Distribution network gray-start and emergency recovery strategy with pumped storage unit under a typhoon☆

0 Introduction

Typhoons can damage distribution network infrastructures or cause blackouts under the threat of global warming [1,2].Thus far, extensive research has been conducted on emergency recovery strategies after a typhoon, with some of them focusing on black-start strategies after a blackout event [3,4].However, the entire distribution network may not always suffer from a blackout.In most scenarios,only fault-distributed generation(DGs)are cut off,while the normally operating DGs remain connected.Therefore, we define a partial outage state in the distribution network as a ‘‘gray state” and investigate the graystart strategy as well as distribution network emergency recovery approach in detail.

Current studies extensively focus on pre- and postdisaster scenarios.Predicting wind and photovoltaic generation is important in pre-disaster research.Predicting wind and photovoltaic generation before a typhoon strike can aid in post-disaster emergency recovery scheduling.The research on wind and photovoltaic generation prediction based on a hybrid model that combines machine and deep learning approaches was reviewed in[5].In[6],the authors proposed a wind and photovoltaic generation prediction method based on deep belief networks with a swarm spider optimization method.Further, a system that combines decomposition algorithms and deep learning for ultrashort-term wind and photovoltaic generation prediction was proposed in in [7].In [8], the authors collected complex time-series data on wind and photovoltaic generation and used an ensemble learning model for predicting wind and photovoltaic generation.However, these studies focus on the prediction model for hybrid algorithms, with limited consideration of the lack of meteorological and environmental information.Experimental results suggest that the prediction accuracy of the integrated learning model was significantly improved compared to that of the single model.In[9,10],the authors applied wind power generation forecasting data to the day-ahead scheduling of an integrated energy system to improve the energy utilization and minimize the effect on the fishnet.Further,in[11],the authors applied wind and photovoltaic generation predictions for managing the energy of distribution networks.These studies are primarily used for the distribution management of integrated energy.However, research on wind and photovoltaic generation predictions during typhoons is lacking.

Thus far, several post-disaster studies have been conducted on emergency recovery strategies.Mobile emergency generators (MEGs) and DGs such as wind power generators, photovoltaics, and fuel cells play important roles in the recovery of post-disaster distribution networks.In [12,13],the authors proposed a two-stage distribution network emergency recovery optimization model that considers DGs and power grid reconstruction.Further, in [14], the authors considered the effect of DGs,MEGs, and remote-controlled switches on the emergency recovery of the distribution network after a disaster, further improving the resilience of the distribution network.In addition, in [15], the authors comprehensively considered the cooperative recovery of the distribution network between the recovery crew and DGs, and they developed resilient and dynamic emergency recovery schemes.In the aforementioned studies, the output of each DGs was deterministic and unaffected by typhoons.Wind and photovoltaic (PV) generation prediction have not been extensively considered for post-disaster distribution network emergency recovery.

Although many studies have focused on the recovery of post-disaster distribution networks, the distribution network may still experience an outage after a typhoon, and yet,there are relatively few studies on gray-start networks.In [16], the authors proposed a black-start recovery scheme for elastic distribution systems based on model predictive control (MPC).In [17], the authors proposed a black-start recovery scheme for distribution networks based on multi-type DGs.In [18], the authors proposed a dynamic black-start and recovery optimization strategy for a distribution network with automatic sectionalization and flexible reconfiguration.Further, in [19], the authors formulated a scientific and reasonable black-start zoning scheme to accelerate the power-grid recovery process and reduce the power-outage time of users.The aforementioned studies did not consider the various types of DGs that can be quickly integrated into the fault distribution network.Therefore, the probability of a total blackout in the distribution network is relatively low.Researchers have proposed the concept of a gray start for the recovery of gray-state distribution networks.Selecting a reliable gray-start power source is a top priority when developing a gray-start scheme, and it can be subsequently used to restore other units and loads.In[20],the authors proposed a gray-start strategy for AC/DC microgrids in a partially outage state.In[21],the authors proposed a gray-start grid reconstruction model for elastic microgrids, which improved the stability and speed of resilient microgrid recovery.However, there are few studies that focus on the use of pumped storage units as gray-start power sources for participating in distribution network emergency recovery during typhoons.The pumped-storage unit can bear an emergency backup,and therefore,it can effectively improve the safe and stable operation of the power system.This is an important part of a new round of power system reform, with new energy being the focus [22].Therefore, it has advantages of a fast response and stable output to the gray-start power source.

