A distribution network side energy storage collaborative power distribution optimization control method and system

Abstract

The application discloses a kind of distribution network side energy storage collaborative power distribution optimization control method and system, it is related to power distribution network operation control technical field, the present application real-time acquisition each node operation data of distribution network, with beidou satellite time service as time base reference upload to edge computing node, construct time series prediction model and roll over predict node net load, with the sum of power distribution network loss and energy storage operating cost minimum as target to build optimization scheduling model, each energy storage charge-discharge power instruction is obtained by solving, energy storage executes, active instruction is satisfied first, when voltage out-of-limit, local reactive power compensation is called by remaining capacity, actual operation deviation is collected at the end of scheduling period and corrected next cycle model parameter.The present application improves the adaptability of distribution network to distributed power fluctuation, considers distribution network operation efficiency and energy storage service life, guarantees node voltage quality, adapts to the demand of distribution network regulation and control after high proportion distributed photovoltaic access.

Description

Technical Field

[0001] This invention relates to the field of power distribution network operation control technology, and in particular to a method and system for power distribution optimization control of energy storage-coordinated power distribution on the power distribution network side. Background Technology

[0002] With the continuous increase in the scale of distributed photovoltaic and other renewable energy connections on the distribution network side, the power flow distribution and operating characteristics of traditional passive distribution networks have undergone significant changes. The randomness and volatility on both the source and load sides have greatly increased the probability of problems such as voltage deviation at distribution network nodes, feeder current exceeding limits, and increased network losses. The control methods relied upon by traditional distribution networks, such as on-load tap changer adjustment and parallel capacitor bank switching, have coarse adjustment granularity, strong response lag, and can only achieve stepped adjustment. They are difficult to adapt to the dynamic operating requirements of the distribution network after the high proportion of distributed power sources are connected, and cannot respond to short-term source and load fluctuations in a timely manner, which easily leads to problems such as operating indicators exceeding limits due to untimely adjustment.

[0003] The large-scale application of distribution network-side energy storage systems provides flexible adjustment resources for distribution network operation and control. However, existing collaborative control methods for distribution network-side energy storage still have many shortcomings: most optimization scheduling methods only take the operation indicators of the distribution network as optimization targets, without fully considering the operation losses and aging characteristics of the energy storage system itself. Some methods that include energy storage costs only calculate the aging losses corresponding to the charging and discharging capacity, without considering the additional aging effects caused by drastic fluctuations in charging and discharging power. This can easily lead to unreasonable operating states such as frequent high-power charging and discharging and deep charging and discharging of the energy storage system, shortening the actual service life of the energy storage. At the same time, most existing scheduling methods adopt a single time scale optimization mode, with a coarse time granularity for global optimization, making it difficult to cope with random fluctuations in source and load within a short time scale. Some local control strategies rely solely on locally collected operation data for adjustment, without coordinating with the global optimization scheduling plan, which can easily lead to conflicts between local adjustments and global operation objectives.

[0004] Furthermore, in the existing distribution network operation data acquisition system, the time synchronization accuracy of different types of acquisition terminals is not uniform, and the timestamps of operation data from different nodes are biased, resulting in insufficient consistency of the input data time series of the time series prediction model, which in turn affects the accuracy of the prediction results. Most optimized scheduling methods do not set up corresponding deviation feedback correction links, so the deviation between the predicted value and the measured value cannot be corrected in a timely manner during actual operation. Over the long term, the adaptability of the scheduling plan will gradually decrease, making it difficult to adapt to the dynamic changes in the operating characteristics of the distribution network. At the same time, in the existing control methods, the active power scheduling and reactive power compensation of energy storage are mostly designed independently, and the remaining reactive power capacity of the energy storage converter is not fully utilized, resulting in the underutilization of the regulation potential of energy storage equipment. Summary of the Invention

[0005] This invention proposes a distribution network-side energy storage-coordinated power distribution optimization control method and system to solve the problems mentioned in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: a distribution network-side energy storage coordinated power distribution optimization control method, comprising the following steps: S1. Real-time collection of electrical operation data of each node in the power distribution network. Each collection terminal uses the second pulse of Beidou satellite time synchronization as the time reference, and after being marked with a unified timestamp, it is uploaded to the edge computing node through the power distribution communication network. S2. Use edge computing nodes to construct a time-series forecasting model based on historical load data and photovoltaic output data. Use the time-series forecasting model to make rolling forecasts of the load demand and photovoltaic output of each node in the future preset time domain to obtain the net load forecast value. S3. An optimal scheduling model is constructed with the goal of minimizing the sum of distribution network losses and energy storage operating costs. Constraints are set for distribution network power flow equations, node voltage amplitude, feeder current carrying capacity, upper and lower limits of energy storage charge state, and charging and discharging power limits. Under the condition of satisfying all constraints, the preset scheduling period is used as the optimization time domain. The charging and discharging power plans of each energy storage system are solved by rolling optimization to obtain the charging and discharging power command values ​​for each time period. S4. The charging and discharging power command value is sent to the local controller of each energy storage system. The local controller monitors the voltage amplitude of the access node in real time when executing the power command. When the voltage amplitude exceeds the preset deviation threshold, the remaining reactive capacity of the energy storage converter is used to perform local reactive compensation on the node voltage under the premise of prioritizing the active power command. S5. At the end of each scheduling cycle, obtain the actual charging and discharging power curves and state of charge change curves of each energy storage system, calculate the deviation between the actual operation and the optimized scheduling prediction, use the deviation as a feedback correction term to update the objective function of the next cycle, and adaptively correct the parameters of the optimized scheduling model online.

