Resilience enhancement strategies for distribution networks considering the coordination of 5G base stations and multiple flexible resources

Abstract

In recent years, the increasing frequency of extreme natural disasters has significantly exposed the vulnerability of distribution networks. To address this challenge, this study proposes a resilience enhancement strategy that integrates 5G base stations with multiple flexible resources. First, a disaster scenario modeling framework is developed by considering typhoon wind speed, line outage probability, and renewable generation curtailment, and representative scenarios are extracted using Latin Hypercube Sampling (LHS) and K-means++ clustering. Subsequently, a coordinated optimization model is established with the objectives of minimizing critical load loss and economic cost, in which the energy storage of 5G base stations is utilized not only to guarantee communication loads but also to participate in system dispatch. Meanwhile, distributed generation, electric vehicles, and mobile energy storage systems are coordinately scheduled, combined with network reconfiguration to achieve complementary utilization of multiple resources. The bi-objective problem is then reformulated into a mixed-integer second-order cone programming (MISOCP) model using second-order cone relaxation and the weighted-sum method, and efficiently solved. Finally, case studies conducted on a modified IEEE 33-bus system demonstrate that, compared with uncoordinated operation under extreme disaster conditions, the proposed strategy reduces critical load loss by 84.6%, decreases economic losses by 76.9%, and improves the comprehensive objective value by 83.7%. Furthermore, by coordinating 5G base stations with mobile energy storage to support islanded areas, the overall comprehensive objective is further improved by 89.1%.

Introduction

Motivation

In recent years, the frequency and intensity of extreme weather events—such as typhoons, blizzards, and heavy rainfall—have posed severe challenges to power distribution systems. For instance, Typhoon Lekima in 2019 caused power outages affecting more than six million households in eastern China¹, while the 2021 Texas blizzard triggered widespread electricity interruptions, leaving millions of users without power². Traditionally, power systems have been designed with reliability as the primary objective, emphasizing continuous power supply under foreseeable disturbances³. However, when confronted with low-probability yet high-impact extreme events, such systems often prove inadequate. Consequently, the concept of resilience has been introduced into the power system domain to characterize a distribution system’s ability to withstand, recover from, and adapt to extreme disasters^4,5. Therefore, developing realistic disaster scenarios and fault models under multiple hazard factors, as well as exploiting the potential of various flexibility resources, has become a key research focus in recent years.

Literture review

In efforts to enhance the resilience of distribution networks, researchers have primarily focused on the mechanisms of disasters and scenario modeling. The impact analysis of extreme events is commonly conducted using fragility curves⁶, which describe the behavior of lines or components through failure rates⁷, probabilistic modeling⁸, and fault scenarios⁹. For example, Reference¹⁰ investigated the failure risks of overhead lines and towers under extreme weather conditions and proposed models for line-break probability and failure modes based on statistical analysis and physical modeling. Reference¹¹ employed the Monte Carlo method to simulate typhoon scenarios and calculate line failure probabilities using fragility curves. Furthermore, References^12–14 integrated typhoon wind-field models, machine learning, and Monte Carlo simulations to generate comprehensive scenario sets, providing essential tools for resilience assessment. Meanwhile, References^15,16 jointly modeled the uncertainty of renewable generation and disaster scenarios under extreme weather, capturing the coupled effects of disasters on supply–demand balance. Reference¹⁷ focuses on the planning of virtual power plants under extreme events such as earthquakes and floods, providing a new perspective for enhancing system resilience in disaster scenarios. However, most of these studies focus on single factors or limited scenarios, which remain insufficient to comprehensively reflect the multifaceted impacts of extreme disasters on distribution networks.

In recent years, energy management and coordinated dispatch of multiple flexible resources have been widely adopted to enhance the operational performance of distribution networks. Reference¹⁸ proposed a real-time price–based demand response program combined with network reconfiguration and energy management, significantly improving system reliability and economic efficiency. Reference¹⁹ employed Benders decomposition for HVDC expansion planning to optimize network topology and capacity allocation, enhancing both economic and technical performance. Reference²⁰ proposes a stochastic economic dispatch strategy for a renewable-based virtual power plant coupled with electric-spring units, achieving coordinated utilization of diverse flexibility resources under renewable and load uncertainties. Reference²¹ develops a networked virtual power plant energy management model incorporating EV parking lots and price-based demand response, where voltage security, economic performance, and uncertainties are jointly considered. References^22,23 further illustrate—through electro-hydrogen integrated systems and hydrogen-storage–EV coordinated optimization—the potential of multi-energy synergy in enhancing distribution network flexibility and coordination. For scenarios with high penetration of renewable energy, Reference²⁴ proposed a resilience-oriented reconfiguration strategy considering renewable uncertainties, enhancing system stability through coordinated wind–solar dispatch and network reconfiguration. Reference²⁵ achieved economic optimization of renewable microgrids through coordinated scheduling of electric vehicles, energy storage, and demand response. Moreover, Reference²⁶ analyzed the potential of energy storage to enhance social welfare from a market perspective and introduced risk analysis to improve scheduling robustness, while Reference²⁷ applied deep learning to improve the prediction and execution efficiency of demand response in short-term microgrid operation. Meanwhile, Reference²⁸ investigated coordinated optimization between distribution networks and 5G base stations, mainly focusing on communication load migration and reserve capacity management. Although these studies have made significant progress in energy management and multi-resource coordination, most of them primarily focus on economic optimization or local technical performance under normal operating conditions, while limited attention has been paid to resilience challenges driven by multi-source uncertainties under extreme disaster scenarios.