Current studies focus on wind and photovoltaic generation prediction without typhoon disasters and they rarely incorporate them in post-disaster distribution network emergency recovery.Further, recovery studies considered the distribution network in the gray state to start with the pumped storage unit.Therefore, we propose a twostage emergency recovery framework for distribution networks.The first stage is the prediction of pre-disaster wind and photovoltaic generation.The second stage is the postdisaster emergency recovery strategy and gray-tart based on the pumped storage unit.The main contributions of this study are as follows:

1) A pre-disaster wind and photovoltaic generation prediction model is proposed for providing guidance for post-disaster emergency recovery scheduling during typhoons.

2) A post-disaster distribution network emergency recovery strategy is developed to consider the thermal power generating unit, MEGs, and DGs.

3) A gray-start based on a pumped storage unit is proposed as an important first step in the emergency recovery strategy.

The remainder of this paper is organized as follows:Section II describes the framework of this research.Section III introduces the generation output models.Section IV introduces the emergency recovery scheduling model that considers wind and photovoltaic generation prediction and the gray-start of the pumped storage unit.Section V considers the improved IEEE 33 node system in 2022 super typhoon ‘‘Muifa” in Zhoushan, Zhejiang,China as an example to verify the superiority of the proposed method.Finally, Section VI summarizes the study and introduces future research directions.

1 Research framework for emergency recovery strategies in distribution networks

In this study,wind and photovoltaic generation prediction were combined with the gray start of a pumped storage unit.Further,a strategy is proposed for the emergency recovery of distribution networks during typhoons.This process comprises two stages, pre-disaster wind and photovoltaic generation prediction and distribution network gray-start based on pumped storage unit.Fig.1 shows the research framework for emergency recovery strategies in distribution networks during typhoons.

(1) The first stage involves pre-disaster wind and photovoltaic generation prediction.A phase-space reconstruction technique is used for processing the time series of the historical data of the wind and photovoltaic generation.Subsequently, a wind and photovoltaic generation prediction model based on a stacking integrated model is established.The performance of the prediction model is evaluated using error and fitting indicators.Finally, a comparative analysis is conducted between the predicted and actual wind and photovoltaic generation values to verify the accuracy of the proposed method.

(2) The second stage is a gray-start distribution network based on a pumped storage unit.An emergency recovery scheduling model is established considering wind and photovoltaic generation prediction and the pumped storage unit as a gray-start power source in recovery.The model is solved using the CPLEX solver.The improved IEEE 33 node system under the 2022 super typhoon‘‘Mufia”in Zhoushan,Zhejiang,China is considered as an example.The recovery crew and MEG scheduling strategies under different scenarios are obtained, and the effect of each equipment output on the emergency recovery of the distribution network is analyzed.The case study shows that the proposed method has the shortest recovery time and lowest load loss.

2 Pre-disaster wind and photovoltaic generation prediction model

Fig.1 Research framework for emergency recovery strategy in distribution networks during typhoons.

The difficulty in predicting wind and photovoltaic generation during typhoons lies in the difficulty of obtaining the required meteorological and environmental information.However, the prediction accuracy of the model requires further improvement.Given these issues, this study proposes a phase-space reconstruction and stacking integrated model for predicting wind and photovoltaic generation during typhoons.

