Key technologies and applications of cloud-edge collaborative integrated low and medium voltage distributed photovoltaic coordination control

The large-scale distributed photovoltaic access leads to overvoltage, equipment overload, and difficulty in on-site consumption, which causes power to be sent back to the high voltage level, posing hidden dangers to the safe operation of the distribution network. This article proposes optimizing the medium-voltage distribution network at the cloud side with the goal of maximum consumption and operating cost, forming a rolling strategy adjustment based on load forecasting within the day. At the edge side, the distributed photovoltaic control in the low-voltage distribution network within the jurisdiction is adjusted based on the daily adjustment strategy issued by the cloud side, completing the optimization and regulation of distributed photovoltaics on a distribution substation unit basis to achieve the control and integration of distributed power sources and consumption, coordination between operation and side, and coordinated control in the integrated medium and low-voltage zoning, achieving autonomous distribution substations and feeder lines. Field applications have proven that this solution solves problems such as line and transformer overload, and voltage quality caused by distributed photovoltaic power generation.


Introduction
Under the global energy transition, Chinese "carbon peaking and carbon neutrality" goals, and the context of county-wide photovoltaic construction, renewable energy, especially photovoltaics, is experiencing rapid development [1] .According to the "2022 China Distributed Photovoltaic Industry Development White Paper", between January and September 2022 in China, new installations of distributed photovoltaics for industrial and commercial use reached 18.94 GW, an increase of 296% year-on-year.There were also 16.95 GW of new photovoltaic installations for users, a year-on-year increase of 42%.A large number of distributed photovoltaics are connected to the distribution network through transformer substations, leading to issues such as voltage exceeding limits [2][3] , three-phase imbalance [4] , equipment overload [5] , and reverse power flow [6] in the low-voltage distribution network.At the same time, when the users' distributed power generation is too high, it will be sent back to the medium-voltage distribution network through the transformer substation, impacting the economic and safe operation of the medium-voltage distribution network and increasing the difficulty of the distribution network regulation management [7][8] .To fully absorb distributed photovoltaic energy and address the impact of high penetration rate distributed photovoltaic on the power distribution network, Ding et al. establish a "source-grid-load-storage" collaborative optimization model for the power distribution substation area [9] .This approach enhances the absorption capacity of distributed photovoltaic (DPV) from four aspects: reactive power control of photovoltaic inverters on the "source" side, Static Var Generator (SVG) reactive power compensation on the "grid" side, electric vehicle load control on the "load" side, and energy storage device regulation on the "storage" side.Cai et al. propose a "centralized -local" day-ahead and intraday adaptive control architecture [10] , aiming to resolve issues such as transformer overload and voltage exceeding limits in the distribution transformers.Kang et al. use a sequential quadratic programming algorithm to optimize the reactive power output and active power reduction of distributed photovoltaic inverters [11] , achieving a three-phase balance and maximizing the absorption of photovoltaic generation.
In response to the problems brought by the high penetration rate of distributed photovoltaics, current proposals mainly focus on the location, capacity, regulation strategies, and devices for photovoltaic access in low-voltage distribution networks around distribution substations.This article proposes a coordinated control solution featuring cloud-edge collaboration, integration of medium and low voltage, and hierarchical and zonal control, aiming to achieve autonomous operation of distribution substations and feeders.Specifically, optimized operation of the medium-voltage distribution network is conducted at the cloud side, to maximize the accommodation of distributed photovoltaics and minimize operational costs, based on rolling strategies adjusted according to load forecasts within the day.Meanwhile, at the edge side, adjustment strategies directed from the cloud side are implemented to regulate distributed photovoltaics in the local low-voltage distribution network, to achieve control optimization and accommodation of distributed energy sources within the distribution substation area.Real-time monitoring and control between the cloud and edge sides are also employed to promptly address issues such as overload, reverse power flow, and voltage violations in the medium and low voltage distribution networks, ensuring grid safety.

Key technology
To solve the problems caused by the grid-scale distributed photovoltaic access and ensure the safe operation of the power grid, the following tasks typically need to be completed in practical dispatch operation: (1) We accept the commands from the superior scheduler and cut off the back-feed power according to the instructions to eliminate the safety risks brought by back-feed power.
(2) We adjust operational modes to transfer load or regulate output, to alleviate the overloading situation for devices both in forward and reverse directions.
(3) We timely adjust various devices and deal with any arising issues to eliminate power quality problems and ensure user satisfaction in the consumption of electricity.
For this purpose, we carry out key technical research and propose solutions for the above application scenarios.Considering the multi-timescale characteristics of distributed photovoltaic power generation forecasting and load forecasting, as well as the response speeds of various types of equipment, using a single timescale for scheduling fails to reflect the impact of prediction errors in distributed power output and load demand, as well as unplanned instantaneous power fluctuations on the optimized operation of an active distribution network.This does not conform to the actual operational conditions of the distribution network.Therefore, an intraday forecast-based rolling strategy, combined with real-time monitoring and control, is proposed for the optimized control of distributed energy resources.

