Multi-element intelligent control method for multi-target safety scheduling of power grid

Abstract

The invention discloses a multi-element intelligent control method for power grid multi-target safety scheduling. The method comprises the following steps: (1) modeling an active power distribution network based on simulation software OpenDSS; (2) converting a control problem of the active power distribution network into a Markov decision problem, constructing a deep reinforcement learning agent, and defining a state space, an action space and a reward function; (3) defining the weather index of each weather type according to the actual output capability of photovoltaic power generation; combining different weather types together by adopting a random scene generation method to form a plurality of random training scenes; (4) training the intelligent agent by adopting a near-end strategy optimization algorithm PPO; and (5) applying the trained reinforcement learning agent to intelligent control of the active power distribution network. According to the method, the novel active power distribution network with the characteristics of complexity, uncertainty and the like is comprehensively controlled, and the comprehensive goals that the voltage is not out of limit, the network loss is reduced and the control cost is reduced can be achieved.

Description

Technical Field

[0001] The present invention belongs to the technical field of multi-objective dispatch optimization of electric power systems, and in particular relates to a multi-intelligence control method for multi-objective safe dispatch of power grids. Background Art

[0002] As the proportion of distributed photovoltaic and wind power generation increases year by year, the intermittent and uncertain nature of distributed photovoltaic and wind power generation has brought huge challenges to grid dispatch and safe and economical operation. In particular, the integration of distributed photovoltaic and wind power in distribution networks has increased the complexity of grid regulation and requires innovative control strategies.

[0003] As a key solution to the above challenges, active distribution networks have effectively improved the flexibility and operating efficiency of power grids by optimizing and coordinating the capabilities of distributed energy resources and flexible loads, and promoted the development of smart grids. Active distribution networks integrate multiple distributed energy sources, including distributed generation, energy storage systems and flexible loads, control bidirectional currents, and maximize the use of renewable energy, which is an important direction for promoting the transformation of future power systems.

[0004] Compared with traditional distribution networks, active distribution networks have more complex structures and operating modes, with a higher proportion of power electronic equipment and new energy, and more diverse constraints and control equipment types. This requires the formulation of more flexible and targeted control strategies to adapt to changing operational needs. For example, the ideal dispatch concept proposed by PJM in the United States in 2007 emphasizes the optimal start and stop and economic dispatch of units under safety constraints, including "re-dispatch" and "reflection" operations 24 hours before the day.

[0005] In recent years, technologies such as virtual synchronous generators, droop control and PQ control have provided relatively mature local control solutions for active distribution networks, but these solutions lack the ability to coordinate and optimize the overall system.

[0006] Therefore, it is necessary to combine some optimization algorithms, such as linear and nonlinear programming, dynamic programming and multi-objective particle swarm optimization, to achieve global optimization. However, these algorithms often have modeling difficulties and are prone to local optimal limitations when facing the complexity of modern active distribution networks.

[0007] With the rapid development of artificial intelligence, big data analysis and Internet of Things technology, new methods have been provided for the optimization control of active distribution networks. Deep reinforcement learning has become a powerful tool to solve these challenges because it can handle complex problems such as high-dimensionality, large-scale, and uncertainty. Deep reinforcement learning combines the perception ability of deep learning with the decision-making ability of reinforcement learning. It is particularly suitable for dealing with the complexity of modern power systems, especially the multi-dimensional control problems in active distribution networks.

[0008] In power system applications, deep reinforcement learning has been successfully applied to emergency control, automatic generation control, integrated energy system coordination, and demand side management.

[0009] However, the application of deep reinforcement learning in the comprehensive control of new active distribution networks is still in its early stages, and facing challenges such as complexity and uncertainty, there is still huge research potential. Summary of the invention

[0010] The present invention provides a multi-intelligence control method for multi-objective safe dispatching of power grids, which comprehensively controls a new type of active distribution network with characteristics such as complexity and uncertainty, and can achieve the comprehensive goals of voltage not exceeding the limit, network loss reduction, and control cost reduction.