Phase-space reconstruction technology based on chaos theory can ensure that the reconstructed phase space based on the original time series is topologically equivalent to the original dynamical system[23,24].Phase-space reconstruction makes it possible to perform more accurate load forecasting without weather information or multifactor data.The original time-series data C is given by

where Ci （i=1,2,3,...,n） represents the output value of wind power generation or photovoltaic power at the i-th moment, and n represents the length of the original time series data.

The mutual information method is used to obtain the parameters and the pseudo-nearest neighbor method is used for obtaining embedding dimensions [25].The delay time required for phase space reconstruction after embedding the dimension is determined.The reconstructed data are

where C′,τ, and m represent the reconstructed data, delay time, and embedding dimensions, respectively.

Compared to a single model, the stacking ensemble model generalizes output results of multiple models for improving the overall prediction accuracy.The model includes two layers: a base-learning layer and a metalearning layer [26].Random forest (RF) [27], adaptive boosting (AdaBoost) [28], support vector machine(SVM) [29], light gradient boosting machine (LightGBM)[30], and extreme gradient boosting (XGBoost) [31] were used as the base learners.Long short-term memory(LSTM)networks have been widely used in wind and photovoltaic generation forecasting with good predictive performance [32,33].LSTM was selected as the meta-learner for predicting the final wind and photovoltaic generation output data.The prediction process is illustrated in Fig.2.

The performance of the prediction model is evaluated using error and fitting indicators, and the time indicator reflects the response speed of the prediction model.The error indicators are the mean square error (MSE), root mean square error (RMSE), and mean absolute error(MAE) [34].The time indicator represents the time from the start to the end of the prediction model.

Fig.2 Wind and photovoltaic generation prediction process under typhoons.

3 Post-disaster emergency recovery scheduling with a gray start

3.1 Objective function

In the maintenance process of a distribution network,the loss of user load and maintenance time are important indices to measure maintenance efficiency.This study aims to minimize the load loss under emergency recovery scheduling as the optimization objective to reduce losses caused by power outages for users.

where , T, N, ωi, and pi,t represent the minimum load loss under emergency recovery scheduling as the optimization objective, time set, set of power-loss nodes, weight of the load-loss node, and amount of load loss of node i at time t.

3.2 Generation output model

3.2.1 Gray-start pumped storage unit model

The output of wind and photovoltaic generation is not suitable for providing the starting power during a typhoon.Pumped storage units have advantages such as fast response and stable output [35], and therefore, a pumped storage unit is selected as the gray-start power source.The pumped-storage unit can be used as both a generator and an electric motor.The power generation and pumping constraints are

where Pc,p（t）, W and Uc,p（t） represent the pumping power of the pumped storage unit at time t, power of the water pump motor, and 0-1 variable, respectively.If it is 1, the unit pumps water during this period; otherwise, it does not.Pc,g（t） represents the active power generated by the pumped storage unit and ensures that power generation and pumping states do not occur simultaneously.

Assuming the storage capacity is sufficient, the output of the pumped storage unit as a gray-start power source must satisfy

whereand Kc represent the lower limit of the pumping power of the pumped storage unit,upper limit of the pumping power of the pumped storage unit,lower limit of the active power output of the pumped storage unit, upper limit of the active power generated by the pumped storage unit,and climbing rate of the pumped storage unit, respectively.

3.2.2 Thermal power generating unit model

The thermal power-generating unit is initiated by a gray-start power source in the case of a power outage in the main distribution network.This study adopts a typical start climb curve, and at time t1, the auxiliary power is electrified.After tc,the unit is connected to the power grid at t1+tc and provides active power at the climbing rate of Kh.

where and represent the upper and lower limits of the thermal power generating unit active power,respectively, and and represent the upper and lower limits of the thermal power generating unit reactive power,respectively.The change in active power cannot be greater than the climbing slope.

3.2.3 Other generation output unit models

The energy storage system is equipped with wind and photovoltaic generation storage systems to minimize the impact of the power grid.The energy storage system constraints are described in references [36,37].Fuel cells and gas generators provide temporary power for important users; the constraints are presented in [38,39].