To enhance the resilience of distribution networks under extreme disasters, various strategies have been proposed by researchers. Reference²⁹ developed a two-stage robust optimization model for energy storage systems that enhances the disaster resistance of distribution networks under extreme weather conditions, addressing the problem of large-scale power outages caused by natural disasters. Reference³⁰ adopted a coordinated strategy combining network reconfiguration and mobile energy storage to improve post-disaster load restoration efficiency. Reference³¹ proposed an outage management strategy that strengthens the resilience of distribution systems through network reconfiguration and distributed energy scheduling. Reference³² investigated the stochastic scheduling of electric vehicles (EVs) as mobile storage resources to mitigate the uncertainty of charging and discharging behavior, while Reference³³ proposed an optimization method for EV deployment to enhance supply restoration flexibility. Reference³⁴ approaches the problem from the perspective of transmission network structural optimization, providing structural-level insights to support network reconfiguration and resilience planning. However, most of the aforementioned studies focus on single types of resources, and the exploration of emerging flexibility resources remains limited. In recent years, with the widespread deployment of 5G base stations, their backup batteries have demonstrated tremendous potential. Reference³⁵ examined the dispatchable capacity of base station energy storage and revealed its backup value in power system operation. Reference³⁶ proposed a coordinated scheduling strategy for 5G base station energy storage to improve resource utilization. Reference³⁷ further coordinated 5G base station energy storage with distributed generation to achieve rapid recovery during disasters. These studies indicate that 5G base stations can not only ensure reliable power supply for communication networks but also serve as novel flexibility resources participating in system scheduling, thereby providing a new pathway to enhance the resilience of distribution networks. A comparative summary of the literature review is presented in Table 1.

Table 1.

Comparison of literature review.

Reference	Disaster scenario modeling	Multi-Resource coordination	Network reconfiguration	5G Base station energy storage
10,11,12,13,14	√
15,16,17	√
18	×	×	√	×
19	×	√	×	×
20,21,22,23	×	×	√	×
24	×	×	√	×
25	×	√	×	×
26	×	√	√	×
27	×	√	×	×
28	×	√	×	√
29	√	√	√	×
30		√	√	×
31		√	√	×
32		√	×	×
33		√	×	×
34		×	√	×
35		√	×	√
36		√	×	√
37		√	×	√
This paper	√	√	√	√

Research gaps

In summary, existing studies have made significant progress in fault prediction, disaster scenario modeling, and single-type flexibility resource scheduling for distribution networks under extreme weather conditions. However, several limitations remain. First, the comprehensive modeling of multiple uncertainty factors is still insufficient. Second, the potential of 5G base station energy storage has not been fully explored, and its coordinated optimization with electric vehicles, mobile energy storage systems, and distributed generation has yet to be realized.

Contributions

To address the research gaps identified above, this study proposes a resilience enhancement strategy for distribution networks that integrates 5G base stations and multiple flexible resources, as illustrated in Fig. 1. First, a coupled disaster scenario generation framework is developed by integrating typhoon wind fields, line outage probabilities, and wind–solar power curtailment, and representative scenarios are extracted using Latin hypercube sampling and K-means++ clustering. Then, considering that 5G base stations must guarantee communication loads while exhibiting dispatch potential under extreme disasters, a storage capacity partitioning model is proposed, enabling their participation in power system dispatch while ensuring communication reliability, thereby forming a flexible resource pool jointly with distributed generators, electric vehicles, and mobile energy storage. Furthermore, a resilience-oriented optimization model is established with the objectives of minimizing critical load outages and economic losses. By incorporating second-order cone relaxation and the weighted-sum method, an integrated coordinated optimization of multi-type resources, network reconfiguration, and islanding partition is achieved. Finally, simulation studies on a modified IEEE 33-bus system verify the effectiveness of the proposed strategy in enhancing the disaster resistance, self-healing, and recovery capabilities of distribution networks.

Fig. 1.

Framework of the proposed strategy.

The main contributions of this study are summarized as follows:

1. A comprehensive scenario modeling method for extreme typhoon disasters is developed, integrating typhoon wind speeds, line outage probabilities, and wind–solar curtailment, and extracting typical scenarios using Latin hypercube sampling and K-means++ clustering.

2. A storage capacity partitioning and dispatch mechanism for 5G base stations is proposed, enabling energy storage to simultaneously guarantee communication loads and participate in power system dispatch.

3. A resilience-oriented coordinated optimization model integrating multiple flexible resources and network reconfiguration is established, in which 5G base stations, distributed generators, electric vehicles, and mobile energy storage are jointly scheduled. This strategy effectively mitigates wind–solar curtailment and line faults caused by typhoon disasters, while reducing economic losses and enhancing system self-healing and recovery capabilities.

Resources for enhancing distribution network resilience

This paper investigates resilience enhancement from the perspective of power–communication system coordination. The resources are summarized as follows: the distribution network integrates flexible resources such as wind power (WT), photovoltaic generation (PV), electric vehicle charging stations (EVCS), and mobile energy storage (MES), along with network reconfiguration. On the communication side, clusters of 5G base stations equipped with energy storage (BSS) are deployed, which serve not only as critical communication loads but also as dispatchable emergency power supplies. All resources are connected to distribution network nodes according to their geographical locations, thereby forming an integrated power–communication resource configuration, as illustrated in Fig. 2.

Fig. 2.

Resource configuration structure.

K-means++-based fault scenario modeling

This paper develops a fault scenario model under typhoon disasters, which incorporates typhoon wind speeds, line outage probabilities, and renewable generation curtailments. Typical scenarios are extracted using Latin hypercube sampling (LHS) combined with the K-means++ clustering method, providing the basis for subsequent optimization.