[0011] A multi-intelligent control method for multi-objective safe dispatching of power grid includes the following steps:

[0012] (1) Modeling the active distribution network based on the distribution network simulation software OpenDSS;

[0013] (2) Convert the control problem of the active distribution network into a Markov decision problem, construct a deep reinforcement learning agent, and define the state space, action space, and reward function required for agent training;

[0014] (3) Based on the actual output capacity of photovoltaic power generation, the weather index of each weather type is defined; the random scenario generation method is used to combine different weather types to form multiple random training scenarios; these training scenarios simulate the changes in power demand and power generation of the active distribution network under different weather conditions, thereby providing training data for the reinforcement learning agent;

[0015] (4) The proximal policy optimization algorithm PPO is used to train the intelligent agent. First, the photovoltaic power generation data and load data are preprocessed, and then the parameters of the actor network and the critic network are initialized. The actor network is responsible for generating control actions based on the current state input; the critic network is used to evaluate the quality of the control actions generated by the actor network, that is, to estimate the future rewards brought by the action.

[0016] (5) The trained reinforcement learning agent is used for intelligent control of active distribution networks.

[0017] In step (1), the photovoltaic power generation system, energy storage system, load regulation equipment and power electronic equipment of the active distribution network are modeled in detail through OpenDSS, the behavior of various types of equipment under different operating actions is simulated, and key indicators are calculated in real time, including power flow, line loss, and voltage fluctuation.

[0018] In step (2), the state space includes: node voltage amplitude, node active power, node reactive power, on-load voltage regulator gear, parallel capacitor bank switching state, energy storage gear, photovoltaic inverter output reactive power.

[0019] <U m , P i , Q i 、Tapnum OlTC 、State capacior , Q pv 、State storage >

[0020] The action space includes: switching on and off of parallel capacitor banks, selection of on-load voltage regulator gears, charging and discharging of energy storage, and reactive output of photovoltaic inverters.

[0021] The reward function is:

[0022]

[0023] storage difference =batt kw / batt kw_allow

[0024] Among them, M1, M2, and M3 are the voltage over-limit penalty coefficient, network loss penalty coefficient, and control cost penalty coefficient respectively; P loss_ratio is the network loss rate; reg difference pv is the action amplitude of the on-load voltage regulator gear; difference is the control amplitude of photovoltaic reactive power; storage difference The depreciation degree of the energy storage battery is the ratio of the actual discharge capacity of the energy storage to the maximum allowable discharge capacity; cap difference The number of times the capacitor is switched on and off.

[0025] In step (3), the weather index of each weather type is defined as follows:

[0026] The weather index for a sunny day is defined as 5, indicating that the photovoltaic power generation system achieves the highest power generation efficiency on sunny days; the weather index for a cloudy day is defined as 4.1, the weather index for an overcast day is defined as 2.7, the weather index for a rainy day is defined as 1.8, and the weather index for a rainy or snowy day is defined as 1, indicating that the photovoltaic power generation is reduced to the lowest.

[0027] In step (3), the generation process of each training scene includes:

[0028] Select weather type: According to historical meteorological data, randomly select weather types within a certain period of time and assign corresponding weather indexes to each weather type;

[0029] Calculate photovoltaic power generation: According to the weather index corresponding to each weather type, calculate the power generation of the photovoltaic system under the weather conditions; the higher the weather index, the greater the photovoltaic power generation; conversely, the smaller the power generation.

[0030] In step (4), when the proximal policy optimization algorithm PPO is used to train the intelligent agent, the operation of the distribution network must meet the constraints of power flow balance, photovoltaic output limit, energy storage output limit and voltage not exceeding the limit;

[0031] Among them, power flow balance is used to ensure the balance between power input and power output at each node, ensuring that the power flow of the power grid complies with the laws of physics; photovoltaic output limitation is used to ensure that the output of the photovoltaic system is adjusted within a certain range and cannot exceed its maximum power output capacity; energy storage output limitation is used to ensure that the charging and discharging power of the energy storage system must be within the specified range; voltage limit is used to ensure that the voltage amplitude of the node must be maintained within the set safety range.

[0032] Compared with the prior art, the present invention has the following beneficial effects:

[0033] 1. The present invention adopts the distribution network simulation software OpenDSS to model the new active distribution network, transforms the control problem of the active distribution network into a Markov decision problem, constructs a deep reinforcement learning agent training model, and models typical control elements: photovoltaic, parallel capacitor bank, on-load voltage regulator, and energy storage, and transforms them into the action space of deep reinforcement learning.