3.3 Constraints

3.3.1 Recovery crew constraints

All recovery crews depart from the warehouse and ultimately return to the warehouse to ensure that the recovery crew x leaves immediately after the recovery of fault point m.

where γm,n,x,γm,n,x,θ,D,and X represent whether the recovery crew x departed from the fault point m to fault point n,whether the maintenance team x departed from warehouse D to the fault point n, set of fault points, collection of warehouses,and collection of recovery crews,respectively.

Each fault point can only be recovered by one recovery crew, i.e.,

where ηm,x represents whether the fault point m can be recovered using recovery crew x.

The demand for materials during the road recovery process is less than the amount of materials carried by the recovery crew, i.e.,

where ϑm and ϑx represent the amount of materials required to recover the fault point and total amount of materials carried by the recovery crew x, respectively.

3.3.2 MEGs constraints

All MEGs depart from the warehouse and eventually return to the warehouse to ensure that the MEGs y leave immediately after charging the power access point u.

where γu,v,y, γD,u,y, ρ, D, and Y represent whether MEGs y depart from the power access point u to the power access point v, whether MEGs y depart from warehouse point D to power access point u, set of power access points, warehouse collection, and emergency generator set,respectively.

Each power access point u can only be accessed by one MEGs y, that is,

where ηu,y represents whether the power source access point u is connected to MEGs y.

Further, the MEGs need to meet the power constraints

where PMEG,max and QMEG,max represent the upper limits of the active and reactive powers of the MEGs, respectively.

3.3.3 Recovery time constraints

The time constraints for the recovery crew to arrive at the fault point, start recovering from the fault point, and leave the fault point on the way are expressed as

where M, Tm,x, tm,x, tm,n,x, fm,t, and lm,t represent a positive number, time when the recovery crew x reaches the fault point m,dwell time of the recovery crew x at the fault point m,travel time of the recovery crew x from fault point m to fault point n, whether there is a recovery crew arriving at the fault point m at time t,and whether there is a recovery crew leaving the fault point m at time t, respectively.The arrival and departure of the recovery crew must satisfy the requirement ∑fm,t =∑lm,t.

The time constraints for the MEGs to arrive at the power access point, start connecting, and leave the power access point are

where Tu,y, zu,y, tu,v,y, fu,t, and lu,t represent the time when MEGs y arrive at the access point u, dwell time of MEGs y at the access point u,travel time of MEGs y from access point u to access point v,whether there is a MEG arriving at the access point u at time t, and whether there is an MEG leaving the access point u at time t, respectively.The arrival and departure of the MEGs should meet the requirements ∑fu,t =∑lu,t.

3.3.4 Distribution network operation constraints

The nonlinear term caused by the node power-balance constraints caused by system topology changes during the recovery process is linearized.This study uses the Dist-Flow model for linearizing the power flow equation.The power flow constraint of the branch is

whererepresent the active power of the non-fault line mh at time t, active power of the faulty line nm at time t,active power provided by the DGs at time t to the fault point m,active power provided by the emergency generator MEG at time t to the fault point m, active power of fault point m at time t,and load loss of fault point m at time t, respectively.Further,Q represents the corresponding reactive power variable and will not be elaborated in detail.The DGs includes fuel cells and gas generators.

The voltage constraint of the node is

where V n,t,V m,t,Znm,t,rnm,xnm,,and represent the voltage of node n at time t, voltage of node m at time t,available binary variable of line nm at time t, resistance of line nm,reactance of line nm,lower limit of node n voltage, and upper limit of node n voltage, respectively.

The power and node load loss constraints are

where represents the upper limit of the nm active power of the faulty line; and represent the active and reactive powers of the load node m at time t, respectively; and the amount of load loss should not exceed this value.

3.3.5 Distribution network radial topology constraints

The operation of the distribution network meets strict radiation constraints.We use a single-compatibility flow model for constraining the radio topology of the distribution network.Among these, the line-flow constraint is

The total number of closed lines is equal to the total number of nodes minus the number of root nodes.

where Nbus and γm,t represent the total number of root nodes and a binary variable of the root bus state,respectively.