[0007] Furthermore, the objective function of the optimization scheduling model described in S3 consists of two parts: a network loss cost term and an energy storage operation cost term. The energy storage operation cost term is structured as follows: ; In the formula, This represents the total operating cost of energy storage. This represents the total number of energy storage systems on the distribution network side. The total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period. The aging cost coefficient per unit charge / discharge capacity. For single time period duration, This is the power fluctuation penalty coefficient. For the first The rated capacity of each energy storage system.

[0008] Furthermore, in S3, the rolling optimization solution is set as a two-layer time scale. The upper layer uses a first preset duration as the scheduling period and completes the global optimization of the charging and discharging power of each energy storage system based on the first granularity discrete time. The lower layer, based on the optimization results of the upper layer, jointly optimizes the active power and reactive power of the energy storage converter with a second preset duration and a second granularity discrete time. The lower layer optimization satisfies the node voltage constraint and the feeder current carrying capacity constraint, tracks the active power plan value of the upper layer, and performs local adjustment of reactive power. The first preset duration is longer than the second preset duration, and the first granularity is coarser than the second granularity.

[0009] Furthermore, in S1, the electrical operation data of each node of the distribution network are jointly collected by the intelligent fusion terminal, the feeder terminal unit and the photovoltaic grid-connected intelligent circuit breaker. Each collection terminal and the edge computing node exchange data using a communication protocol based on the IEC61850 standard. The edge computing node is deployed in the power distribution room and has a built-in local data storage cache and optimized computing processing unit.

[0010] Furthermore, the time-series prediction model described in S2 adopts a gated recurrent unit deep neural network structure. The model uses historical load data and historical photovoltaic output data of the same period over several consecutive days as input features, and uses the load value and photovoltaic output value of each node in the future preset time domain as prediction output. Before each rolling prediction, the model network weights are incrementally trained using measured data from the most recent scheduling cycles.

[0011] Furthermore, the remaining reactive capacity of the energy storage converter mentioned in S4 is obtained by taking the square root of the difference between the square of the apparent power rating of the energy storage converter at the current moment and the square of the current active power command value. When the remaining reactive capacity is insufficient to adjust the node voltage amplitude back to the preset deviation threshold range, the local controller sends a voltage over-limit alarm message to the upper-level edge computing node and requests the energy storage system of the adjacent node to provide reactive power assistance.

[0012] Furthermore, after receiving the electrical operation data sent by each acquisition terminal, the edge computing node in S1 performs integrity verification and outlier removal on the data. When it detects that a node's data is missing or the value of a data item exceeds the range of change, it uses linear interpolation of valid data at adjacent time points to complete the missing or abnormal data. The data after verification and completion is used as the input data for the time series prediction model in S2 for prediction.

[0013] Furthermore, a distribution network-side energy storage-coordinated power distribution optimization control system includes the following modules: The data acquisition and sensing module is used to collect electrical operation data of each node of the power distribution network in real time. The data acquisition and sensing module uses the Beidou satellite time synchronization second pulse as the time reference and uploads the data to the edge computing module through the power distribution communication network. The edge computing module is used to receive electrical operation data sent by the acquisition and sensing module. It has a built-in time series prediction unit and an optimization scheduling solution unit. The time series prediction unit performs rolling prediction of the load demand and photovoltaic output of each node in the future preset time domain and calculates the net load prediction value. The optimization scheduling solution unit takes minimizing the sum of distribution network loss and energy storage operation cost as the optimization objective and solves the charging and discharging power command value of each energy storage system in each time period under the condition of simultaneously satisfying power flow constraints, voltage constraints, feeder current carrying capacity constraints and energy storage operation constraints, and sends it to the local control module. The local control module is used to receive the charging and discharging power command value issued by the edge computing module, and control the energy storage converter to execute the active power command. At the same time, it monitors the voltage amplitude of the access node in real time and uses the remaining reactive capacity of the energy storage converter to perform local reactive power compensation adjustment when the voltage amplitude exceeds the preset deviation threshold. The feedback correction module is used to acquire the actual charge and discharge power curves and state of charge change curves of each energy storage system at the end of each scheduling cycle, calculate the deviation between the actual operating data and the predicted data of the optimization scheduling model, and input the deviation as a feedback correction term into the optimization scheduling solution unit in the edge computing module to implement online adaptive correction of the model parameters.

[0014] Furthermore, the objective function of the optimization scheduling solution unit in the edge computing module consists of two parts: a network loss cost term and an energy storage operation cost term. The energy storage operation cost term is structured as follows: ; In the formula, This represents the total operating cost of energy storage. This represents the total number of energy storage systems on the distribution network side. The total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period. The aging cost coefficient per unit charge / discharge capacity. For single time period duration, This is the power fluctuation penalty coefficient. For the first The rated capacity of each energy storage system.

[0015] Furthermore, the reactive power capacity of the energy storage converter in the local control module is obtained by taking the square root of the difference between the square of the apparent power rating at the current moment and the square of the current active power command value. When the local control module detects that the reactive power capacity is insufficient to adjust the node voltage amplitude to the preset deviation threshold range, it sends a voltage over-limit alarm message to the edge computing module and requests the distribution network side energy storage system of the adjacent node to provide reactive power assistance.

[0016] Compared with existing technologies, the beneficial effects of this invention are: This invention provides a unified time reference for all acquisition terminals through the second pulse time synchronization of the BeiDou satellite, ensuring the consistency of the timing of operational data collected by different nodes. This provides reliable basic data support for subsequent load and photovoltaic output prediction and optimized scheduling solutions, avoiding analysis errors caused by timestamp deviations. The time series prediction model adopted supports incremental training, enabling iterative updates of model parameters based on the latest measured operational data. It adaptively tracks the changing patterns of load and photovoltaic output, improving the adaptability of prediction results to actual operating conditions and reducing the deviation of net load prediction.