[0034] 2. The present invention uses the PPO algorithm of deep reinforcement learning to train the intelligent agent, in which the deep neural network adopts a fully connected mode. The intelligent agent continuously interacts with the environment of the active distribution network and finds suitable training parameters in multiple trainings. The trained intelligent agent is tested with new data samples, and the test results also achieve the comprehensive goals of voltage not exceeding the limit, network loss reduction, and control cost reduction.

[0035] 3. The present invention proposes a new active distribution network intelligent control method that is adaptable to weather changes, robust and adaptable. By introducing weather type index to generate photovoltaic data, and using these data for training, good results have been achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 This is a flow chart of a multi-intelligence control method for multi-objective safety dispatch of power grid according to an embodiment of the present invention;

[0037] Figure 2 It is the photovoltaic curve diagram of the method of the present invention.

[0038] Figure 3 It is a schematic diagram of voltage situation before optimization;

[0039] Figure 4 It is a schematic diagram of voltage after optimization;

[0040] Figure 5 This is a schematic diagram comparing network loss before and after optimization;

[0041] Figure 6 It is a schematic diagram of control cost using OpenDSS control;

[0042] Figure 7 Schematic diagram of control cost using the control method of the present invention. DETAILED DESCRIPTION

[0043] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be pointed out that the embodiments described below are intended to facilitate the understanding of the present invention and do not have any limiting effect on the present invention.

[0044] The present invention defines the multi-objective optimization control problem of the power grid as a Markov decision problem, and proposes a multi-intelligence control method for multi-objective safe scheduling of the power grid to solve the problems of voltage over-limit, increased network loss, increased control cost, etc. of the power grid to which a high proportion of photovoltaic and power electronic equipment is connected.

[0045] For the new active distribution network, the present invention uses the distribution network simulation software OpenDSS for modeling, and uses the COM interface to automatically interact with Python. For the complex system control process, this paper models the Markov decision process, and uses the parallel capacitor bank, on-load voltage regulator, energy storage, and photovoltaic inverter as control elements on the premise of 100% photovoltaic absorption, and takes voltage limit, network loss reduction, and control cost reduction as reward functions. For the training process, the present invention uses real load data and photovoltaic power generation data as training sets, and introduces a weather type index to adapt to weather changes. Finally, according to the case analysis, the intelligent agent that has been trained many times verifies the effectiveness, adaptability, and robustness of the control strategy.

[0046] like Figure 1 As shown, it is a flow chart of a multi-intelligence control method for multi-objective safe dispatching of power grid according to an embodiment of the present invention. The embodiment of the present invention is realized by the following specific technical solutions:

[0047] The present invention first models the new active distribution network based on the distribution network simulation software OpenDSS:

[0048] New active power distribution networks usually include a variety of distributed energy, energy storage systems, power electronic devices, and flexible load adjustment equipment. Compared with traditional distribution networks, active distribution networks face more dynamic changes and uncertainties, so a comprehensive modeling tool is needed to handle these complex system characteristics.

[0049] The present invention uses OpenDSS to unify the modeling of these complex control devices and operating conditions to form a complete distribution network simulation model. By modeling the operation of photovoltaic power generation, energy storage, load regulation equipment and power electronic equipment in detail in OpenDSS, it is possible to simulate the behavior of various types of equipment in the power grid under different operating conditions and calculate key indicators such as power flow, line loss, and voltage fluctuation in real time.

[0050] Compared with traditional static power system modeling, OpenDSS provides dynamic power flow calculation functions, which enables the intelligent agent to continuously adjust the control strategy under changing environmental conditions to cope with real-time load changes, photovoltaic output fluctuations, and uncertainties in energy storage charging and discharging. This refined modeling improves the authenticity and adaptability of the model, and greatly enhances the application effect of deep reinforcement learning algorithms in complex power systems.

[0051] This high-precision distribution network modeling method lays the foundation for the application of intelligent control methods, enabling deep reinforcement learning algorithms to be trained and optimized in a simulation environment to obtain effective control strategies.

