consumption. This achieves a global deployment balance of emergency resources while ensuring the continuity of base station power supply.

[0046] In one embodiment of the present invention, for further illustration and limitation, such as Figure 4 As shown, the predicted lower limit of the range is used as a constraint on the arrival time of the power supply equipment. Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the power supply equipment, including: 1041. Based on the expected power supply travel time, connection start-up time, and safety buffer time of each schedulable power supply device, as well as the predicted lower limit of the endurance time, calculate the safety margin of each schedulable power supply device, and select the schedulable power supply devices whose safety margin meets the preset margin threshold as candidate power supply devices.

[0047] 1042. Multi-objective optimization is performed with the travel time, fuel consumption, task switching cost, and power outage risk cost of the candidate power supply equipment as optimization objectives to obtain the target power supply equipment.

[0048] In this embodiment of the invention, after determining each schedulable power supply device, the total expected time from the issuance of the scheduling command to the actual start of power supply for each device is calculated by superimposing the expected power supply journey time, wiring start time, and preset safety buffer time for each schedulable power supply device. The total expected time is then subtracted from the predicted lower limit of the battery life to obtain the safety margin for each schedulable power supply device. The expected power supply journey time includes the travel time from the current location to the target base station; the wiring start time includes the on-site operation time such as cable connection, power-on preheating, and parallel operation preparation after the device is in place; the preset safety buffer time is used to cope with uncertainties such as road condition fluctuations or on-site operation delays. The preset margin threshold can be 0. For example, a generator... and power outage base stations As a combination that requires evaluation of safety margins, the calculation is performed from the current moment. Travel time required from start to finish connecting the generator to power supply Wiring start time and safety buffer time The lower bound of the base station's predicted power outage time is the actual power outage time. The sum of the predicted lower limit of battery life and the result of the summation. Then this combination must satisfy: .

[0049] Therefore, the safety margin is defined as .when When this combination is ready, it can proceed to the next multi-objective optimization operation; when If the combination is removed, the reason is recorded. Unlike simply lowering the scheduling score, the removed combination cannot participate in the current round of task allocation. If a base station has no feasible combinations, the system outputs a resource shortage alarm, which may trigger cross-regional calls, power-down of non-critical loads, or manual intervention.

[0050] Based on this, a multi-objective optimization function is constructed using each candidate power supply device as a decision variable. This multi-objective optimization function includes at least the following optimization dimensions: travel time cost reflecting scheduling response speed; fuel consumption cost reflecting economic efficiency; and task switching cost to suppress execution chaos caused by frequent rescheduling, which characterizes the resource adjustment cost required to reassign the device from its original task to the current base station. Furthermore, the optimization dimension may also include the downtime risk cost of a power outage base station, which characterizes the loss due to power outages determined by the base station's protection level and service importance. The objective function is expressed as: ; in, This indicates that base station i did not obtain feasible resources in this round; The cost of a site outage is determined based on the level of protection and the impact on business operations. This indicates that power supply equipment j is allocated to base station i; Indicates the time cost of the trip; This indicates the cost of fuel consumption; This represents the task switching cost. The solver determines the device with the optimal overall cost from the candidate power supply devices as the target power supply device and issues a scheduling command. The solver can employ integer programming, the Hungarian algorithm, heuristic search, or rule-based solutions that satisfy constraints. Regardless of the solver used, the safe arrival constraint is treated as a hard constraint and executed before cost. Through the above objective optimization, while satisfying the hard constraint of safe arrival, it balances response time, economic cost, scheduling stability, and business assurance requirements, achieving refined scheduling decisions for emergency power supply resources.

[0051] In one embodiment of the present invention, for further explanation and limitation, in the case of multiple power outage base stations at the same time, in response to the occurrence of any rolling rescheduling trigger event, rolling rescheduling of the power supply equipment is performed. The rolling rescheduling process includes: Based on the real-time power supply information of the power supply equipment, the power supply scheduling task is divided into switchable tasks and non-switchable tasks; The non-switchable tasks are frozen, and multi-objective optimization is performed on the switchable power supply equipment and the power outage base station to be rescheduled for the non-frozen tasks, so as to determine the updated power supply equipment from the switchable power supply equipment for each of the rescheduled power outage base stations.