An ultra-short-term intraday rolling strategy
Based on the 2-hour load forecast data and the distributed power generation forecast data, an optimization strategy is formulated with the objectives of maximizing the consumption of distributed power sources, minimizing the operational costs of the distribution grid, and reducing the costs of controllable loads.This strategy takes into account various constraints such as the operation of the distribution grid and equipment specifications to determine the switching states of circuit breakers (CB), the charging and discharging states of energy storage systems, and the power output levels for distributed generation (DG) and controllable loads.The strategy is then dispatched to intelligent integrated terminals.The optimization objective function is calculated according to Formulas (1) to (5): where  ,  , and  are the cost of distributed power source abandoning wind and light, operating costs, and controllable load shedding costs at time interval ∆, respectively; T is daily time periods (the time can be divided into 5 minute intervals, making up a total of 24 intervals);  is distributed power source (including distributed power source, energy storage, back-feed power transformer/micro-grid);  , is the subsidized price per unit electricity generated by the distributed generation;  is the compensation price for shedding a unit of controllable load; P , is the absorbed amount of the distributed power source at time period  under the constraint conditions, while  , is the maximum power output forecasted a day earlier from the distributed power source within the time period t;  , and  , are the unit price at time period  and network loss quantity, respectively;  is the branch where the closed switch is located in the distribution network at time period ;  is the cost of adjusting the number of CB throw-in groups once;  ,  ,  , and  are the branch resistance, active power, reactive power, and voltage of the branch, respectively; ∆ is the duration of each time period.
When optimizing the objective, it is necessary to satisfy the following constraints: load flow constraints, nodal voltage constraints, line capacity constraints, transformer reverse power flow constraints, feeder reverse power flow constraints, capacitor bank switching constraints for groupconnected capacitors, energy storage constraints, distributed generation capacity, and output constraints [9, 12]   .

Autonomous and coordinated control of power distribution substation areas based on cloud-edge collaboration
The edge side receives intraday rolling power dispatch strategies for the grid and distribution substation tie lines from the cloud side, using them as optimization constraints.It controls distributed photovoltaic (PV) systems within the low-voltage power distribution network on a per-substation basis, achieving autonomy within the distribution substation area.The edge side considers the safety operational constraints of the substation area's low-voltage distribution network and the intraday rolling power dispatch strategies for the substation tie lines, with the optimization goal of minimizing energy curtailment and load shedding of adjustable resources within the distribution substation area.It develops scheduling strategies for each distributed PV and energy storage system within the distribution substation area and issues them to the PV inverters and energy storage devices.The optimization objective function is shown in Formula (6): =  ( ) +   (7) where  and  represent the unit costs of load shedding and energy curtailment respectively; ∆ and ∆ are the total amounts of load shedding and energy curtailment at time t;  is the charging device's charging and discharging operational cost;  and  are the coefficients for the maintenance and operation costs of the energy storage system;  is the charging and discharging power of the energy storage system at time t.
When optimizing strategies for distribution substation areas, it is necessary to meet the constraints for secure operation of the electrical grid and equipment.The constraints to be satisfied include [12] : (1) Power balance constraint within the distribution substation area.
(4) Energy storage system charging and discharging power constraints as well as capacity constraints.
(5) Adjustable distributed photovoltaic (PV) operation and capacity constraints.( 6) Interruptible load shedding capacity and timing constraints.( 7) Distribution transformer capacity constraints.(8) The power constraint on the tie line between the grid and the distribution substation.