[0052] The present invention proposes a multivariate control method based on the PPO algorithm, comprising the following steps:

[0053] First, the constraints of this method are set. In the operation of the distribution network, various constraints such as power flow balance, photovoltaic output limit, energy storage output limit and voltage limit must be met, as follows:

[0054] Power flow balance constraint: This constraint ensures the balance between power input and power output at each node, ensuring that the power flow of the power grid conforms to the laws of physics.

[0055]

[0056]

[0057] Photovoltaic power generation is affected by weather and light conditions, so the output of the photovoltaic system needs to be adjusted within a certain range and cannot exceed its maximum power output capacity.

[0058] P pv(i) ≤P pvmax

[0059]

[0060] The charging and discharging power of the energy storage system must be within the specified range to ensure the life of the energy storage equipment and the stable operation of the system.

[0061] P storage(i) ≤P storagemax

[0062] To ensure the safe operation of the power grid, the voltage amplitude of all nodes must be kept within the set safety range to avoid excessively high or low voltage affecting equipment operation.

[0063] U min ≤U i ≤U max

[0064] Where P pv(i) represents the active power injected by PV at node i, P storage(i) represents the charging power of energy storage at node i, P load(i) represents the active power of the load at node i. Similarly, Q solar(i) represents the inductive reactive power absorbed by the photovoltaic at node i, Q storage(i) represents the inductive reactive power absorbed by the energy storage at node i, Q load(i) Represents the inductive reactive power generated by the load at node i. The intelligent control method of the present invention will realize the control effectiveness verification during the operation of the distribution network based on the above constraints.

[0065] Based on the principle of Markov decision process and combined with the relevant characteristics of the active distribution network environment, the parameters such as state space, action space, reward function and so on required for the training of the intelligent agent in the present invention are defined.

[0066] In order to comprehensively observe the active distribution network, the state space s of the active distribution network is set to include: node voltage amplitude, node active power, node reactive power, on-load voltage regulator gear, shunt capacitor bank switching state, energy storage gear, and photovoltaic inverter output reactive power.

[0067] <U m , P i , Q i 、Tapnum OlTC 、State capacior , Q pv 、State storage >

[0068] In order to make full use of the control resources of the active distribution network, the present invention also incorporates photovoltaics and energy storage into the control means. The action space of the active distribution network is mainly determined by the switching of parallel capacitor groups, the selection of on-load voltage regulator gears, the charging and discharging of energy storage, and the reactive output of photovoltaic inverters.

[0069] The control goal of this intelligent control method is to achieve voltage within the limit, reduce network losses, and control costs in an active distribution network environment with a high proportion of photovoltaic power generation to ensure that the active distribution network develops safely and efficiently towards the goal of cleanliness and environmental protection. Therefore, the reward function of the present invention is:

[0070]

[0071] storage difference =batt kw / batt kw_allow

[0072] Among them, M1, M2, and M3 are the voltage over-limit penalty coefficient, network loss penalty coefficient, and control cost penalty coefficient respectively; P loss_ratio is the network loss rate; reg difference pv is the action amplitude of the on-load voltage regulator gear; difference is the control amplitude of photovoltaic reactive power; storage difference The depreciation degree of the energy storage battery is the ratio of the actual discharge capacity of the energy storage to the maximum allowable discharge capacity; cap difference The number of times the capacitor is switched on and off.

[0073] The present invention proposes a method for generating random training scenarios according to weather types. The method comprises the following steps:

[0074] The present invention first defines the change ratio of photovoltaic power generation under different weather conditions as a "weather index". In order to simulate the power generation capacity under different weather conditions, the present invention sets the weather index of rainy and snowy days to 1, indicating that the output of photovoltaic power generation is the weakest at this time. For other weather types, the weather index of each weather type is defined according to the actual output capacity of photovoltaic power generation. The definitions of each weather type and the corresponding weather index are as follows:

[0075] Sunny day: The weather index is 5, which means that the photovoltaic power generation system can achieve the highest power generation efficiency on sunny days.

[0076] Cloudy: The weather index is 4.1, which means that the power generation of the photovoltaic power generation system in cloudy weather is slightly lower than that in sunny days, but still maintains a high output.