[0017] The optimized scheduling model incorporates distribution network losses and energy storage operating costs into the optimization objectives. During the optimization process, it takes into account both the operational needs of the distribution network and the operational characteristics of the energy storage system, avoiding the adverse effects of unreasonable charging and discharging strategies on the lifespan of energy storage. A two-layer time-scale rolling optimization mode is adopted. The upper-layer coarse-grained global optimization can coordinate the charging and discharging plans of various energy storage systems at the overall distribution network level, ensuring the rationality of overall regulation. The lower-layer fine-grained optimization can adapt to short-term source-load fluctuations based on the upper-layer plan, and simultaneously complete the joint regulation of active and reactive power, improving the timeliness of dispatch response.

[0018] The local control mechanism prioritizes active power dispatch commands and utilizes the remaining reactive power capacity of energy storage converters for local voltage regulation. This fully leverages the regulation potential of existing energy storage equipment, enabling rapid voltage regulation at nodes without the need for additional reactive power compensation devices. When local reactive power capacity is insufficient, it can request energy storage systems from adjacent nodes to collaboratively participate in voltage regulation, improving the reliability of voltage regulation. The included feedback correction mechanism incorporates the deviation between actual and predicted data from the previous dispatch cycle into the optimization model for the next cycle, correcting parameters and preventing the continuous accumulation of prediction deviations that could cause the dispatch plan to deviate from actual operational needs, thus ensuring the long-term adaptability of the dispatch model. The entire control process achieves synergy between global optimization and local response, balancing the operational needs of the distribution network with the operating characteristics of the energy storage system, and is adaptable to the distribution network control requirements following the integration of a high proportion of distributed power sources. Attached Figure Description

[0019] Figure 1This is an overall flowchart of a distribution network-side energy storage coordinated power distribution optimization control method proposed in this invention; Figure 2 This is a flowchart of the data preprocessing and net load forecasting process of the present invention; Figure 3 This is a flowchart of the two-layer time-scale rolling optimization scheduling process of the present invention; Figure 4 This is a flowchart of the local control and reactive power voltage collaborative compensation process of the present invention. Figure 5 This is a flowchart of the online adaptive feedback correction process for parameters in this invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0022] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, features defined with "first" and "second" may explicitly or implicitly include one or more of the stated features. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified. Furthermore, the terms "installed," "connected," and "linked" should be interpreted broadly; for example, they may refer to a fixed connection, a detachable connection, or an integral connection; they may refer to a mechanical connection or an electrical connection; they may refer to a direct connection or an indirect connection through an intermediate medium; and they may refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances. The invention will now be described in further detail with reference to the accompanying drawings.

[0023] Reference Figures 1 to 5A method for coordinated distribution control of energy storage on the distribution network side, comprising the following steps: S1. Real-time collection of electrical operation data of each node in the distribution network, including at least the effective values ​​of three-phase voltage and three-phase current at the head end of each feeder, the active and reactive power on the low-voltage side of each distribution transformer, the active output of each distributed photovoltaic access point, and the state of charge, charging and discharging power and terminal voltage of the energy storage system on the distribution network side. Each acquisition terminal uses the Beidou satellite time synchronization second pulse as the time reference and is marked with a unified timestamp before uploading to the edge computing node through the power distribution communication network. S2. Utilize edge computing nodes to construct a time-series forecasting model based on historical load data and photovoltaic output data. Through the time-series forecasting model, perform rolling forecasts on the load demand and photovoltaic output of each node in the future preset time domain to obtain the net load forecast value. The net load forecast value is the difference between the load demand forecast value and the photovoltaic output forecast value. S3. An optimal scheduling model is constructed with the goal of minimizing the sum of distribution network losses and energy storage operating costs. Constraints are set for distribution network power flow equations, node voltage amplitude, feeder current carrying capacity, upper and lower limits of energy storage charge state, and charging and discharging power limits. Under the condition of satisfying all constraints, the charging and discharging power plan of each energy storage system is rolled and optimized in the preset scheduling period as the optimization time domain to obtain the charging and discharging power command value for each time period. S4. The charging and discharging power command value is sent to the local controller of each energy storage system. The local controller monitors the voltage amplitude of the access node in real time when executing the power command. When the voltage amplitude exceeds the preset deviation threshold, the remaining reactive capacity of the energy storage converter is used to perform local reactive compensation on the node voltage under the premise of prioritizing the active power command. S5. At the end of each scheduling cycle, obtain the actual charging and discharging power curves and state of charge change curves of each energy storage system, calculate the deviation between the actual operation and the optimized scheduling prediction, and incorporate the deviation as a feedback correction term into the objective function of the next cycle to implement online adaptive correction of the optimized scheduling model parameters.

[0024] This invention also includes: the objective function of the optimized scheduling model described in S3 consists of two parts: a network loss cost term and an energy storage operation cost term. The energy storage operation cost term is structured as follows: ; In the formula, This represents the total operating cost of energy storage, expressed in monetary units. This represents the total number of energy storage systems on the distribution network side. This represents the total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period, with charging as positive and power as the dimension; The aging cost coefficient per unit charge / discharge capacity is expressed in monetary units per unit of energy. The duration of a single time period is measured in units of time. This is the power fluctuation penalty coefficient, with dimensions of monetary units multiplied by the square of time; For the first The rated capacity of an energy storage system is expressed in units of energy. By integrating the cumulative aging loss of charging and discharging power and the additional aging loss due to power fluctuations into a unified objective function, the rolling optimization solution can simultaneously take into account the lifespan of energy storage and the economic efficiency of power distribution network operation.