Real-time monitoring and adjustment
Real-time monitoring and control refer to the real-time collection of the operating conditions of the distribution network, monitoring the status of the medium and low voltage distribution systems, detecting any over-limit voltages, and implementing voltage-reactive power control.When the main distribution station detects an over-limit voltage in the medium voltage distribution network, it will initiate voltage-reactive power control within the cluster to which the limited node belongs.Similarly, when the smart integrated terminal detects an over-limit situation in the low voltage distribution system within the jurisdiction of a distribution substation, it treats the area as a cluster and performs intra-cluster voltage-reactive power control.The voltage-reactive power control process first adjusts the reactive power output of distributed energy resources within a cluster.If the ability to adjust reactive power from distributed energy resources within the cluster is insufficient, then it moves on to curtail the active power output of the distributed energy sources.The final goal is to ensure that the voltage at all nodes within the cluster meets the required limits.The steps of this process are as follows: (1) Selection of regulation node: we select the node with the largest voltage deviation within the cluster as the key node i, and calculate the voltage deviation ∆ .
(2) Selection of regulation equipment: based on the reactive power voltage sensitivity matrix, we select the distributed generation source with the highest reactive power voltage sensitivity value  , which has reactive power regulation capability.We calculate the available reactive power for regulation  and the reactive power  needed for correction.
) Determination of reactive power regulation output: if the distributed generation source's available reactive power for regulation  is greater or equal to  , then we adjust this distributed generation source's reactive output to  .If the distributed generation source's available reactive power for regulation  is less than  , we adjust the distributed generation source's reactive output to  .(4) Check if there are still over-voltage nodes in the cluster: if there are no over-voltage nodes within the cluster, then the voltage-reactive power control for this cluster is concluded.If over-voltage conditions persist and there are still distributed generation sources available for reactive power regulation, we repeat steps (1) to (4).If over-voltage conditions persist and there are no distributed generation sources available to adjust reactive power, we proceed to step (5).
(5) Selection of regulation node: we select the node with the largest voltage deviation within the cluster as the key node , and calculate the voltage deviation ∆ .
(6) Selection of power-cutting equipment: based on the active power voltage sensitivity matrix, we select the distributed generation source with the highest active power voltage sensitivity value  , which has active power regulation capability.The required amount of active power regulation  is calculated according to Formula (10).
(7) Check if there are still over-voltage nodes in the cluster: after trimming the active power of distributed generation sources according to  , we check if there are still over-voltage nodes within the cluster.If over-voltage conditions persist, we repeat from steps (5) to step (7).Otherwise, the voltage-reactive power control adjustment for this cluster is completed.

Control architecture for integrated medium and low voltage distribution network based on cloudedge collaboration
With the large-scale integration of distributed photovoltaics, it is challenging to locally absorb its generated power when it is solely managed unit-wise at distribution substations.This difficulty leads to power back-feed to medium voltage and even higher voltage levels.To address this issue, we refer to the architecture system of the Internet of Things and propose a cloud-edge collaboration, integrating coordinated control between medium and low-voltage distribution networks (as shown in Figure 1).This layered and zonal coordination control structure enables autonomous operation at different tiers and zones within a medium and low-voltage distribution network.It not only guarantees the safe operation of the distribution network but also maximizes the absorption of power from distributed sources.
(1) The term "cloud" refers to the power distribution main station or cloud-based main station, which is responsible for the global optimization of medium and low-voltage power distribution networks.It obtains the power generation prediction data and load prediction data of the power distribution network, considers various constraints such as the output of distributed power and the structure of the power distribution network, and formulates a day-ahead dispatch plan for the medium and low-voltage power distribution network.Based on real-time collected information, short-term load prediction results, and power prediction results, it adjusts the intraday dispatch strategy on a rolling basis, monitors the operation status of the power distribution network in real time, and promptly handles malfunctions and anomalies in the power distribution network.
(2) "Edge" refers to the intelligent fusion terminal.The edge, with the power distribution station area as the autonomous unit, manages and controls the distributed photovoltaic power in its low-voltage distribution network to maximize the absorption of distributed power.Based on the daily dispatch plan and intraday dispatch strategy for the power station area and grid interconnection line issued by the "Cloud", as well as load forecasting results and photovoltaic power generation forecasting results, the edge optimizes the distributed photovoltaic output and energy storage charging/discharging power in the power distribution station area to achieve self-regulation and absorption of distributed power.It monitors the voltage quality of each node in the low voltage distribution network of the power distribution station area, adjusts reactive power, three-phase unbalance device, etc., to ensure that the voltage quality of this power distribution station area meets requirements and operates safely and economically in real-time.
(3) "End" refers to the distribution equipment, data acquisition devices, and various sensors: installed on the side of the distribution network (as shown in Figure 2), it collects the operation status of lowvoltage distribution network equipment and distributed power supplies, communicates with the smart integrated terminals, sends operational information and parameters of the distribution network and distributed power upstream, and executes control commands and strategies issued by the smart integrated terminals downstream.