[0077] Cloudy day: The weather index is 2.7. On cloudy days, the sunlight is blocked and the photovoltaic power generation is significantly lower than that on sunny and cloudy days.

[0078] Rainy days: The weather index is 1.8. In rainy weather, due to the reduction in sunlight, photovoltaic power generation further decreases and the output capacity of the photovoltaic system is greatly weakened.

[0079] Rainy and snowy days: The weather index is 1, which means that photovoltaic power generation has almost no output and the power generation drops to the minimum.

[0080] According to the above weather types and their corresponding weather indexes, the present invention adopts a random scene generation method to combine different weather types together to form multiple random training scenes, such as Figure 2These training scenarios simulate the changes in power demand and power generation of the distribution network under different weather conditions, thereby providing more representative training data for the reinforcement learning agent. The generation process of each training scenario includes:

[0081] Select weather type: Based on historical meteorological data, randomly select weather types within a certain period of time (such as weather changes in a day), and assign a corresponding weather index to each weather type.

[0082] Calculate photovoltaic power generation: According to the weather index corresponding to each weather type, calculate the power generation of the photovoltaic system under the weather conditions. The higher the weather index, the greater the photovoltaic power generation; conversely, the smaller the power generation.

[0083] The present invention proposes a training process based on OpenDSS and Python joint simulation. The specific steps are as follows:

[0084] Before training begins, the photovoltaic power generation data and load data must first be preprocessed. Since the amplitude range of these data may vary greatly, in order to improve the training efficiency and stability of the deep learning model, all input data (photovoltaic output power, load power, etc.) need to be normalized. Specifically, the photovoltaic power generation and load data are processed by maximum and minimum normalization to make their value ranges consistent, avoiding the impact of data scale differences on the training process. The normalized data ensures that the intelligent agent can better perceive changes in the environment and reduce the problem of gradient disappearance or explosion during training.

[0085] Then initialize the parameters of the Actor network and the Critic network. After the data is processed, the reinforcement learning network is initialized. In order to enable the deep reinforcement learning agent to handle the complex dispatching tasks of the distribution network, two neural networks are initialized first: the Actor network and the Critic network.

[0086] The Actor network is responsible for generating control actions based on the current state input, that is, selecting and adjusting the operation mode of each control device in the power grid (photovoltaic inverter, energy storage, OLTC, capacitor, etc.). The input of the Actor network is the current state of the distribution network, including the voltage, load, photovoltaic power generation, energy storage status, etc. of each node, and the output is the control strategy (such as adjusting the OLTC gear, energy storage charging and discharging status, etc.).

[0087] The Critic network is used to evaluate the quality of the actions generated by the Actor network, that is, to estimate the future rewards brought by the action. The task of the Critic network is to provide feedback signals to the Actor network to guide it to improve its strategy. The input of the Critic network is the current state and the action taken by the agent, and the output is the value estimate of the action.

[0088] After initializing the Actor and Critic networks, the next step is to interact with the distribution network simulation environment. The present invention uses OpenDSS as a simulation platform to simulate the operating status of the distribution network. OpenDSS can simulate the operating conditions of various types of equipment in the distribution network in real time, such as voltage level, power flow, equipment control status, etc. By interacting with the OpenDSS environment, the agent can make control decisions based on the current state and receive feedback from the environment.

[0089] After each interaction between the agent and the environment, the generated experience (including the current state, the action taken, the immediate reward, and the next state) is stored in the experience pool. The experience pool collects all interaction records of the agent during the training process and provides samples for subsequent network training.

[0090] After adopting the intelligent control strategy proposed by the present invention, the over-limit situation of each bus voltage is controlled within the range of 5% above and below; Figure 3 and Figure 4 The voltage conditions before and after optimization are shown in Figure 1. It can be seen intuitively that the voltage limit situation has been greatly improved. The network loss has been greatly reduced, and the average network loss has been reduced by 17.14%. Figure 5 As shown; the control costs of using OpenDSS and this control strategy are as follows Figure 6 and Figure 7 As shown in the figure, the control cost is greatly reduced. Compared with the control cost of the control function of the distribution network simulation software OpenDSS itself, the total control amplitude of the on-load voltage regulator is reduced by 1710 times, and the total switching times of the capacitor is reduced by 183 times.