Probabilistic modeling of typhoon intensity

In this paper, the wind speed probability density function of typhoon disasters is fitted using the Extreme Value Type III (EV-III) distribution, as given by the following equation

1

Where denotes the typhoon wind speed in region s at time t; represents the scale parameter; is the location parameter; and is the shape parameter

Line outage probability modeling for distribution networks

In this paper, a probabilistic model of distribution line faults under typhoon disasters is proposed to predict their failure probability , which can be expressed as follows:

2

Where denotes the maximum wind speed that the distribution line in region s can withstand at time t; represents the maximum wind load of the distribution line; is the wind pressure non-uniformity coefficient of the conductor; is the wind pressure height variation coefficient; denotes the shape coefficient of the distribution line; is the line diameter; represents the angle between the line and the typhoon wind direction; and is the effective line length under wind load.

Modeling of renewable generation curtailmen

1. Wind power curtailment probability model under typhoon disasters

The relationship between the power captured by the wind turbine from the wind and the current wind speed can be expressed as:

3

Where is the cut-in wind speed; is the rated wind speed; is the cut-out wind speed; denotes the rated output of the wind turbine; is the wind energy utilization coefficient; and represents the current air density.

• 2)PV curtailment probability model under typhoon disasters

Under typhoon conditions, the solar irradiance deviates from that under normal conditions. By modeling this deviation with a probabilistic model that follows a normal distribution, the PV output can be expressed as:

4

where denotes the solar irradiance; represents the irradiance under standard test conditions (1000W/m²); is the deviation between solar irradiance under typhoon weather and that under normal weather conditions; and denotes the rated output power of the photovoltaic system.

Latin hypercube sampling-based scenario generation algorithm

Assuming that the sampling size of the t time period is , the steps of scenario sampling using LHS for any time period can be described as follows, in conjunction with its sampling process diagram:

Step 1: Fit the wind and photovoltaic data to their corresponding probability distributions to obtain the scale, location, and shape parameters of the Extreme Value Type III distribution, as well as the shape parameters of the Beta distribution.

Step 2: Based on the probability distribution parameters obtained in Step 1, calculate the cumulative distribution function , and divide into sub-intervals, each with an interval length of .

Step 3: Generate a random number within the range [0,1]. For the interval among the sub-intervals, the corresponding cumulative distribution function value of interval can then be determined as:

5

Step 4: Assuming that the inverse function of the cumulative distribution function is , the sampling value of the random variable can be obtained by substituting into this function, that is:

6

Through the above steps, the sampling scenario of the random variable for time period can be obtained.

Typical scenario selection based on K-means++ clustering

Step 1: Based on the scenario reduction algorithm using K-means++ clustering, the clustering effect is optimized by improving the selection of initial cluster centers. First, the number of typical scenarios is specified, and the first cluster center is selected randomly. The remaining initial centers are then selected sequentially according to a probability distribution. Samples are assigned to the nearest cluster center, and the cluster centers are updated to the mean of their assigned samples. This process is repeated until the change in cluster centers is less than a threshold or the maximum number of iterations is reached. The final cluster partitions and the corresponding typical scenarios are then obtained, completing the scenario reduction.

Step 2: Following the above procedure, probability scenarios for line outage failures and renewable generation curtailments under typhoon disasters can be derived. Two typical scenarios, namely “network reconfiguration” and “islanded operation,” together with their probabilities, are obtained. These provide fault scenarios for developing subsequent strategies to enhance distribution network resilience.

Resilience enhancement strategy for distribution networks through the coordination of 5G base stations and multiple flexible resources

This paper proposes a coordinated optimization strategy that comprehensively considers 5G base station energy storage, distributed generation, electric vehicles, and mobile energy storage. The strategy is designed from the perspectives of objective function formulation, capacity modeling, system constraints, and optimization methods and procedures.

Objective function

The objective function includes the Critical Load Loss of the System and the Comprehensive Economic Cost

1. Critical load loss of the system

In power systems, loads can be classified into different importance levels, generally including first-level, second-level, and third-level loads. Let the set of nodes be denoted as , where larger numerical values indicate higher load criticality, and let the time window be denoted as . The expression of the system critical load loss is given by (7):

7

Where denotes the importance weight of node n; represent the curtailed active and reactive power loads of node , respectively (MW).

• (2)Comprehensive economic cost

The comprehensive economic cost (10⁴CNY) consists of two components: the direct economic loss (10⁴CNY) and the cost of economic measures (10⁴CNY) Its expression is given by (8):

8

Where denotes the unit curtailed load cost coefficient of node (10⁴CNY/MW); represents the overall effect of total demand response and the corresponding electricity price variation(10⁴CNY); is the total charging amount within the scheduling period(MW); is the total discharging amount(MW); denotes the switching state of line at time ; and represents the switching state of line at time .

The coefficients in (8) are employed to convert different physical quantities into a unified equivalent loss index, thereby reflecting their relative cost levels under disaster scenarios. Among them, the coefficient 0.5 indicates that the impact of reactive power curtailment on system resilience is weaker than that of active power curtailment. The values −0.05, 0.1, and 0.05 are normalized coefficients derived from demand response compensation, energy storage operating costs, and typical switching operation and maintenance costs, respectively. In particular, the coefficient 0.05 represents the extremely low cost of a single switching operation relative to critical load curtailment, which is introduced to suppress excessive network partitioning and frequent switching actions, while preserving realistic engineering significance.