The constraint of node inflow and outflow balance,and whether the node is at both ends of the faulty line is given by

where Fmh,t represents the virtual load on line mh at time t;γm,t,μm,t,and χm,t represent binary variables; Gm represents the DGs installation status; and Gm =1 indicates that bus m is installed with DGs, otherwise, Gm =0.Each binary variable must satisfy the following requirements.

4 Case study

4.1 Case background

Assuming it is an improved IEEE 33 node system, we consider the distribution network in Zhoushan, Zhejiang,China [40].This paper chose Typhoon No.12 ‘‘Muifa,”which landed in Zhoushan, Zhejiang, China in 2022, as the background for calculation.It began at 17:00 on September 14, 2022.The track of Typhoon 12 ‘‘Muifa”is shown in Fig.3.

We consider the IEEE 33 node system as a representative system because the distribution network in Zhoushan,Zhejiang, China is similar to the improved IEEE 33 node system network.An improved IEEE 33 node system topology diagram with the load classification, fault points, and equipment connection relationships is shown in Fig.4.Distribution lines 8-21, 9-15, 12-22, 18-33, and 25-29 are equipped with remote-controlled switches.

4.2 Pre-disaster wind and photovoltaic generation prediction

Fig.3 Typhoon No.12 ‘‘Muifa” track.

The historical data of wind and photovoltaic time series is the real data in Zhoushan, Zhejiang, China during the landing of Typhoon ‘‘Muifa.” The starting and ending times of the historical wind and photovoltaic generation data are from 0:00 on September 11, 2022, to 12:00 on September 15, 2022.The time step for the historical data on wind and photovoltaic generation is T = 15 min because of the time span of T = 15 min during the postdisaster emergency recovery phase.Historical data for wind and photovoltaic generation are provided by the Electric Power Research Institute of State Grid Zhejiang Electric Power Co., Ltd.The delay time is obtained by mutual information method τ embedding dimension m obtained by the pseudo nearest neighbor method shown in Table 1.

Historical wind and photovoltaic generation data in the time series are reconstructed.The reconstructed data for the stacking integrated model are used to predict wind and photovoltaic generation data during typhoon landfall.The performance evaluation indicators are listed in Table 2.

Fig.4 Typhoon No.12 ‘‘Muifa” landing path.

Table 1 Delay time and embedding dimension.

Data typeτm Wind power generation data93 Photovoltaic data454

Table 2 indicates the MSE, RMSE, and MAE error indicators of the wind and photovoltaic generation prediction data are relatively low.Therefore, the stackingintegrated model has a good prediction performance.The typhoon circle impact has an effect on the distribution network of Zhoushan, Zhejiang, China at 14:00 on September 14th.Therefore, starting from 14:00 on September 14th, the prediction of the wind and photovoltaic generation data for the next 32 T period is shown in Fig.5.

Fig.5 shows that the stacking integrated modelpredicted data for wind and photovoltaic generation are very close to the actual values.The method proposed in this paper can accurately predict the power of wind and photovoltaic generation even in the absence of key meteorological and environmental information under typhoon disasters.

4.3 Post-disaster emergency recovery strategies analysis

Four different scenarios are constructed for the simulation analysis to verify the superiority of the emergency electricity recovery model:

(1) Scenario 1 considers wind and photovoltaic generation prediction and the pumped storage unit graystart (method used in this study).

(2) Scenario 2 considers wind and photovoltaic generation predictions without a gray start.

(3) Scenario 3 considers wind and photovoltaic generation prediction and the gray start of the pumped storage unit without energy storage equipment.

(4) Scenario 4 considered a gray start without wind and photovoltaic generation predictions.

As listed in Table 3, we obtained the completion time,load loss, and simulation model running time for emergency recovery.

