A distributed control-based optical storage charging energy management method

Abstract

The application discloses a kind of photovoltaic storage charging energy management methods based on distributed control, it is related to distributed energy management technical field, comprising: step one, photovoltaic node, energy storage node and charging node operation data are collected, and local state vector is constructed;Step two, generate sub-risks power interval and construct scheduling token;Step three, exchange scheduling token and construct neighborhood token set;Step four, construct risk sub-state set and energy alliance potential space;Step five, input the improved DreamerV2 model to energy alliance candidate structure, execute counterfactual trajectory rehearsal, generate rehearsal trajectory set;Step six, construct alliance state evolution sequence and calculate alliance stability, determine target energy alliance, generate power scheduling result;Step seven, perform consistency correction and output control instruction.The application realizes dynamic collaborative scheduling and alliance evolution prediction of photovoltaic storage charging system.

Description

Technical Field

[0001] This invention relates to the fields of deep learning and energy management technology, and in particular to a photovoltaic energy management method based on distributed control for photovoltaic energy storage and charging. Background Technology

[0002] With the large-scale integration of photovoltaic (PV) power generation, energy storage systems, and electric vehicle charging facilities, integrated PV-storage-charging systems are gradually becoming an important component of new power distribution networks. Due to the volatility and randomness of PV output, the continuous changes in the operating status of energy storage devices, and the significant time-varying characteristics of charging loads, achieving coordinated scheduling among PV nodes, energy storage nodes, and charging nodes in distributed scenarios has become a crucial research direction in the energy internet field. Currently, energy management methods for PV-storage-charging systems mainly employ centralized optimization scheduling, multi-agent collaborative control, and reinforcement learning predictive control. These methods collect node operating data and generate scheduling strategies to achieve energy supply and demand balance and power distribution control.

[0003] However, existing technologies generally use single state variables or deterministic prediction results as the basis for scheduling, which makes it difficult to effectively characterize the risk propagation process caused by fluctuations in photovoltaic output, changes in the state of charge of energy storage, and uncertainties in charging load. Most existing distributed energy management methods are based on fixed topology structures for collaborative scheduling, lacking the ability to model the dynamic changes in the collaborative relationships between nodes, and failing to reflect the evolution characteristics of energy alliance structures under different operating conditions. Existing reinforcement learning or model predictive control methods usually only predict the future state of a single path, lacking a counterfactual trajectory pre-simulation mechanism for multiple potential alliance structures, resulting in limited adaptability of scheduling decisions to complex operating scenarios. In addition, existing methods lack quantitative evaluation methods for the stability of alliance structures, making it difficult to select target energy alliances suitable for the current operating state from multiple candidate collaborative structures, thus limiting the further improvement of the collaborative scheduling capability of distributed photovoltaic-storage-charging systems.

[0004] Therefore, how to provide a distributed control-based photovoltaic energy storage and charging management method is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0005] One objective of this invention is to propose a distributed control-based photovoltaic-storage-charging energy management method. This invention achieves a unified representation of the collaborative relationships among photovoltaic nodes, energy storage nodes, and charging nodes by constructing quantile risk power ranges, risk quantile state sets, and potential energy alliance spaces. Combining the alliance topology evolution mechanism and counterfactual alliance splitting mechanism in the improved DreamerV2 model, it performs parallel trajectory prediction and stability assessment on multiple energy alliance candidate structures. Finally, based on the target energy alliance, it generates and corrects the power scheduling results, realizing risk perception, alliance evolution prediction, and dynamic collaborative scheduling of the distributed photovoltaic-storage-charging system.