[0052] In this embodiment of the invention, when multiple base stations are simultaneously experiencing power outages, continuous monitoring of multi-source dynamic information such as power outage events, battery life prediction, road traffic, and generator status is conducted. When any rolling rescheduling trigger event occurs, a rolling rescheduling process is initiated to re-coordinate and optimize the power supply schemes for all currently out-of-power base stations, ensuring that the scheduling scheme remains dynamically adaptable to real-time situational changes. Rolling rescheduling trigger events include: the addition of a new out-of-power base station (i.e., a new out-of-power base station is connected), restoration of normal power supply (i.e., the mains power supply on which the base station was originally based is restored), battery life prediction lower limit duration being less than a lower limit duration threshold, target power supply equipment arrival delay exceeding a preset delay threshold, base station security level updates, and target power supply equipment malfunctions, such as generator failure or insufficient fuel. The lower limit duration threshold represents the minimum duration that no current power supply equipment can meet, and the preset delay threshold represents the maximum allowable delay for power supply equipment to arrive later than the set duration. Both thresholds can be customized based on actual scheduling experience. Base station security level updates affect the battery life prediction lower limit duration; therefore, rescheduling is also required when the level is updated.

[0053] During the scheduling process, based on the real-time power supply information of each power supply device, the generated power supply scheduling tasks are divided into switchable and non-switchable tasks. If a power supply device has already been connected and is continuously generating power, or if it has not yet been connected but its distance from the target base station is less than a preset distance threshold and forced reassignment would cause the target base station to face a power supply gap and a higher risk of power outage, then the task currently being executed by that device is marked as a non-switchable task and frozen, and will no longer participate in the resource reallocation of this rescheduling, in order to ensure the stability and continuity of existing power supply tasks. The tasks corresponding to other devices that have not yet been actually connected and whose reassignment will not cause an urgent power outage risk are marked as switchable tasks, and the power supply devices corresponding to these tasks are released from the original scheduling tasks and included in the available resource pool of this rolling rescheduling.

[0054] Subsequently, for the base stations awaiting rescheduling that are not yet covered by the frozen tasks, and using the calculated lower limit of the predicted endurance time for each base station as a hard arrival time constraint, switchable power supply equipment capable of reliably reaching the target base station before the battery runs out is selected from the available resource pool, constructing a candidate power supply equipment set for each base station. Based on this, for each base station awaiting rescheduling, multi-objective optimization is performed with the goal of minimizing the overall operating cost of the power supply equipment. This comprehensively considers fuel consumption converted from transportation distance, power generation cost per unit power supply time, and the impact of equipment availability on the overall balance of emergency reserves. Updated target power supply equipment is then determined from the candidate equipment, and dispatch instructions are issued.

[0055] In addition, to suppress repeated task switching of equipment caused by frequent rescheduling, a reassignment cost penalty term can be introduced into the objective optimization function, or a minimum task hold time can be set for equipment that is already in a near-ready state, so as to ensure the agility of emergency response while taking into account the actual executability of the scheduling scheme and the stability of resource utilization.

[0056] In practical applications, such as provincial power supply platforms, data on base station power outages, batteries, loads, temperatures, and alarms can be read from the operation and maintenance monitoring system via interfaces. Road travel times can be read through map services, and generator location, fuel, power, and task status can be read through the generator management system. The system generates tasks based on predicted range duration and lower limit range prediction, and these tasks are then distributed via maintenance personnel terminals or generator control terminals. An example of risk mapping is shown in Table 1 below.

[0057] Table 1:

[0058] Assuming the current calculation time is 10:30, there are four power outage base stations B1-B4 and three available mobile generators G1-G3. The endurance prediction model has provided the predicted endurance duration, and the corresponding residual quantiles can be found based on the operating condition group and support level. Details are shown in Table 2 below.

[0059] Table 2:

[0060] For example, the lower limit of the B1's predicted range is 180. 40 = 140 minutes, the predicted lower limit of power outage time is 10:00 + 140 = 12:20. Calculate the safety margin for each generator-base station combination. As shown in Table 3 below.

[0061] Table 3:

[0062] Taking G1→B1 as an example, the safety margin is 12:20. 10:30 60 15 15 = 20 minutes, therefore, this combination is feasible. Taking G2→B1 as an example, the safety margin is 12:20. 10:30 100 15 15= The time limit is 20 minutes, therefore this combination must be removed from the feasible set. Assuming the scheduling objective is to first minimize the level-weighted risk of unsecured sites, then minimize travel time, and prioritize those with smaller safety margins, the initial allocation is G1→B4, G2→B2, G3→B1; B3 entered the waiting set relatively late due to its predicted lower limit of power outage. At 10:40, a mains power restoration event is received for B4. The G1→B4 task has not yet entered the uninterruptible phase and can therefore be cancelled. Simultaneously, the road service increases the travel time of G3→B1 by 25 minutes, and its safety margin changes from 5 minutes to... 20 minutes. The system freezes the G2→B2 task, which has entered the uninterruptible phase, releases G1 and G3, and regenerates the remaining combinations. If the new estimated journey from G1 to B1 at 10:40 is 45 minutes, then the new safety margin is 12:20. 10:40 45 15 15 = 25 minutes, therefore G1 is reassigned to B1; G3 is released, and B3 continues to be included in subsequent scheduling. This result demonstrates a local rolling rescheduling triggered by both power restoration and traffic delays.