Modeling of dispatchable capacity of 5G base stations

A 5G base station typically consists of communication equipment, air conditioning equipment (AC), and power supply equipment, with its typical structure illustrated in Fig. 3

Fig. 3.

Typical structure of a 5G base station.

The communication equipment includes transmission devices, active antenna units (AAU), and baseband units (BBU), where the AAU is installed on the tower outside the equipment room. The communication equipment is mainly responsible for transmitting and processing wireless signals, serving as the interface between mobile terminals and the 5G network.

The power supply equipment mainly consists of switching power supplies and energy storage batteries, which provide direct current (DC) power to the main devices of the 5G base station. The primary power source is the external grid; when a distribution network fault causes an outage, the energy storage battery serves as a backup power supply. The air conditioning equipment is used to maintain the indoor temperature of the 5G base station and is affected by ambient temperature once the base station room has been constructed.

In this paper, both the load characteristics and the backup requirements are considered to describe the dual functions of communication support and grid scheduling.

Linear load model of 5G base stations based on data traffic

The power load model of 5G base stations is decomposed into static and dynamic loads.

9

Where denotes the static load of the 5G base station, and represents the dynamic load of the 5G base station.

1. Static load model

The static load mainly includes the BBU load, the baseline load of the AAU, and the power supply load, which are primarily used to maintain the basic functions of the 5G base station, and can be expressed as:

10

Where denotes the power consumption of the BBU, represents the baseline power consumption of the AAU, refers to the AAU power consumption caused by downlink signaling, and denotes the power consumption of the power supply.

• (2)Dynamic load model

The dynamic load consists of the AAU incremental load and the air conditioning equipment load :

11

The AAU incremental load is sensitive to the communication traffic. As the traffic increases, the power consumption of the AAU in the main equipment rises accordingly. The relationship between the AAU load and the communication traffic is illustrated in Fig. 4, where the blue area represents the static baseline load and the orange area denotes the incremental load, whose magnitude depends on the level of communication traffic.

Fig. 4.

Relationship between communication traffic and AAU power consumption.

The number of active 5G user terminals and the volume of 5G data traffic within the coverage area of a 5G base station are key factors affecting its communication load. The total data traffic in the coverage area of a 5G base station can be expressed as:

12

Where denotes the total data traffic in the region during period (Byte); represents the number of active 5G user terminals in the region during period ; and refers to the total 5G data traffic consumed by the active user terminal during period .

According to the relationship between data traffic and downlink data rate, the equivalent downlink data rate of a single 5G base station at time can be derived from the total data traffic within the coverage area of the 5G base station during period as follows:

13

Where denotes the time granularity, and represents the equivalent downlink data rate(bps).

The communication load ratio is defined as the ratio of the total downlink data rate of all users within the coverage area of a 5G base station to the maximum downlink data rate of the base station, which can be expressed as:

14

Where denotes the inflection point of the downlink data rate carried by the 5G base station, indicating the change in the linear growth trend of power consumption after the base station sustains a certain communication load; and represents the maximum downlink data rate that the 5G base station can support.

In summary, the dynamic incremental load of the AAU equipment can be expressed as:

15

Where denotes the inflection point of the AAU equipment power consumption, and represents the maximum power consumption of the AAU equipment.

The load of the air conditioning equipment is mainly related to temperature and can be expressed as:

16

Where denotes the surface area of the peripheral building structure of the equipment room; represents the heat transfer coefficient; is the indoor temperature of the equipment room; is the ambient temperature; denotes the total power consumption of the AAU and BBU in the 5G base station; represents the efficiency coefficient of the AAU and BBU energy consumption that is not converted into heat; denotes the conversion efficiency of air conditioning power; and is the performance coefficient of the air conditioning equipment.

Therefore, the dynamic load model of the 5G base station can be expressed as:

17

Modeling of dispatchable capacity considering backup demand for power outages

Power–capacity coupling constraint of 5G base station energy storage:

18

Where and denote the charging and discharging power of storage unit in the 5G base station during period ; is the charging/discharging indicator, with 1 for charging and 0 for discharging; and represent the maximum charging and discharging power of storage unit during period ; and 、 are the charging and discharging efficiencies of the 5G base station storage, respectively.

After an unexpected interruption of the external power supply, the energy storage of the 5G base station can serve as an uninterruptible power supply (UPS) to ensure the load demand of the base station, while the surplus energy can be scheduled for dispatch. The storage capacity of the 5G base station is divided into two parts: backup capacity and dispatchable capacity, as illustrated in Fig. 5.

Fig. 5.

Schematic diagram of the capacity division of 5G base station energy storage.

Where and are the minimum and maximum SOC thresholds set to prevent battery overcharging and over-discharging. The dispatchable capacity depends on the configured capacity, state of health, and backup capacity of the storage battery.

The backup batteries of 5G base stations are required to meet the emergency power supply standards of the telecommunications industry under disaster conditions. According to the application specifications for backup lithium batteries in base stations and typical engineering practices of network operators, base stations are required to sustain communication loads for at least 1–2 hours under extreme conditions; therefore, a certain proportion of non-dispatchable capacity must be reserved. Considering the rated power of base stations and battery configuration, the is set to 20% in this study to ensure autonomous power supply capability. Meanwhile, lithium batteries operating at high SoC levels are prone to accelerated capacity degradation and increased safety risks. As a result, communication backup power systems generally limit the is set to 90%. Accordingly, the upper SoC limit for the participation of 5G base station energy storage in system dispatch is set to 90% in this study, so as to balance operational safety and battery lifetime.

19

Where denotes the dispatchable capacity of storage unit in the 5G base station at time ; represents the state of health (SOH) of the battery, ; denote the rated capacity of the battery at the time of initial installation.