[0025] In this invention, the method further includes: in S3, the rolling optimization solution is set as a two-layer time scale. The upper layer uses a first preset duration as the scheduling period to perform global optimization of the charging and discharging power of each energy storage system at the first granularity discrete time. The lower layer, based on the optimization results of the upper layer, performs joint optimization of the active power and reactive power of the energy storage converter at the second preset duration at the second granularity discrete time. The lower layer optimization tracks the planned value of the active power of the upper layer and performs local fine adjustment of the reactive power under the premise of satisfying the node voltage constraint and the feeder current carrying capacity constraint. The first preset duration is longer than the second preset duration and the first granularity is coarser than the second granularity.

[0026] In this invention, the electrical operation data of each node of the distribution network mentioned in S1 are jointly collected by the intelligent fusion terminal deployed on the low-voltage side of the distribution transformer, the feeder terminal unit deployed at the feeder head, and the photovoltaic grid-connected intelligent circuit breaker deployed at the photovoltaic grid-connected point. Each collection terminal and the edge computing node exchange data using a communication protocol based on the IEC61850 standard. The edge computing node is deployed in the power distribution room and has a built-in local data storage cache and optimized computing processing unit.

[0027] In this invention, the time-series prediction model described in S2 adopts a deep neural network structure based on gated recurrent units. The model uses historical load data and historical photovoltaic output data of the same period over several consecutive days as input features and uses the load value and photovoltaic output value of each node in the future preset time domain as prediction output. Before each rolling prediction, the network weights of the model are incrementally trained using the measured data of the most recent scheduling cycles to achieve adaptive tracking of load change patterns and photovoltaic output change patterns.

[0028] In this invention, the remaining reactive power capacity of the energy storage converter in S4 is obtained by taking the square root of the difference between the square of the apparent power rating of the energy storage converter at the current moment and the square of the current active power command value. When the remaining reactive power capacity is insufficient to adjust the node voltage amplitude back to the preset deviation threshold range, the local controller sends a voltage over-limit alarm message to the upper-level edge computing node and requests the energy storage system of the adjacent node to provide reactive power assistance.

[0029] In this invention, after receiving the electrical operation data sent by each acquisition terminal, the edge computing node in S1 performs integrity verification and outlier removal on the data. When it detects that a node's data is missing or the value of a data item exceeds the preset reasonable variation range for that item, it uses linear interpolation of valid data at adjacent time points to complete the missing or abnormal data. The data after verification and completion processing is used as the input data for the time series prediction model in S2 for prediction.

[0030] This invention includes the following modules: The data acquisition and sensing module is used to collect electrical operation data of each node of the distribution network in real time, including at least the effective values ​​of three-phase voltage and three-phase current at the head end of each feeder, the active power and reactive power on the low-voltage side of each distribution transformer, the active power output of each distributed photovoltaic access point, and the state of charge, charging and discharging power and terminal voltage of the energy storage system on the distribution network side. The data acquisition and sensing module uses the Beidou satellite time synchronization second pulse as the time reference and uploads the data to the edge computing module through the power distribution communication network. The edge computing module is used to receive electrical operation data sent by the acquisition and sensing module. It has a built-in time series prediction unit and an optimization scheduling solution unit. The time series prediction unit performs rolling prediction of the load demand and photovoltaic output of each node in the future preset time domain and calculates the net load prediction value. The optimization scheduling solution unit takes minimizing the sum of distribution network loss and energy storage operation cost as the optimization objective and solves the charging and discharging power command value of each energy storage system in each time period under the condition of simultaneously satisfying power flow constraints, voltage constraints, feeder current carrying capacity constraints and energy storage operation constraints, and sends it to the local control module. The local control module is used to receive the charging and discharging power command value issued by the edge computing module and control the energy storage converter to execute the active power command. At the same time, it monitors the voltage amplitude of the access node in real time and uses the remaining reactive capacity of the energy storage converter to perform local reactive power compensation adjustment when the voltage amplitude exceeds the preset deviation threshold. The feedback correction module is used to acquire the actual charge and discharge power curves and state of charge change curves of each energy storage system at the end of each scheduling cycle, calculate the deviation between the actual operating data and the predicted data of the optimization scheduling model, and input the deviation as a feedback correction term into the optimization scheduling solution unit in the edge computing module to implement online adaptive correction of the model parameters.

[0031] In this invention, the objective function of the optimization scheduling solution unit in the edge computing module consists of two parts: a network loss cost term and an energy storage operation cost term. The energy storage operation cost term is structured as follows: ; In the formula, This represents the total operating cost of energy storage, expressed in monetary units. This represents the total number of energy storage systems on the distribution network side. This represents the total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period, with charging as positive and power as the dimension; The aging cost coefficient per unit charge / discharge capacity is expressed in monetary units per unit of energy. The duration of a single time period is measured in units of time. This is the power fluctuation penalty coefficient, with dimensions of monetary units multiplied by the square of time; For the first The rated capacity of an energy storage system is expressed in units of energy. The squared term in the above energy storage operating cost item penalizes the drastic fluctuations in the amplitude of charging and discharging power during the scheduling cycle to suppress the deep charging and discharging behavior of the energy storage system, thereby extending the actual operating life of the energy storage system.

[0032] In this invention, the reactive power capacity of the energy storage converter in the local control module is obtained by taking the square root of the difference between the square of the apparent power rating at the current moment and the square of the current active power command value. When the local control module detects that the reactive power capacity is insufficient to adjust the node voltage amplitude to the preset deviation threshold range, it sends a voltage over-limit alarm message to the edge computing module and requests the distribution network side energy storage system of the adjacent node to provide reactive power assistance to collaboratively complete the voltage amplitude adjustment.