[0063] Furthermore, as a response to the above Figure 1 The implementation of the method shown in this invention provides a base station emergency power supply scheduling system based on endurance prediction residuals, such as... Figure 5 As shown, the system includes: Data processing module 31 is used to respond to any power outage event of a base station by extracting feature vectors from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and constructing a multi-dimensional operating condition feature vector based on the feature vector extraction results. The operating condition determination module 32 is used to predict the predicted battery life of the power outage base station based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, and to determine the target operating condition based on the multi-dimensional operating condition feature vector. The lower limit duration calculation module 33 is used to calculate the lower limit duration of the power outage base station based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station and the duration prediction duration. The sample prediction residual sequence is constructed based on the duration prediction residual of each sample base station under the target operating condition. The power supply equipment screening module 34 is used to use the predicted lower limit of the battery life as the arrival time constraint of the power supply equipment, perform multi-objective optimization with the goal of minimizing the comprehensive operating cost of the power supply equipment, determine the target power supply equipment from the schedulable power supply equipment, and schedule the target power supply equipment to provide emergency power supply to the power outage base station.

[0064] Furthermore, the system also includes a rescheduling module, which is used to perform rolling rescheduling of power supply equipment in response to the occurrence of any rolling rescheduling trigger event when multiple power outage base stations exist simultaneously. The rolling rescheduling trigger event includes adding a new power outage base station, restoration of normal power supply, the predicted lower limit of the battery life duration being less than the lower limit of the duration threshold, the arrival delay of the target power supply equipment being greater than the preset delay threshold, the base station protection level being updated, and the target power supply equipment operating abnormally. The rolling rescheduling process includes: Based on the real-time power supply information of the power supply equipment, the power supply scheduling task is divided into switchable tasks and non-switchable tasks; The non-switchable tasks are frozen, and multi-objective optimization is performed on the switchable power supply equipment and the power outage base station to be rescheduled for the non-frozen tasks, so as to determine the updated power supply equipment from the switchable power supply equipment for each of the rescheduled power outage base stations.

[0065] Furthermore, the operating condition determination module 32 includes: The retrieval unit is used to retrieve a pre-constructed working condition cluster, wherein the working condition cluster is obtained by clustering the historical multidimensional working condition feature vectors of all sample base stations in the target scheduling area; The first calculation unit is used to calculate the distance between the multi-dimensional working condition feature vector and the center of each working condition cluster, and to take the working condition corresponding to the working condition cluster with the smallest distance as the target working condition of the power outage base station. The system also includes a working condition forwarding module, which updates the historical multidimensional working condition feature vectors of each sample base station in any working condition cluster at a preset time interval, and calculates the offset distance between the updated historical multidimensional working condition feature vector and the center of the cluster. If the offset distance of any sample base station is greater than the preset working condition offset distance threshold, the sample base station is reassigned to the cluster with the smallest distance to realize working condition migration.

[0066] Furthermore, the lower limit duration calculation module 33 includes: The second calculation unit is used to match the allowable power outage risk level parameter based on the protection level, and use the complement of the allowable power outage risk level parameter as the quantile, wherein the protection level is negatively correlated with the allowable power outage risk level parameter; The quantile extraction unit is used to extract residual quantiles from the sample predicted residual sequence based on the quantiles; The third calculation unit is used to calculate the difference between the predicted range duration and the residual quantile. If the difference is greater than the preset lower limit duration threshold, the difference is used as the lower limit duration of the predicted range. If the difference is less than or equal to the preset lower limit duration threshold, model prediction anomaly warning information is generated and fed back.