[0091] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A multi-intelligence control method for multi-objective safe dispatching of power grid, characterized in that: The following steps are involved: (1) Modeling the active distribution network based on the distribution network simulation software OpenDSS; (2) Convert the control problem of the active distribution network into a Markov decision problem, construct a deep reinforcement learning agent, and define the state space, action space, and reward function required for agent training; (3) Based on the actual output capacity of photovoltaic power generation, define the weather index of each weather type; use the random scene generation method to combine different weather types together to form multiple random training scenes; These training scenarios simulate the changes in power demand and generation of active distribution networks under different weather conditions, thus providing training data for reinforcement learning agents; (4) The proximal policy optimization algorithm PPO is used to train the intelligent agent. First, the photovoltaic power generation data and load data are preprocessed, and then the parameters of the actor network and the critic network are initialized. The actor network is responsible for generating control actions based on the current state input; the critic network is used to evaluate the quality of the control actions generated by the actor network, that is, to estimate the future rewards brought by the action. (5) The trained reinforcement learning agent is used for intelligent control of active distribution networks.

2. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 1 is characterized in that: In step (1), the photovoltaic power generation system, energy storage system, load regulation equipment and power electronic equipment of the active distribution network are modeled in detail through OpenDSS, the behavior of various types of equipment under different operating actions is simulated, and key indicators are calculated in real time, including power flow, line loss, and voltage fluctuation.

3. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 1 is characterized in that: In step (2), the state space includes: node voltage amplitude, node active power, node reactive power, on-load voltage regulator gear, parallel capacitor bank switching state, energy storage gear, photovoltaic inverter output reactive power.

4. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 3 is characterized in that: In step (2), the action space includes: switching on and off of the parallel capacitor bank, selection of the on-load voltage regulator gear, charging and discharging of energy storage, and reactive output of the photovoltaic inverter.

5. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 4 is characterized in that: In step (2), the reward function is: storage difference =batt kw / batt kw_allow Among them, M1, M2, and M3 are the voltage over-limit penalty coefficient, network loss penalty coefficient, and control cost penalty coefficient respectively; P loss_ratio is the network loss rate; reg difference pv is the action amplitude of the on-load voltage regulator gear; difference is the control amplitude of photovoltaic reactive power; storage difference The depreciation degree of the energy storage battery is the ratio of the actual discharge capacity of the energy storage to the maximum allowable discharge capacity; cap difference The number of times the capacitor is switched on and off.

6. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 1 is characterized in that: In step (3), the weather index of each weather type is defined as follows: The weather index for a sunny day is defined as 5, indicating that the photovoltaic power generation system achieves the highest power generation efficiency on sunny days; the weather index for a cloudy day is defined as 4.1, the weather index for an overcast day is defined as 2.7, the weather index for a rainy day is defined as 1.8, and the weather index for a rainy or snowy day is defined as 1, indicating that the photovoltaic power generation is reduced to the lowest.

7. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 1 is characterized in that: In step (3), the generation process of each training scene includes: Select weather type: According to historical meteorological data, randomly select weather types within a certain period of time and assign corresponding weather indexes to each weather type; Calculate photovoltaic power generation: According to the weather index corresponding to each weather type, calculate the power generation of the photovoltaic system under the weather conditions; the higher the weather index, the greater the photovoltaic power generation; conversely, the smaller the power generation.

8. The multi-intelligence control method for multi-objective safety dispatch of power grid according to claim 1 is characterized in that: In step (4), when the proximal policy optimization algorithm PPO is used to train the intelligent agent, the operation of the distribution network must meet the constraints of power flow balance, photovoltaic output limit, energy storage output limit and voltage not exceeding the limit; Among them, power flow balance is used to ensure the balance between power input and power output at each node, ensuring that the power flow of the power grid complies with the laws of physics; photovoltaic output limitation is used to ensure that the output of the photovoltaic system is adjusted within a certain range and cannot exceed its maximum power output capacity; energy storage output limitation is used to ensure that the charging and discharging power of the energy storage system must be within the specified range; voltage limit is used to ensure that the voltage amplitude of the node must be maintained within the set safety range.

Images