Software architecture and primary functions
The coordinated control software for the integrated medium and low voltage distribution network is divided into two parts, "cloud" and "edge" (as shown in Figure 3).
The "cloud" side software is developed based on the existing distribution main station/distribution cloud main station to achieve rapid business iteration and decoupling from the original software and hardware.It adopts a micro-service architecture, mainly including basic algorithm libraries composed of particle swarm optimization algorithm service, neural network algorithm service, LSTM algorithm service, SVM algorithm service, etc.; grid analysis algorithm library composed of topology analysis service, state estimation service, power flow calculation service, etc.; and business application layer composed of load forecasting service, photovoltaic power generation forecasting service, cluster division service, network reconstruction service, source-grid-load-storage coordinated control service, heavy overload monitoring and early warning service, power quality analysis and early warning service, group adjustment and control optimization strategy service, acceptable capacity evaluation service, etc.
The "edge" side software is deployed on smart integrated terminals in the form of an app using virtual container technology, mainly including APP such as coordinated control, distributed photovoltaic power generation forecasting, load forecasting, distributed photovoltaic integration, heavy overload governance of distribution transformers in substation areas, and power quality governance.

On-site use
Due to a large number of distributed photovoltaic (PV) power generation systems being connected, causing reverse power flow and overloading in the medium-voltage distribution network, the 10 kV northern main line of the power supply bureau is being retrofitted with PV grid interconnection boundary switches, protocol converters, and other equipment, along with the renovation of 27 distribution substations and 271 inverters.Additionally, a distributed PV optimization and cluster control software, as well as intelligent integrated terminals for distributed PV cluster control and voltage-reactive power control applications, are being deployed at the distribution cloud master station to achieve "partition autonomy and hierarchical coordination" for optimized operation and control of the medium-and lowvoltage integrated distribution network.
Example 1 is the effect of overload regulation on the circuit.On October 10, 2023, at 10:30 AM, when the maximum current reached 595.4 A, exceeding the rated capacity of the line which is 503 A, resulting in a maximum load rate of 118.3%, a reverse overloading occurred.The distribution cloud master station initiated optimization calculations and sent active power strategies to the intelligent integrated terminals.Upon receiving the power strategy from the distribution cloud master station, each substation's intelligent integrated terminal activated the distributed PV optimization and cluster control software, adjusting the active power output of each PV inverter based on the power requirements communicated by the distribution cloud master station.As a result, after issuing the control commands, the reverse line current decreased by 200 A and the reverse load rate dropped to below 80% (as shown in Figure 4).Example 2 is the effect of overvoltage control in the distribution substation area.On October 13, the control strategy was not activated all day, and the photovoltaic reverse transmission caused the peak voltage at noon to exceed 250 V (taking endpoint 6 as an example).At 0:00 on October 14, the control strategy was activated, the voltage levels of each node were monitored in real-time and the output of the photovoltaic was adjusted, so the area's voltage level remained stable throughout the 14th.At 14:00 on October 15, the control strategy was stopped, and the area voltage quickly rose again, exceeding the limit value (as shown in Figure 5).

Conclusions
To address issues caused by the large-scale connection of distributed photovoltaic power generation to the low-voltage distribution network, such as voltage over-limit, reverse overload of distribution transformers, and power backflow to the medium-voltage distribution network, an analysis of the practical engineering application requirements is conducted.A coordinated control architecture based on edge-cloud synergy and integrated control of the medium and low-voltage distribution networks is proposed.A zoning and hierarchical coordination control strategy is formulated on the cloud side to optimize daily rolling strategies for maximum absorption and minimal operating costs.On the edge side, distributed photovoltaic power generation in the local low-voltage distribution network is controlled based on the daily rolling strategies issued by the cloud to achieve optimized adjustment.Real-time monitoring is used to promptly address anomalies and ensure the safe operation of the power grid.This solution effectively addresses issues of overload, power backflow, and voltage over-limit in the medium and low-voltage distribution networks, and will be further refined and improved based on on-site applications.

Figure 1 .
Figure 1.Coordination and control architecture for medium and low voltage base on cloud-edge.

Figure 4 .
Figure 4.The process of line power reverse regulation.Example 2 is the effect of overvoltage control in the distribution substation area.On October 13, the control strategy was not activated all day, and the photovoltaic reverse transmission caused the peak voltage at noon to exceed 250 V (taking endpoint 6 as an example).At 0:00 on October 14, the control strategy was activated, the voltage levels of each node were monitored in real-time and the output of the photovoltaic was adjusted, so the area's voltage level remained stable throughout the 14th.At 14:00 on October 15, the control strategy was stopped, and the area voltage quickly rose again, exceeding the limit value (as shown in Figure5).