The size of the backup capacity of a 5G base station is influenced by the minimum backup time. When the base station is powered by an external supply and the communication load is considered, the minimum required backup capacity can be expressed as:

20

Where denotes the power consumption of base station during period , and represents the minimum backup time of base station .

For the cluster of 5G base stations connected to distribution network node , the feasible region of the aggregated backup energy storage is given by:

21

Where denotes the number of 5G base stations connected to node ; and represent the maximum charging and discharging power of the energy storage cluster of 5G base stations at node j; is the energy of the storage cluster at node during period ; and denotes the maximum dispatchable capacity of the storage cluster at node .

System operation constraints

(1) The power constraint of line at time t and the voltage drop balance equation can be expressed as:

22

Where and denote the active and reactive power flows through branch at time ; and are the active and reactive power flows of branch at time ; and represent the resistance and reactance of branch ; is the current through branch ; and denote the active and reactive power flowing out of node ; and are the active and reactive power purchased by node from the main grid; is the reactive power demand of node ; and are the voltages of nodes and respectively.

(2) The node voltage constraints and branch current constraints can be expressed as:

23

Where and denote the lower and upper voltage limits of node at time ; when node is a slack node, condition applies; and represents the max allowable current of branch at time .

(3) The constraint on the number of line reconnections can be expressed as:

24

Where and denote the on/off status indicators of tie line at time ; and represents the daily limit on the number of tie line reconfigurations.

(4) Constraints on the charging and discharging power of mobile energy storage:

25

Where and denote the charging and discharging power of the mobile energy storage unit during period ; and represents its maximum charging/discharging power.

(5) Modeling and constraints of electric vehicle load shifting capability stimulated by price regulation:

To enhance the resilience of urban power grids under typhoon conditions, it is essential to fully exploit the load regulation potential of electric vehicles. A load-shifting control strategy can be developed by considering the interactions between multi-period incentive prices and response energy, which not only strengthens the grid’s ability to guide EV charging loads but also reduces the charging costs for EV users.

is defined as the response coefficient of the load variation in period i to the price variation in period , given by:

26

Where and denote the baseline charging energy and its variation in period ; and represent the original price and its variation, respectively.

27

By adding the variations of price and power to their initial values, the final power and price after the EV price response can be obtained. To avoid excessively high elastic prices that may cause user dissatisfaction, an upper limit should be imposed on the final price.

28

Where denotes the final electricity price after the EV load response, while and represent the minimum and maximum prices, respectively.

Model formulation and solution

(1) The optimization model developed in this study for enhancing the resilience of distribution networks is formulated as a multi-objective Mixed-Integer Second-Order Cone Programming (MISOCP) problem. Since the two objectives often conflict in practical scheduling, a weighted-sum approach is adopted to convert the bi-objective problem into a single-objective one. The resulting integrated objective function is expressed as:

29

Where and denote the weighting coefficients of the two objectives, with . Reference³⁸ adopted the CRITIC (Criteria Importance Through Intercriteria Correlation) objective weighting method to determine the optimal weights based on the contrast intensity and inter-criteria correlations of the evaluation indicators. The results revealed that under disaster scenarios, resilience-related indicators such as “recovery capability” and “damage tolerance” were automatically assigned higher weights, indicating that critical performance indices should be prioritized for optimization in emergency conditions. In this study, the weight coefficients between critical load loss and economic cost are determined as follows: .The transformed single-objective optimization model can be efficiently solved using commercial solvers such as Gurobi.

(2) The branch power flow model is inherently nonlinear and can be reformulated using Second-Order Cone Relaxation (SOCR). The transformation process introduces new variables and to replace the squared terms of voltage and current and in the original constraints, thereby converting the nonlinear quadratic constraints involving active and reactive power flows:

30

Upon further relaxation of the above expression, the following is derived:

31

Through further equivalent reformulation, the following is obtained:

32

Where denotes the 2-norm of a matrix.

After the above steps, the nonlinear constraints in the branch power flow model can be converted into linear constraints.

Resilience enhancement process of distribution networks

To systematically enhance the resilience of distribution networks under typhoon disasters, this study proposes the resilience enhancement procedure illustrated in Fig. 6, namely the Resilience Enhancement Strategy of Distribution Networks Considering the Coordination of 5G Base Stations and Multiple Flexible Resources.

1. Disaster Scenario Generation: Probability models of typhoons, line faults, and WT/PV curtailment are constructed. Typical fault scenarios are generated and extracted through Latin hypercube sampling and K-means++ clustering, including “reconfigurable network” and “islanded autonomy” cases.

2. Objective Function Formulation: A bi-objective optimization model is developed with the goals of minimizing critical load loss and minimizing the overall economic cost.

3. Resource Modeling: Models of distributed generation, electric vehicle charging stations, and mobile energy storage are established. Special emphasis is placed on the energy storage of 5G base stations, whose capacity is divided into backup capacity for communication and schedulable capacity for the grid.

4. Model Processing and Solution: The bi-objective model is transformed into a single-objective one using the weighted-sum method. Nonlinear power flow constraints are relaxed via Second-Order Cone Relaxation (SOCR), leading to a MISOCP formulation that can be efficiently solved by Gurobi.

5. Result Analysis: The levels of critical load restoration and economic cost under different strategies are compared to verify the effectiveness of 5G base stations and multi-resource coordination in enhancing distribution network resilience.

Fig. 6.

Procedure of the resilience enhancement strategy for distribution networks.