The emergency recovery method proposed in this study had the shortest completion time.The minimum load losses are listed in Table 3.The method proposed in this study has advantages and a short simulation time, which meets the requirements of practical applications.A detailed analysis of the emergency recovery strategies for the four scenarios is provided below.

4.3.1 Post-disaster emergency recovery strategy

Scenario 1 considers wind and photovoltaic generation output prediction and pumped storage unit gray start,which includes energy storage equipment.The scheduling strategy for the recovery crew and MEGs is shown in Fig.6(a),and the output of each piece of the power equipment is shown in Fig.6 (b).

Scenario 1 completed the emergency recovery of the distribution network after 27 periods of 405 min, which results in a load loss of 386579.17 kWh.Fig.6 (a) shows that the itinerary of recovery crew 1 is 6—⑤—①—④—②, recovery crew 2 is 6—⑥—③—⑦, and itinerary of MEGs is 6-22.Fig.6 (b) shows that the pumped storage unit provides a gray-start power source for the power consumption of the thermal power-generating unit at T1, and the thermal power-generating unit is connected to the power grid for providing power for the distribution network at T6.In the process of recovering electricity, more electricity is stored in energy storage equipment when wind and photovoltaic generation are high.Fuel cells no longer provide power to the distribution network in the later stages with the recovery of the distribution network.

4.3.2 Pumped storage unit gray-start impact

Scenario 2 considers wind and photovoltaic generation predictions without a gray start.The scheduling strategy for the recovery crew and MEGs is shown in Fig.7 (a),and the output of each piece of the power equipment is shown in Fig.7 (b).

Scenario 2 completes the emergency recovery of the distribution network after 30 periods of 450 min, resulting in a load loss of 669399.68 kWh.We can only wait for the main network to recover power in the distribution network because of the lack of a pumped storage unit as a graystart power source, which results in an increase in the distribution network recovery time and load loss.

4.3.3 Energy storage impact

Scenario 3 considers wind and photovoltaic generation prediction and the pumped storage unit gray-start but without wind and photovoltaic generation energy storageequipment.The scheduling strategy of the recovery crew and MEGs is shown in Fig.8 (a), and the output of each piece of the power equipment is shown in Fig.8 (b).

Table 2 Model evaluation indicator results.

Evaluation indicatorWind power generation dataPhotovoltaic data MSE0.990.78 RMSE1.211.15 NAE0.520.47 Time(s)0.490.63

Fig.5 Wind and photovoltaic generation prediction data.

Scenario 3 completes emergency recovery and distribution network recovery after 28 periods of time (420 min)with a load loss of 506258.98 kWh.Fig.8 (a) shows that the itinerary of recovery crew 1 is 6—⑤—①—④, itinerary of recovery crew 2 is 6—⑥—⑦—②—③, and itinerary of MEGs is 6-22.Compared to Scenario 1, the number of recovery points for Recovery Crew 1 decreased,whereas Recovery Crew 2 had more recovery points.Fig.8(b) shows that wind and photovoltaic generation provide full power for the recovery of the distribution network.The outputs of the wind and photovoltaic generation are shown in Fig.9 to analyze the impact of wind and photovoltaic generation on the recovery of the distribution network.

Fig.9 shows that there is a large amount of abandoned air starting from T9 because of the lack of wind power generation and energy storage equipment.There is a significant shortage of electricity during the recovery process of the distribution network because of the lack of photovoltaic and energy storage equipment.Wind and photovoltaic generation plays an important role in the emergency recovery of post-disaster distribution networks.The distribution network department must reasonably match the energy storage equipment for wind and photovoltaic generation.

Fig.6 Scenario 1 analysis of the emergency recovery power strategy.

4.3.4 Wind and photovoltaic generation impact

Scenario 4 considers the pumped storage unit gray-start but without wind and photovoltaic generation.The scheduling strategy for the recovery crew and MEGs is shown in Fig.10 (a), and the output of each piece of the power equipment is shown in Fig.10 (b).