[0067] Furthermore, the lower limit duration calculation module 33 also includes: The penalty coefficient determination unit is used to retrieve the sample penalty coefficient based on the sample quantity or the missing ratio if the number of samples in the sample prediction residual sequence under the target working condition is less than a preset sample quantity threshold, or if the missing ratio of key fields in the historical sample data on which the sample prediction residual sequence is based is greater than a preset ratio threshold. The correction unit is used to negatively correct the allowable power loss risk level parameter based on the sample penalty coefficient, and use the complement of the corrected allowable power loss risk level parameter as the new quantile, and return to execute the step of extracting residual quantiles from the sample predicted residual sequence based on the quantile, so as to achieve adaptive updating of residual quantiles.

[0068] Furthermore, the data processing module 31 includes: The fourth calculation unit is used to extract the operation monitoring data of the power outage base station within a series of preset historical days before the current power outage time, and within the historical sliding window corresponding to the current power outage time each day. It also calculates the average operating energy consumption and standard deviation of operating energy consumption for each historical sliding window, and calculates the tidal load deviation characteristics based on the average operating energy consumption, standard deviation of operating energy consumption for different historical sliding windows, and the instantaneous load at the time of the power outage. The fifth calculation unit is used to extract charge and discharge data for a preset number of times from the battery monitoring data, and calculate the average voltage drop slope and the estimated transient internal resistance based on the charge and discharge data, and use the average voltage drop slope and the estimated transient internal resistance as dynamic characteristics of battery degradation. The feature construction unit is used to extract the nominal capacity and years of operation from the basic information as basic features, and construct a multi-dimensional operating condition feature vector based on the tidal load deviation feature, the battery degradation dynamic feature and the basic features.

[0069] Furthermore, the multi-dimensional operating condition feature vector is predicted based on the trained dual-stream architecture endurance prediction model, wherein the trained dual-stream architecture endurance prediction model includes a one-dimensional causal convolutional network, a multilayer perceptron, a fusion network, and a fully connected layer. The operating condition determination module 32 includes: The feature extraction unit is used to extract features from the time-series monitoring data matrix based on the one-dimensional causal convolutional network to obtain time-series discharge features, and to extract features from the multi-dimensional operating condition feature vector based on the multilayer perceptron to obtain structured battery performance features. The prediction unit is used to stitch together the time-series discharge characteristics and the structured battery performance characteristics through the fusion network, and to map the stitched fusion characteristics through the fully connected layer to obtain the predicted range duration.

[0070] Furthermore, the power supply equipment screening module 34 includes: The screening unit is used to calculate the safety margin of each schedulable power supply device based on the expected power supply travel time, connection start-up time, and safety buffer time of each schedulable power supply device, as well as the predicted lower limit of the endurance time, and selects the schedulable power supply devices whose safety margin meets the preset margin threshold as candidate power supply devices. The optimization solution unit is used to perform multi-objective optimization with the travel time, fuel consumption, task switching cost, and power outage risk cost of the candidate power supply equipment as optimization objectives, and solve for the target power supply equipment.

[0071] Furthermore, the system also includes: The residual calculation module is used to obtain the actual battery support time under the circumstances of normal power restoration, power supply equipment access, or battery undervoltage reaching the protection threshold, and to use the difference between the predicted battery life and the actual battery support time as the predicted residual of the power outage base station. The sample construction module is used to update the sample prediction residual sequence of the target operating condition with the residual of the power outage base station as an incremental sample, so as to calculate the lower limit of the battery life prediction based on the updated sample prediction residual sequence during the subsequent power supply scheduling process.

[0072] It is obvious to those skilled in the art that the modules or steps of the present invention described above can be implemented using general-purpose computing systems. They can be centralized on a single computing system or distributed across a network of multiple computing systems. Optionally, they can be implemented using computer-executable program code, which can then be stored in a storage system for execution by the computing system. In some cases, the steps shown or described can be performed in a different order than those described herein, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0073] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A base station emergency power supply scheduling method based on endurance prediction residuals, characterized in that, include: In response to a power outage event of any base station, feature vectors are extracted from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and a multi-dimensional operating condition feature vector is constructed based on the feature vector extraction results. Based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, the predicted battery life of the power outage base station is predicted, and the target operating condition is determined based on the multi-dimensional operating condition feature vector. Based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station, and the predicted battery life duration, the lower limit of the predicted battery life duration of the power outage base station is calculated. The sample prediction residual sequence is constructed based on the positive prediction residual of the battery life duration of each sample base station under the target operating condition. The predicted lower limit of the endurance duration is used as the arrival time constraint for the power supply equipment. The sum of the power outage time of the base station and the predicted lower limit of the endurance duration is used as the predicted lower bound of the power outage. The current time, the expected power supply journey time of each schedulable power supply equipment, the connection start-up time, and the safety buffer time are subtracted from the predicted lower bound of the power outage to obtain the safety margin of each schedulable power supply equipment. Schedulable power supply equipment whose safety margin meets a preset margin threshold is selected as candidate power supply equipment. Power supply equipment that does not meet the preset margin threshold, combined with the base station, does not participate in this round of task allocation. Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the candidate power supply equipment. A target power supply equipment is determined from the candidate power supply equipment, and the target power supply equipment is scheduled to provide emergency power to the base station. The calculation process for the predicted lower limit of the power outage base station's remaining power includes: matching the allowable power outage risk level parameter based on the protection level, and using the complement of the allowable power outage risk level parameter as the quantile, wherein the protection level is negatively correlated with the allowable power outage risk level parameter; Based on the quantile, the residual quantile is extracted from the sample prediction residual sequence; the difference between the predicted battery life and the residual quantile is calculated; if the difference is greater than a preset lower limit duration threshold, the difference is used as the lower limit duration for battery life prediction.

2. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, In the event of multiple power outage base stations, the power supply equipment is rescheduled in response to any rolling rescheduling trigger event. The rolling rescheduling trigger events include adding a new power outage base station, restoration of normal power supply, the predicted lower limit of the battery life duration being less than the lower limit threshold, the arrival delay of the target power supply equipment being greater than the preset delay threshold, the base station protection level being updated, and the target power supply equipment operating abnormally. The rolling rescheduling process includes: Based on the real-time power supply information of the power supply equipment, the power supply scheduling task is divided into switchable tasks and non-switchable tasks; The non-switchable tasks are frozen, and multi-objective optimization is performed on the switchable power supply equipment and the power outage base station to be rescheduled for the non-frozen tasks, so as to determine the updated power supply equipment from the switchable power supply equipment for each of the rescheduled power outage base stations.

3. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, Determining the target working condition based on the multi-dimensional working condition feature vector includes: Retrieve a pre-constructed working condition cluster, wherein the working condition cluster is obtained by clustering the historical multidimensional working condition feature vectors of all sample base stations within the target scheduling area; Calculate the distance between the multidimensional working condition feature vector and the center of each working condition cluster, and take the working condition corresponding to the working condition cluster with the smallest distance as the target working condition of the power outage base station; The method further includes: For any operating condition cluster, the historical multidimensional operating condition feature vector of each sample base station in the cluster is updated at a preset time interval, and the offset distance between the updated historical multidimensional operating condition feature vector and the center of the cluster is calculated. If the offset distance of any sample base station is greater than the preset operating condition offset distance threshold, the sample base station is reassigned to the cluster with the smallest distance to realize operating condition migration.

4. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, The method further includes: If the difference is less than or equal to the preset lower limit time threshold, then model prediction anomaly warning information is generated and fed back.

5. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, The method further includes: If the number of samples in the sample prediction residual sequence under the target working condition is less than the preset sample number threshold, or the missing proportion of key fields in the historical sample data on which the sample prediction residual sequence is based is greater than the preset proportion threshold, then the sample penalty coefficient is retrieved based on the sample number or the missing proportion. The permissible power loss risk level parameter is negatively corrected based on the sample penalty coefficient, and the complement of the corrected permissible power loss risk level parameter is used as the new quantile. The process then returns to the step of extracting residual quantiles from the sample predicted residual sequence based on the quantiles, thereby achieving adaptive updating of residual quantiles.

6. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, Feature vectors are extracted from the historical operation monitoring data and battery monitoring data of the power outage base station, as well as the basic information of the power outage base station. Based on the feature vector extraction results, a multi-dimensional operating condition feature vector is constructed, including: Extract the operation monitoring data of the power outage base station within a series of preset historical days before the current power outage time, and within the historical sliding window corresponding to the current power outage time each day. Calculate the average operating energy consumption and standard deviation of operating energy consumption corresponding to each historical sliding window. Based on the average operating energy consumption, standard deviation of operating energy consumption corresponding to different historical sliding windows, and the instantaneous load at the time of the power outage, calculate the tidal load deviation characteristics. Extract charge and discharge data for a preset number of times from the battery monitoring data, and calculate the average voltage drop slope and the estimated transient internal resistance based on the charge and discharge data. Use the average voltage drop slope and the estimated transient internal resistance as dynamic characteristics of battery degradation. The nominal capacity and years of operation are extracted from the aforementioned basic information as basic features; A multi-dimensional operating condition feature vector is constructed based on the tidal load deviation characteristics, the battery degradation dynamic characteristics, and the basic characteristics.

7. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 6, characterized in that, The multi-dimensional operating condition feature vector is predicted based on the trained dual-stream architecture endurance prediction model, wherein the trained dual-stream architecture endurance prediction model includes a one-dimensional causal convolutional network, a multilayer perceptron, a fusion network, and a fully connected layer. Based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, the predicted battery life of the power outage base station is calculated, including: Based on the one-dimensional causal convolutional network, feature extraction is performed on the time-series monitoring data matrix to obtain time-series discharge features. Based on the multilayer perceptron, feature extraction is performed on the multi-dimensional operating condition feature vector to obtain structured battery performance features. The time-series discharge characteristics and the structured battery performance characteristics are spliced ​​together by the fusion network, and the spliced ​​fusion characteristics are mapped by the fully connected layer to obtain the predicted driving range.

8. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the candidate power supply equipment, including: Multi-objective optimization is performed with the travel time, fuel consumption, task switching cost, and power outage risk cost of the candidate power supply equipment as optimization objectives to obtain the target power supply equipment.

9. The base station emergency power supply scheduling method based on endurance prediction residuals according to claim 1, characterized in that, The method further includes: Under the circumstances of normal power restoration, power supply equipment access, or battery undervoltage reaching the protection threshold, the actual battery support time is obtained, and the difference between the predicted battery life and the actual battery support time is used as the prediction residual of the power outage base station. The residuals of the power outage base stations are used as incremental samples to update the sample prediction residual sequence of the target operating condition, so that the lower limit of the battery life prediction can be calculated based on the updated sample prediction residual sequence during subsequent power supply scheduling.

10. A base station emergency power supply dispatching system based on endurance prediction residuals, characterized in that, include: The data processing module is used to respond to any power outage event of the base station, extract feature vectors from the operation monitoring data, battery monitoring data and basic information of the power outage base station in the historical period, and construct a multi-dimensional operating condition feature vector based on the feature vector extraction results. The operating condition determination module is used to predict the predicted battery life of the power outage base station based on the multi-dimensional operating condition feature vector and the time-series monitoring data matrix within a preset period after the power outage, and to determine the target operating condition based on the multi-dimensional operating condition feature vector. The lower limit duration calculation module is used to calculate the lower limit duration of the power outage base station's predicted battery life based on the sample prediction residual sequence under the target operating condition, the protection level of the power outage base station, and the predicted battery life duration. The sample prediction residual sequence is constructed based on the positive prediction residuals of the battery life duration of each sample base station under the target operating condition. The calculation process of the lower limit duration includes: matching an allowable power outage risk level parameter based on the protection level, and using the complement of the allowable power outage risk level parameter as a quantile, wherein the protection level is negatively correlated with the allowable power outage risk level parameter; extracting residual quantiles from the sample prediction residual sequence based on the quantiles; calculating the difference between the predicted battery life duration and the residual quantiles; if the difference is greater than a preset lower limit duration threshold, then the difference is used as the lower limit duration of the predicted battery life duration. The power supply equipment screening module is used to use the predicted lower limit of the endurance duration as a constraint on the arrival time of the power supply equipment. Specifically, the sum of the power outage time of the base station and the predicted lower limit of the endurance duration is used as the predicted lower bound of the power outage. The current time, the expected power supply journey time of each schedulable power supply equipment, the connection start-up time, and the safety buffer time are subtracted from the predicted lower bound of the power outage to obtain the safety margin of each schedulable power supply equipment. Schedulable power supply equipment whose safety margin meets a preset margin threshold is selected as candidate power supply equipment. Power supply equipment that does not meet the preset margin threshold, combined with the base station, does not participate in the current task allocation. Multi-objective optimization is performed with the goal of minimizing the overall operating cost of the candidate power supply equipment. A target power supply equipment is then selected from the candidate power supply equipment, and the target power supply equipment is scheduled to provide emergency power to the base station.

Citation Information

Patent Citations

  • Emergency generator car scheduling method and device and electronic equipment

    CN121638807A

  • Smart city base station power supply emergency supervision Internet of Things large model system and method

    CN122292653A

  • Multi-target dynamic scheduling distribution method and system for communication network emergency resources

    CN122457452A

  • Transformer substation equipment fault intelligent diagnosis method and system based on multi-source time sequence data

    CN122548581A

  • Load interval prediction method based on conditional residual distribution modeling and local adaptive calibration

    CN122638991A