Case study analysis

Case study parameter settings

In this study, simulations are conducted on a modified IEEE 33-bus distribution system. The operating scenario of the network is shown in Fig. 7. The system consists of 33 buses, 37 branches, and 5 tie switches. The maximum active and reactive power demands are 6.3 MW and 3.0 MW, respectively. Two wind turbines, WT1 and WT2, are connected at buses 11 and 33, while two photovoltaic plants, PV1 and PV2, are located at buses 18 and 21. Electric vehicle charging stations (EVCs) are deployed at buses 5 and 16. In addition, 5G base station clusters with embedded energy storage (BSS) are installed at buses 8, 15, 22, and 33. The relevant parameters of base station power consumption and rated power are provided in Table 2, while the actual operating power varies dynamically with communication load levels and environmental conditions.The key parameters of the distribution network system used in the case study are provided in Table 3.

Fig. 7.

Operating scenario of the distribution network.

Table 2.

5G base station parameters.

Parameters	Value
	0.3kw
	0.6kw
	1.2kw
	0.3kw
	0.15kw
	3
	0.85
	750bps
	2000bps

Table 3.

System operating parameters.

Parameters	Value
	4MW
	4MW
	1MW
	1.05p.u.
	0.95p.u.
	10 times per day
	300A
	0.5MW
	0.9MW
	1.2MWh

Setup establishment of typical fault scenarios

A coordinate system is established with (0,0) as the origin. The typhoon is assumed to originate at (−120 km, −120 km) and moves at a speed of 25 km/h along a direction forming a 30 angle with the horizontal axis. Other parameters of the typhoon are listed in Table 4.

Table 4.

Typical typhoon disaster parameters.

Characteristic parameters	Value
Start Time	13: 00
Time Duration()	12
Maximum Wind Speed()	32
Moving Speed()	25

In the typical scenarios generated based on Latin hypercube sampling and K-means++ clustering, the unavailability of lines 4, 12, 15, and 29 is significantly higher than the system average, as illustrated in Fig. 8.

Fig. 8.

Unavailability distribution of critical lines under typhoon disasters.

The high failure probabilities of these lines result in two typical scenarios, as shown in Fig. 9: (1) Reconfigurable Network Scenario: After the outages of lines 4 and 12, power supply is restored to the main grid through network reconfiguration via tie switches. (2) Islanded Autonomy Scenario: The outages of lines 15 and 29 cause certain areas to be disconnected from the main grid, requiring autonomous islanded operation.

Fig. 9.

Typical fault scenarios.

In this study, typical wind and photovoltaic curtailment scenarios are generated using Latin hypercube sampling and K-means++ clustering. The probabilistic curtailment characteristics of photovoltaic and wind power outputs under typhoon disasters are quantified, as illustrated in Figs. 10 and 11.

Fig. 10.

Typical photovoltaic output curtailment scenario.

Fig. 11.

Typical wind power output curtailment scenario.

As shown in Figs. 10 and 11, both photovoltaic and wind power outputs exhibit significant uncertainty and curtailment characteristics under typhoon disasters. Following the disaster at 13:00, the outputs of photovoltaic and wind power decrease sharply, simulating the impact of extreme weather conditions on the stability of renewable energy.

Based on the aforementioned typical fault scenarios, a comparative analysis is conducted to examine the relationship between total system load and power generation, as well as the load shedding rate, under different disaster intensities. The results are illustrated in Figs. 12, 13, 14.

Fig. 12.

Case without typhoon impact.

Fig. 13.

Scenario with typhoon impact and renewable generation curtailment.

Fig. 14.

Scenario with typhoon impact, renewable curtailment, and line outages.

As shown in Fig. 12, without typhoon impact, wind and photovoltaic outputs remain stable. Local distributed generation can support most loads, resulting in low dependence on the main grid and a well-balanced supply–demand condition. In Fig. 13, typhoon disturbances significantly curtail wind and solar outputs. The system thus relies more on the main grid, and the purchase curve rises sharply, reflecting the adverse impact of typhoons on renewable stability. In Fig. 14, when line outages are added, branch failures further restrict local generation. Although the main grid remains the main supply source, its capacity is insufficient, creating supply–demand gaps in the load curve. These results demonstrate that the combined effect of typhoon-induced line outages and renewable curtailment severely undermines the self-sufficiency and resilience of the distribution system.

To verify the effectiveness of the Latin Hypercube Sampling (LHS) and K-means++ clustering–based scenario generation method adopted in this study, three representative methods were designed to compare their computational efficiency in fault scenario generation. The comparison results are shown in Table 5.

Table 5.

Comparison of computational efficiency of scenario generation methods.

Method	Sampling time(s)	Clustering time (s)	Total time (s)
Method A	10.1	9.1	19.2
Method B	12.5	8.3	20.8
Method C	12.5	5.9	18.4

Method A: Traditional Monte Carlo (MC) simulation with K-means clustering

Method B: LHS with K-means clustering

Method C: LHS with K-means++ clustering

Method A exhibits a shorter sampling time due to its algorithmic simplicity. In contrast, the LHS-based methods require stratified sampling, resulting in a sampling time approximately 24% longer than that of the conventional MC approach. However, the superiority of LHS lies in its ability to provide a more uniformly distributed and spatially well-covered sample set for the subsequent clustering stage, thereby improving clustering efficiency. Specifically, the clustering time of Method B is reduced by 8.8% compared with Method A. More importantly, the K-means++ algorithm fully exploits the high-quality distribution of the LHS samples by optimizing the initialization of cluster centers, which significantly accelerates convergence. As a result, the clustering time of Method C is reduced by 35.2% and 28.9% compared with Methods A and B, respectively. Although LHS sacrifices part of the sampling efficiency, its combination with K-means++ achieves a net improvement in the overall computational time by substantially enhancing the clustering process. This demonstrates that the relatively “slower” sampling of LHS is a trade-off for higher sample quality, which is subsequently exploited by K-means++.