[0033] The specific embodiments of the present invention are further illustrated below: Example 1:

[0034] This embodiment is applied to the power distribution network of a suburban industrial park. The power distribution network includes 3 10kV feeders, 21 distribution transformers, a total distributed photovoltaic installed capacity of 12MW, and 6 sets of distribution network-side energy storage systems with a total capacity of 8MW / 16MWh.

[0035] Each node in the distribution network deploys data acquisition and sensing modules, and feeder terminal units are deployed at the feeder heads to collect the effective values ​​of three-phase voltage and three-phase current. Each distribution transformer's low-voltage side deploys intelligent fusion terminals to collect active and reactive power. Each distributed photovoltaic grid-connected point deploys a photovoltaic grid-connected intelligent circuit breaker to collect photovoltaic active power output. Each energy storage system is equipped with a battery management system to collect state of charge, charging and discharging power, and terminal voltage. All acquisition devices have built-in BeiDou timing modules, using the BeiDou satellite timing second pulse as the time reference, synchronizing with a local clock once per second. After adding a unified timestamp to all acquired data, it is uploaded to the edge computing node deployed in the central distribution room of the park via the distribution fiber optic communication network using the IEC61850 standard communication protocol. The edge computing node has built-in local data storage cache and optimized computing processing units. After receiving data, it first performs integrity verification and outlier removal. When data is missing or the value exceeds the corresponding preset reasonable variation range, linear interpolation of valid data from adjacent times is used to complete the data. The processed data is stored in the local cache for subsequent steps.

[0036] The edge computing module integrates a time-series prediction unit and an optimized scheduling solution unit. The time-series prediction unit employs a deep neural network structure based on gated recurrent units (GRUs). The network consists of an input layer, two GRU layers, and a fully connected output layer. The preset input features are historical load data and historical photovoltaic (PV) output data for the same time period over the past seven consecutive days, with an input data time granularity of 15 minutes. The outputs are the predicted load demand and PV output for each node in the next 24 hours. The net load prediction is the predicted load demand minus the predicted PV output. Before each rolling prediction, the time-series prediction unit performs incremental training using measured operating data from the most recent seven scheduling cycles. The training process uses a learning rate of 0.001, a batch size of 32, a mean absolute error loss function, and an upper limit of 50 iterations. After training, the network weights are updated to adaptively track the load and PV output change patterns within the park.

[0037] The optimization scheduling solution unit constructs an optimization scheduling model with the objective of minimizing the sum of distribution network losses and energy storage operating costs. Energy storage operating costs include aging losses corresponding to charging and discharging power and additional aging losses corresponding to charging and discharging power fluctuations. The model's constraints include distribution network power flow equation constraints, node voltage amplitude constraints, feeder current carrying capacity constraints, upper and lower limits of energy storage state of charge constraints, and charging and discharging power limit constraints. Specifically, the allowable fluctuation range of node voltage amplitude is ±7% of the rated voltage; the energy storage state of charge range is 20% to 90%; the maximum charging and discharging power amplitude does not exceed 1.1 times the rated power of the energy storage converter; and the feeder current carrying capacity does not exceed the rated current carrying capacity of the corresponding feeder. The rolling optimization solution adopts a two-layer time scale setting, with an upper-layer scheduling cycle of 1 hour and a time granularity of 15 minutes. A mixed-integer linear programming algorithm is used to obtain the planned active power values ​​for each energy storage system in each time period within the scheduling cycle. The lower-level optimization, based on the upper-level active power plan, uses a 5-minute scheduling interval and a 1-minute time granularity to jointly optimize the active and reactive power of the energy storage converter. The solution process prioritizes tracking the upper-level active power plan value while satisfying node voltage constraints and feeder current carrying capacity constraints. After optimization, the charging and discharging power command values ​​for each time period are obtained and sent to the local control modules of each energy storage system.

[0038] The local control module, configured as a local controller for each energy storage system, receives the charging and discharging power command value and prioritizes controlling the energy storage converter to execute the active power command. Simultaneously, it collects the voltage amplitude of the energy storage access node in real time. When the voltage amplitude exceeds the preset ±7% deviation threshold, it calculates the remaining reactive capacity of the current energy storage converter. The remaining reactive capacity is obtained by taking the square root of the difference between the square of the apparent power rating of the energy storage converter at the current moment and the square of the currently executed active power command value. This remaining reactive capacity is then used to perform local reactive power compensation for the node voltage. When the remaining reactive capacity is insufficient to adjust the node voltage back within the deviation threshold range, the local controller sends a voltage over-limit alarm message to the edge computing node. The edge computing node then schedules the energy storage systems of adjacent nodes to utilize their remaining reactive capacity to provide assistance and collaboratively complete voltage regulation.

[0039] At the end of each 1-hour scheduling cycle, the feedback correction module calls the actual charging and discharging power curves and state of charge change curves of each energy storage system, calculates the deviation between the actual operating data and the predicted data of the optimized scheduling model, and incorporates the deviation as a feedback correction term into the objective function of the optimized scheduling model for the next cycle, adaptively adjusting the penalty coefficient parameters of the optimized model to complete the online correction.

[0040] This embodiment is adapted to the operation scenario of high-proportion distributed photovoltaic access in suburban industrial parks. Beidou unified time synchronization ensures the time sequence consistency of data collected by multiple nodes. The incremental training mechanism of the gated cyclic unit prediction model can adapt to the load and photovoltaic output change patterns of the park on weekdays and weekends. The dual-layer time scale optimization takes into account both the overall operation requirements of the distribution network and the response requirements of short-term source load fluctuations. Local reactive power regulation fully taps the idle regulation capacity of the energy storage converter. The feedback correction link can avoid the continuous accumulation of prediction deviations, effectively improve the adaptability of the scheduling plan, reduce unreasonable charging and discharging behavior of the energy storage system, extend the service life of energy storage, and improve the stability of distribution network operation.