Scenario 3 required 32 time periods (480 min) to complete emergency recovery and distribution network emer-gency recovery because of the lack of wind and photovoltaic generation to provide power support for the distribution network, thereby resulting in a load loss of up to 710,650 kWh.Compared with the method proposed in this study, the recovery time and load loss were increased significantly.Fig.9 (a) shows that the itinerary of recovery crew 1 is 6—⑥—⑦—④, itinerary of the recovery crew 2 is 6—①—③—②—⑤, and itinerary of the MEGs is 6-22—8.Thus, the lack of wind and photovoltaic generation to provide power support for important loads can increase the burden of the MEGs recovery.

Table 3 Analysis of simulation results in different scenarios.

Ta = 15 min

Scenario typeRecovery completion time/TLoss of load/kWhRuntime/s Scenario 127386,579.1748.312 Scenario 230669,399.683,984.355 Scenario 332710,650.001,350.988 Scenario 428506,258.98145.785

Fig.7 Scenario 2 analysis of emergency recovery power strategy.

Fig.8 Scenario 3 analysis of emergency recovery power strategy.

The method proposed in this study has the shortest recovery time and least load loss.As a gray-start power source, the pumped storage unit reduces load loss in the distribution network.The analysis indicated that wind and photovoltaic generation predictions play a guiding role in formulating strategies for the emergency recovery of post-disaster distribution networks.Energy storage equipment plays an important role in the recovery of power in a distribution network, and the power grid department should configure energy storage equipment for wind and photovoltaic generation.

4.4 Remote-controlled switch status analysis

The improved IEEE 33 node system has tie lines 34-38,which are controlled by remote-controlled switches.We obtained the state of each remote-controlled switch for the four scenarios shown in Fig.11.

When the remote-controlled switch is 1, it was closed,and when the remote-controlled switch was zero, it was open.Fig.11 shows that in Scenario 1, each remotecontrolled switch changed 32 times.In scenario 2, each remote-controlled switch was switched 21 times.In Scenario 3, each remote-controlled switch was changed 38 times.In Scenario 4, each remote-controlled switch was changed 23 times.Therefore, a remote-controlled switch plays an important role in emergency recovery and can effectively improve the efficiency of power source recovery for users.

5 Conclusion

Fig.9 Output of the wind and photovoltaic generation energy storage systems.

Fig.10 Scenario 4 analysis of emergency recovery power strategy.

Fig.11 Remote-controlled switch status.

This study proposes an emergency recovery strategy for the gray-start distribution networks based on a pumped storage unit with wind and photovoltaic prediction during typhoons.Before the typhoon makes landfall, the lack of meteorological and environmental information can provide accurate predictions of wind and photovoltaic generation data.The post-disaster distribution network emergency recovery strategy considers the gray start of the pumped storage unit an important first step.The load loss decreased by 57.75 % for the pumped-storage unit gray start impact.The wind and photovoltaic generation impacts decreased by 76.36 %.Therefore, the case study verified that the proposed method had the shortest recovery time and least load loss.

However, the economy of each equipment output should be investigated in future studies because each equipment output in emergency recovery scheduling can be designed for each generation location and capacity.The author will conduct further research on how fluctuations in wind and photovoltaic power outputs affect emergency resource dispatch and load loss in future.

CRediT authorship contribution statement

Zhenguo Wang: Data curation.Hui Hou: Writing -review&editing,Methodology.Chao Liu:Software.Shaohua Wang: Data curation.Zhengtian Li: Validation.Xiangning Lin: Validation.Te Li: Data curation.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This work was supported in part by the National Natural Science Foundation of China (52177110), Key Program of the National Natural Science Foundation of China (U22B20106, U2142206), Shenzhen Science and Technology Program (JCYJ20210324131409026), the Science and Technology Project of the State Grid Corporation of China (5200-202319382A-2-3-XG) and State Grid Zhejiang Elctric Power Co., Ltd.Science and Technology Project(B311DS24001A).