Optimization results of distribution network resilience enhancement

To verify the effectiveness of the proposed resilience enhancement strategy, four scenarios are designed for analysis:

Scenario 1: Under extreme typhoon conditions with wind/PV curtailment and line outages, no flexible resources are coordinated.

Scenario 2: Multiple flexible resources, except 5G base stations, are coordinated to enhance distribution network resilience.

Scenario 3: Both 5G base stations and multiple flexible resources are coordinated to improve distribution network resilience.

Scenario 4: Considering that islanded areas cannot exchange energy with the main grid, 5G base stations and mobile energy storage are further coordinated to enhance resilience.

Analysis of the schedulable capacity of 5G base stations

The charging and discharging behavior of the 5G base station energy storage system is simulated to analyze its energy transfer capability and scheduling characteristics, as illustrated in Fig. 15.

Fig. 15.

Analysis of the schedulable capacity of the 5G base station energy storage system.

Figure 15 illustrates the 24-hour dynamic distribution of charging and discharging power of storage systems configured at different 5G base stations, used to analyze their scheduling capability and operating patterns. Positive values represent charging, while negative values indicate discharging. Overall, during the pre-disaster stage 0–12 h, most 5G base station storage units adopt a “pre-charging” strategy, actively storing energy during low-risk periods to prepare for post-disaster supply. In the disaster stage approximately 13–24 h, as load uncertainty increases, the storage systems discharge to support load operation. This scheduling pattern highlights the significant spatiotemporal energy-shifting capability of 5G base stations as flexible resources, with typical “peak shaving and valley filling” characteristics. Furthermore, they provide emergency support to islanded areas during disasters, enhancing load flexibility and local self-sufficiency.

To further analyze the energy consumption characteristics of 5G base stations during the resilience enhancement process, the power consumption of internal functional units of the base stations was statistically investigated. The total load of a 5G base station was decomposed into five components: the static power consumption of the BBU, the baseline power consumption of the AAU, the dynamic power consumption of the AAU, the power consumption of air-conditioning equipment, and the power losses associated with power conversion. The statistical results are illustrated in Fig. 16.

Fig. 16.

Power consumption of 5G base station load.

As illustrated in Fig. 16, the disaggregated loads of the 5G base station exhibit pronounced differences before and after the disaster event. The static power consumption of the BBU remains stable throughout the entire dispatch period, maintaining a nearly constant level of approximately 0.30 kW. Similarly, the baseline power consumption of the AAU remains steady at around 0.60 kW, serving as the fundamental load for base station operation. In contrast, the dynamic power consumption of the AAU shows significant fluctuations in response to variations in communication traffic. Its power ranges from 0.15 kW to 0.38 kW, with a marked increase after the disaster. A peak value of 0.38 kW is observed at 14:5 h, after which the power remains at a relatively high level of 0.32–0.36 kW during 16–18 h, which is significantly higher than the pre-disaster range of 0.15–0.26 kW. The power consumption of the air-conditioning system varies between 0.10 kW and 0.23 kW and exhibits a slight upward trend after the disaster. After 13:00, it remains above 0.20 kW and reaches relatively high levels during 14–16 h, reflecting the impact of equipment heat dissipation and environmental changes on the energy consumption of the base station during disaster conditions.

Analysis of resilience enhancement effects in distribution networks

Based on the previously defined scenarios, a comparative analysis of the system load shedding rates in Scenarios 1–3 is conducted, and the results are shown in Fig. 17.

Fig. 17.

System load shedding rates under scenarios 1–3.

As shown in the comparison results in Fig. 17, under Scenario 1 without coordinated resource utilization, the system load outage rate remains relatively stable within the range of 0.52–0.60 p.u., with an evident peak during the disaster period. After introducing coordinated dispatch of distributed generation and electric vehicles in Scenario 2, this indicator is reduced to approximately 0.22–0.28 p.u.; In Scenario 3, where 5G base station energy storage is further incorporated with capacity partitioning and dispatch, the outage rate is further decreased to about 0.14–0.20 p.u. Based on the average level over the entire time horizon, Scenario 3 achieves an approximate reduction of 65% compared with Scenario 1 and still provides about a 30% improvement compared with Scenario 2.

Furthermore, Scenario 4 considers the configuration of 5G base stations and mobile storage in islanded areas, with the resulting load shedding rates illustrated in Figs. 18 and 19.

Fig. 18.

Load shedding rate without 5G base station mobile storage.

Fig. 19.

Load shedding rate with 5G base station and mobile storage.

As shown in Fig. 18, without 5G base stations and mobile storage, the islanded areas exhibit insufficient self-supply capability. The load shedding rate remains high during the disaster period, peaking at nearly 0.8, indicating that flexible resources are limited by their spatial distribution and cannot effectively support loads across line outages.

Figure 19 presents the results after deploying 5G base stations at nodes 15 and 33 and relocating mobile storage units to critical islanded nodes post-disaster. This configuration significantly enhances local autonomous supply capacity, reducing the load shedding rate throughout the time horizon, with the peak decreasing to about 0.65 and the average reduced by more than 15%. These results verify the feasibility of integrating 5G base stations and mobile storage, providing a practical pathway to improve the resilience of islanded areas under extreme disasters.

The values of the integrated objective function for the four scenarios settings are summarized in Table 6.