[0041] Example 2:

[0042] This embodiment is applied to the distribution network of an urban residential community. The distribution network includes two 10kV feeders, 37 unit distribution transformers, a total household distributed photovoltaic installed capacity of 3.2MW, and 12 sets of distribution network-side energy storage systems with a total capacity of 2MW / 8MWh.

[0043] Each node in the distribution network deploys data acquisition and sensing modules, and feeder terminal units are deployed at the feeder heads to collect the effective values ​​of three-phase voltage and three-phase current. Intelligent fusion terminals are deployed on the low-voltage side of each unit's distribution transformer to collect active and reactive power. Smart grid-connected circuit breakers are deployed at each household photovoltaic grid-connected point to collect photovoltaic active power output. Each energy storage system is equipped with a battery management system to collect state of charge, charging and discharging power, and terminal voltage. All acquisition devices have built-in BeiDou timing modules, using the BeiDou satellite timing second pulse as the time reference, synchronizing with a local clock once per second. After adding a unified timestamp to all acquired data, it is uploaded to the edge computing node deployed in the community's central distribution room via the power line carrier communication network using the IEC61850 standard communication protocol. The edge computing node has built-in local data storage cache and optimized computing processing units. After receiving data, it first performs integrity checks and outlier removal. When data is missing or exceeds the corresponding preset reasonable variation range, linear interpolation of valid data from adjacent times is used to complete the data. The processed data is stored in the local cache for subsequent steps.

[0044] The edge computing module integrates a time-series prediction unit and an optimized scheduling solution unit. The time-series prediction unit employs a deep neural network structure based on gated recurrent units (GRUs). The network consists of an input layer, three GRU layers, and a fully connected output layer. The preset input features are historical load data and historical photovoltaic (PV) output data for the same time period over the past 14 consecutive days, with an input data time granularity of 10 minutes. The outputs are the predicted load demand and PV output values ​​for each node in the next 12 hours. The net load prediction is the predicted load demand minus the predicted PV output. Before each rolling prediction, the time-series prediction unit performs incremental training using measured operating data from the most recent 14 scheduling cycles. The training process uses a learning rate of 0.0005, a batch size of 16, a mean absolute error loss function, and an upper limit of 40 iterations. After training, the network weights are updated to adaptively track changes in residential electricity load and household PV output patterns.

[0045] The optimization scheduling solution unit constructs an optimization scheduling model with the objective of minimizing the sum of distribution network losses and energy storage operating costs. Energy storage operating costs include aging losses corresponding to charging and discharging power and additional aging losses corresponding to fluctuations in charging and discharging power. The model's constraints include distribution network power flow equation constraints, node voltage amplitude constraints, feeder current carrying capacity constraints, upper and lower limits of energy storage state of charge constraints, and charging and discharging power limit constraints. Specifically, the allowable fluctuation range of node voltage amplitude is ±5% of the rated voltage; the energy storage state of charge operating range is 15% to 95%; the maximum amplitude of charging and discharging power cannot exceed the rated power of the energy storage converter; and the feeder current carrying capacity cannot exceed the rated current carrying capacity of the corresponding feeder. The rolling optimization solution adopts a two-layer time scale setting, with an upper-layer scheduling cycle of 30 minutes and a time granularity of 10 minutes. A second-order cone programming algorithm is used to obtain the planned active power values ​​for each energy storage system in each time period within the scheduling cycle. The lower-level optimization, based on the upper-level active power plan, uses a 2-minute scheduling interval and a time granularity of 30 seconds to jointly optimize the active and reactive power of the energy storage converter. The solution process prioritizes tracking the upper-level active power plan value while satisfying node voltage constraints and feeder current carrying capacity constraints. After optimization, the charging and discharging power command values ​​for each time period are obtained and sent to the local control modules of each energy storage system.

[0046] The local control module, configured as a local controller for each energy storage system, receives the charging and discharging power command value and prioritizes controlling the energy storage converter to execute the active power command. Simultaneously, it collects the voltage amplitude of the energy storage access node in real time. When the voltage amplitude exceeds a preset ±5% deviation threshold, it calculates the remaining reactive capacity of the current energy storage converter. The remaining reactive capacity is obtained by taking the square root of the difference between the square of the apparent power rating of the energy storage converter at the current moment and the square of the currently executed active power command value. This remaining reactive capacity is then used to perform local reactive power compensation for the node voltage. When the remaining reactive capacity is insufficient to adjust the node voltage back within the deviation threshold range, the local controller sends a voltage over-limit alarm message to the edge computing node. The edge computing node then schedules adjacent energy storage systems to utilize their remaining reactive capacity to assist in coordinating voltage regulation.

[0047] At the end of each 30-minute scheduling cycle, the feedback correction module calls the actual charging and discharging power curves and state of charge change curves of each energy storage system, calculates the deviation between the actual operating data and the predicted data of the optimized scheduling model, and incorporates this deviation as a feedback correction term into the objective function of the optimized scheduling model for the next cycle, adaptively adjusting the penalty coefficient parameters of the optimized model to complete the online correction.

[0048] This embodiment is adapted to the operation scenario of high proportion of household photovoltaic access in urban residential communities. The unified time synchronization of Beidou ensures the consistency of data timing of multiple distributed acquisition devices. The incremental training mechanism of the gated cyclic unit prediction model can adapt to the changing patterns of residential electricity consumption on weekdays, holidays and peak and valley periods. The shorter double-layer optimization time granularity can quickly respond to the short-term random fluctuations of residential load and household photovoltaic. The strict voltage deviation threshold can meet the voltage quality requirements of residential electricity consumption. The feedback correction link can adjust the prediction deviation in a timely manner to avoid the problem of overload during peak electricity consumption periods. At the same time, it takes into account the rationality of energy storage operation, reduces unnecessary charging and discharging losses, and adapts to the public service attributes of residential distribution networks.