Table 6.

Comparison of distribution network resilience enhancement effects.

Scenario	Critical load loss of the system (MW)	Comprehensive economic cost (104CNY)	Integrated objective function
Scenario 1	83.91	24.51	66.09
Scenario 2	20.64	5.52	16.10
Scenario 3	12.96	5.64	10.76
Scenario 4	8.06	5.17	7.19

As indicated by the results in Table 6, the weighted-sum method is employed in this study to unify the objectives of minimizing critical load loss and minimizing economic cost. Under the corresponding weighting scheme, the obtained comprehensive objective values effectively reflect the resilience improvement of the distribution network under different strategies.

In Scenario 1, where no coordinated measures are adopted, the critical load loss reaches 83.91 MW and the economic loss amounts to 24.51×10⁴CNY, yielding the highest comprehensive objective value and indicating the poorest system resilience. In Scenario 2, after introducing conventional flexible resources, the critical load loss is reduced by 63.21 MW and the economic loss is decreased by 18.99×10⁴CNY, resulting in a 49.99% improvement in the comprehensive objective value. In Scenario 3, with the further incorporation of 5G base station energy storage, the critical load loss is reduced by 70.95 MW, corresponding to an 84.6% decrease compared with Scenario 1, while the economic loss is reduced by 18.87×10⁴CNY, representing a 76.9% reduction. The comprehensive objective value is improved by 83.7%. In Scenario 4, where 5G base stations and mobile energy storage are coordinated to support islanded areas, the critical load loss is further reduced to 8.06 MW, corresponding to a 90.4% reduction relative to Scenario 1, and the economic loss decreases to only 5.17×10⁴CNY, representing a 78.9% reduction. The overall improvement in the comprehensive objective value reaches 89.1%, achieving the best performance among the four scenarios.

The results clearly demonstrate that, with the increasing degree of coordination among flexible resources, both load recovery capability and economic performance of the distribution network are significantly enhanced. In particular, the introduction of 5G base station energy storage plays a critical role in improving system resilience.

Conclusion

To address severe challenges caused by extreme typhoon disasters, including line outages and sharp renewable power reductions, this study proposes a resilience enhancement strategy for distribution networks integrating 5G base stations and multiple flexible resources. Based on systematic modeling, optimization, and simulation, the main conclusions are as follows:

1. Disaster scenario modeling: A comprehensive scenario generation framework incorporating typhoon wind speed, dynamic line outage probability, and stochastic wind–PV curtailment is developed. By combining Latin hypercube sampling (LHS) and K-means++ clustering, two representative scenario types, namely reconfigurable grid-connected and islanded autonomous modes, are effectively extracted.

2. Flexible resource modeling and 5G storage partitioning: Models of wind power, photovoltaic generation, EV charging stations, and mobile energy storage are established. A capacity partitioning scheme for 5G base station energy storage is proposed, explicitly distinguishing backup and dispatchable capacities, enabling simultaneous communication reliability and participation in power system dispatch.

3. Coordinated resilience-oriented optimization: A coordinated optimization framework integrating multi-type flexible resources and network reconfiguration is developed. Through joint scheduling of 5G base stations, distributed generation, electric vehicles, and mobile storage, the strategy effectively mitigates renewable curtailment and line faults while reducing economic losses and enhancing system self-healing and recovery capability.

4. Case study validation: Simulation results demonstrate significant resilience improvements. Critical load loss is reduced from 83.91 MW to 12.96 MW, achieving a reduction of 84.6%, and economic loss is reduced from 24.51×104CNY to 5.64×104CNY, achieving a reduction of 76.9%, with an improvement of 83.7% in the comprehensive objective. Further coordination of 5G base stations and mobile energy storage reduces critical load loss to 8.06 MW and economic loss to 5.17×104CNY, and improves the comprehensive objective by 89.1%. These results confirm the effectiveness of the proposed strategy in enhancing local autonomy, self-healing capability, and multi-resource coordination capability.

Despite the effectiveness of the proposed method, several limitations remain:

1. The disaster scenarios do not consider the spatiotemporal evolution of typhoon trajectories, and the actual impacts of disasters may vary dynamically across time and regions;

2. The availability of mobile energy storage and electric vehicles is affected by practical response times and traffic conditions, and their uncertainties have not been modeled in a robust manner;

3. Communication networks themselves may be damaged under extreme disasters, and the impact on dispatch communication has not been fully investigated.

Future research can be further conducted in the following directions:

1. Development of spatiotemporal evolutionary disaster scenario models incorporating the dynamic propagation of typhoon paths and their impacts;

2. Investigation of multi-network coordinated resilient dispatch mechanisms among power, communication, and transportation systems;

3. Exploration of multi-agent game-theoretic and incentive mechanisms among 5G base station energy storage, user-side storage, and virtual power plants.

Author contributions

H.W: Conceptualization, Data Curation, Formal Analysis, Writing-Original Draft, Software, Funding Acquisition Jz.G: Formal Analysis, Investigation, Writing - Original Draft, Software Yq.Z: Resources, Writing - Original Draft J.G: Resources, Supervision, Formal Analysis Cg L: Visualization, Writing - Review & Editing C L: Visualization, Data Curation.

Funding

This project is funded by Jilin Province’s Science and Technology Development Program – Research on Technologies for Enhancing the Disaster Resilience of Distribution Networks with Multiple Types of Emergency Energy Storage (project number: 20250203006SF).

Data availability

The data that support the findings of this study are available within the article. Additional data not presented here are available from the corresponding author upon reasonable request.

Declarations

Competing interests

The authors declare no competing interests.