[0049] Reference Figure 1 This flowchart illustrates the closed-loop execution logic of the energy storage-coordinated power distribution optimization control method on the distribution network side. In the initial stage, the system collects electrical operating parameters at high frequency through multi-source terminals, and uploads them to the edge computing nodes in the distribution room after unifying the timestamp using BeiDou second pulses. At the decision-making level, the edge nodes run deep learning prediction algorithms and collaborative optimization algorithms, performing global optimization scheduling based on net load prediction results with the goal of minimizing network losses and operational lifespan losses, calculating and issuing control commands. At the execution level, the energy storage controller tracks the active power plan while also considering node voltage safety, and finally, at the end of the cycle, feeds back the operating deviation to the control front end for adaptive parameter correction, forming a self-evolving, fully dynamic optimization control closed loop.

[0050] Reference Figure 2This flowchart details the preprocessing and calibration of electrical data at the sensing layer, as well as the future-oriented rolling net load forecast. Because field communication and sensors are highly susceptible to uncertain disturbances, edge computing nodes, upon receiving data from various workstations, first perform rigorous integrity checks and range reviews, then use linear interpolation to smooth out any sudden missing data or abnormal pulses. The cleaned, high-quality dataset is fed into a deep neural network model based on gated recurrent units. The model undergoes high-frequency incremental training by absorbing historical measured data from the most recent cycles. This adaptive learning mode can accurately extrapolate future load demand and photovoltaic output, ultimately outputting a stable rolling net load forecast through differential calculation.

[0051] Reference Figure 3 This flowchart clearly illustrates the two-layer time-scale solution logic for rolling optimization scheduling within edge computing. The optimization model not only considers conventional distribution network loss costs in its objective function but also innovatively introduces an aging penalty term caused by drastic fluctuations in charging and discharging power amplitude. In terms of time scale, the model is divided into two layers: the upper layer focuses on strategic active power planning for global charging and discharging power at a coarser time granularity; the lower layer, based on the optimization boundary conditions of the upper layer, performs joint deep optimization of converter active and reactive power at a finer time granularity. Through this multi-round feedback solution involving interleaved long and short cycles and constraint verification of power flow equations, refined energy storage commands that balance equipment lifespan and grid economy are output.

[0052] Reference Figure 4 This flowchart focuses on the local control terminal's autonomous anti-disturbance adjustment and multi-node collaborative reactive power compensation logic after receiving the dispatch plan. Under normal conditions, the local controller drives the energy storage converter to strictly execute the active power command. If the high-frequency monitoring network detects that the voltage amplitude of its grid-connected node exceeds the limit due to a sudden change in the output of distributed photovoltaic power, the control priority will be adjusted immediately. The converter will quickly extract its rated apparent power and current active power to dynamically calculate the current reactive power adjustment potential. If the remaining capacity is sufficient, local reactive power compensation will be directly implemented to suppress voltage fluctuations; if the local adjustment capacity is exhausted, the local system will immediately trigger an alarm to the edge layer and request neighboring energy storage units to cooperate in reactive power adjustment across nodes through the upper-level dispatch network.

[0053] Reference Figure 5This flowchart reveals the data closed-loop audit and model parameter self-calibration mechanism of the optimized control system at the end of the scheduling cycle. The control method does not blindly trust previous simulation results, but rather performs a complete review and extracts the actual charging and discharging power trajectories and dynamic state-of-charge curves of each energy storage system at the end of each scheduling time window. By comparing the variance with the theoretical expected value output by the model in the previous cycle in multiple dimensions, the system can quantify the theoretical deviations caused by sudden environmental changes and equipment aging. When this deviation triggers the correction judgment condition, the system automatically converts the value into an algebraic feedback term and injects it into the objective function of the optimization model, completing the real-time online correction of the underlying inference parameters. This ensures that the production line maintains high-precision control quality even when facing long-term, variable operating conditions.

[0054] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for coordinated distribution control of energy storage on the distribution network side, characterized in that, Includes the following steps: S1. Real-time collection of electrical operation data of each node in the power distribution network. Each collection terminal uses the second pulse of Beidou satellite time synchronization as the time reference, and after being marked with a unified timestamp, it is uploaded to the edge computing node through the power distribution communication network. S2. Utilize edge computing nodes to construct a time-series forecasting model based on historical load data and photovoltaic output data. Use the time-series forecasting model to perform rolling forecasts of the load demand and photovoltaic output of each node in the future preset time domain to obtain the net load forecast value. S3. An optimal scheduling model is constructed with the goal of minimizing the sum of distribution network losses and energy storage operating costs. Constraints are set for distribution network power flow equations, node voltage amplitude, feeder current carrying capacity, upper and lower limits of energy storage charge state, and charging and discharging power limits. Under the condition of satisfying all constraints, the preset scheduling period is used as the optimization time domain. The charging and discharging power plans of each energy storage system are solved by rolling optimization to obtain the charging and discharging power command values ​​for each time period. S4. The charging and discharging power command value is sent to the local controller of each energy storage system. The local controller monitors the voltage amplitude of the access node in real time when executing the power command. When the voltage amplitude exceeds the preset deviation threshold, the remaining reactive capacity of the energy storage converter is used to perform local reactive compensation on the node voltage under the premise of prioritizing the active power command. S5. At the end of each scheduling cycle, obtain the actual charging and discharging power curves and state of charge change curves of each energy storage system, calculate the deviation between the actual operation and the optimized scheduling prediction, use the deviation as a feedback correction term to update the objective function of the next cycle, and adaptively correct the parameters of the optimized scheduling model online.

2. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 1, characterized in that, Also includes: The objective function of the optimization scheduling model described in S3 consists of two parts: network loss cost and energy storage operation cost. The energy storage operation cost is structured as follows: ； In the formula, This represents the total operating cost of energy storage. This represents the total number of energy storage systems on the distribution network side. The total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period. The aging cost coefficient per unit charge / discharge capacity. For single time period duration, This is the power fluctuation penalty coefficient. For the first The rated capacity of each energy storage system.

3. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 1, characterized in that, Also includes: In S3, the rolling optimization solution is set as a two-layer time scale. The upper layer uses the first preset duration as the scheduling period and completes the global optimization of the charging and discharging power of each energy storage system based on the first granularity discrete time. Based on the optimization results of the upper layer, the lower layer jointly optimizes the active power and reactive power of the energy storage converter with the second preset duration and the second granularity discrete time. The lower-level optimization satisfies node voltage constraints and feeder current carrying capacity constraints, tracks the upper-level active power plan value, and performs local reactive power adjustment; the first preset duration is longer than the second preset duration, and the first granularity is coarser than the second granularity.

4. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 1, characterized in that, The electrical operation data of each node of the distribution network in S1 are jointly collected by intelligent fusion terminals, feeder terminal units and photovoltaic grid-connected intelligent circuit breakers. Each collection terminal and the edge computing node exchange data using a communication protocol based on the IEC61850 standard. The edge computing node is deployed in the power distribution room and has built-in local data storage cache and optimized computing processing unit.

5. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 2, characterized in that, The time-series prediction model described in S2 adopts a gated recurrent unit deep neural network structure. The model uses historical load data and historical photovoltaic output data of the same period over several consecutive days as input features, and uses the load value and photovoltaic output value of each node in the future preset time domain as prediction output. Before each rolling prediction, the model network weights are incrementally trained using the measured data of the most recent scheduling cycles.

6. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 1, characterized in that, The remaining reactive capacity of the energy storage converter mentioned in S4 is obtained by taking the square root of the difference between the square of the apparent power rating of the energy storage converter at the current moment and the square of the current active power command value. When the remaining reactive capacity is insufficient to adjust the node voltage amplitude back to the preset deviation threshold range, the local controller sends a voltage over-limit alarm message to the upper-level edge computing node and requests the energy storage system of the adjacent node to provide reactive power assistance.

7. The distribution network-side energy storage coordinated power distribution optimization control method according to claim 3, characterized in that, After receiving electrical operation data from each acquisition terminal, the edge computing node described in S1 performs integrity verification and outlier removal on the data. When it detects that a node's data is missing or the value of a data item exceeds the range of variation, it uses linear interpolation of valid data at adjacent time points to complete the missing or abnormal data. The data after verification and completion is used as the input data for the time series prediction model described in S2 for prediction.

8. A distribution network-side energy storage-coordinated power distribution optimization control system, characterized in that, The distribution network-side energy storage coordinated power distribution optimization control method according to any one of claims 1 to 7 includes the following modules: The data acquisition and sensing module is used to collect electrical operation data of each node of the power distribution network in real time. The data acquisition and sensing module uses the Beidou satellite time synchronization second pulse as the time reference and uploads the data to the edge computing module through the power distribution communication network. The edge computing module is used to receive electrical operation data sent by the acquisition and sensing module. It has a built-in time series prediction unit and an optimization scheduling solution unit. The time series prediction unit performs rolling prediction of the load demand and photovoltaic output of each node in the future preset time domain and calculates the net load prediction value. The optimization scheduling solution unit takes minimizing the sum of distribution network loss and energy storage operation cost as the optimization objective and solves the charging and discharging power command value of each energy storage system in each time period under the condition of simultaneously satisfying power flow constraints, voltage constraints, feeder current carrying capacity constraints and energy storage operation constraints, and sends it to the local control module. The local control module is used to receive the charging and discharging power command value issued by the edge computing module, and control the energy storage converter to execute the active power command. At the same time, it monitors the voltage amplitude of the access node in real time and uses the remaining reactive capacity of the energy storage converter to perform local reactive power compensation adjustment when the voltage amplitude exceeds the preset deviation threshold. The feedback correction module is used to acquire the actual charge and discharge power curves and state of charge change curves of each energy storage system at the end of each scheduling cycle, calculate the deviation between the actual operating data and the predicted data of the optimization scheduling model, and input the deviation as a feedback correction term into the optimization scheduling solution unit in the edge computing module to implement online adaptive correction of the model parameters.

9. A distribution network-side energy storage-coordinated power distribution optimization control system according to claim 8, characterized in that, The objective function of the optimization scheduling solution unit in the edge computing module consists of two parts: a network loss cost term and an energy storage operation cost term. The energy storage operation cost term is structured as follows: ； In the formula, This represents the total operating cost of energy storage. This represents the total number of energy storage systems on the distribution network side. The total number of time periods within the scheduling cycle. For the first The energy storage system in the first The charging and discharging power command value for each time period. The aging cost coefficient per unit charge / discharge capacity. For single time period duration, This is the power fluctuation penalty coefficient. For the first The rated capacity of each energy storage system.

10. A distribution network-side energy storage-coordinated power distribution optimization control system according to claim 8, characterized in that, The reactive power capacity of the energy storage converter in the local control module is obtained by taking the square root of the difference between the square of the apparent power rating at the current moment and the square of the active power command value executed at the current moment. When the local control module detects that the reactive power capacity is insufficient to adjust the node voltage amplitude to the preset deviation threshold range, it sends a voltage over-limit alarm message to the edge computing module and requests the distribution network side energy storage system of the adjacent node to provide reactive power assistance.

Images