[0006] A photovoltaic energy storage and charging management method based on distributed control according to an embodiment of the present invention includes the following steps: Step 1: Collect photovoltaic output data from photovoltaic nodes, energy storage status data from energy storage nodes, and charging load data from charging nodes to construct the local state vector of each node; Step 2: Generate quantile risk power intervals based on the local state vector, and construct scheduling tokens based on the quantile risk power intervals; Step 3: Send the scheduling token to the distributed control node in the communication neighborhood, and receive the scheduling token sent by the distributed control node in the communication neighborhood to construct a neighborhood token set; Step 4: Construct a risk quantile state set based on the scheduling token and neighborhood token set, and construct an energy alliance potential space based on the risk quantile state set; Step 5: Generate multiple energy alliance candidate structures based on the energy alliance potential space, input each energy alliance candidate structure into the improved DreamerV2 model to perform counterfactual trajectory pre-simulation, and generate a set of pre-simulated trajectories; Step 6: Construct a coalition state evolution sequence based on the pre-simulated trajectory set, calculate the coalition stability of each energy coalition candidate structure based on the coalition state evolution sequence, determine the target energy coalition based on the coalition stability, and generate power scheduling results; Step 7: Collect actual power data and perform consistency correction on the power scheduling results to generate corrected power scheduling results and output control commands.

[0007] Optionally, step one specifically includes: Collect photovoltaic output power, photovoltaic array voltage, photovoltaic array current and irradiance data of photovoltaic nodes during continuous sampling periods; Collect state-of-charge data, battery pack terminal voltage data, battery pack current data, and battery pack temperature data of the energy storage node during the continuous sampling period; The charging node collects charging load data, charging current data, number of charging vehicles connected to the grid data, and number of charging vehicles disconnected from the grid data during the continuous sampling period. The photovoltaic node data, energy storage node data, and charging node data are synchronized and arranged according to timestamps to construct a node operation data sequence; Each time step of the node's running data sequence is used as a state element, and the data are combined in chronological order to form a local state vector.

[0008] Optionally, step two specifically includes: Read the state elements of consecutive time steps in the local state vector and construct the state evolution sequence in chronological order; For the state evolution sequence, calculate the difference sequence of photovoltaic power, energy storage state of charge and charging load at each time step. The difference is the difference between adjacent time step values, and form a power change sequence in time order. The power change sequence is sorted in ascending order of numerical values. The values ​​of the tenth, fiftieth, and ninetieth percentiles of the sorted sequence are taken to form the lower quantile, median, and upper quantile values, which constitute the quantile risk power interval. By combining the quantile risk power range with the node type identifier, a distributed control-based photovoltaic-storage-charging energy management method is proposed. The node type identifier includes photovoltaic nodes, energy storage nodes, and charging nodes. The method also calculates the values ​​of each field in the node scheduling token, where the fields include the lower quantile value, median value, upper quantile value, and the encoding of the node type identifier. The node scheduling tokens are stored in chronological order, associated with their corresponding timestamps and node identifiers, to form a node scheduling token sequence.

[0009] Optionally, step three specifically includes: The scheduling token is encapsulated with the node identifier and timestamp to form a node token data packet; The node token data packet is sent to the distributed control node in the communication neighborhood, and the node token data packet sent by the distributed control node in the communication neighborhood is received. The received node token data packets are classified according to the node identifier to form a photovoltaic node token set, an energy storage node token set, and a charging node token set; The token sets of each node are time-aligned based on the timestamp to form a group of node tokens corresponding to the same time step. The tokens are combined according to the order in which the node identifiers are arranged in the node token group to form a neighborhood token set. The neighborhood token set is stored in chronological order to form a neighborhood token sequence.

[0010] Optionally, step four specifically involves: Align the local scheduling token with the neighborhood token set according to the timestamp to form a node token group containing multiple time steps. Each node token group contains a photovoltaic node scheduling token, an energy storage node scheduling token, and a charging node scheduling token. Extract the lower quantile, median, and upper quantile values ​​from the scheduling tokens of each node, and construct photovoltaic quantile power sequences, energy storage quantile power sequences, and charging quantile power sequences according to node type and time order. For the corresponding quantile values ​​of different nodes within the same time step, calculate the quantile difference value, construct a node coupling relationship table based on each quantile difference value, and generate a node coupling weight matrix according to the degree of association between each node in the node coupling relationship table. The photovoltaic quantile power sequence, energy storage quantile power sequence and charging quantile power sequence corresponding to each time step are mapped to the corresponding node positions in the node coupling weight matrix to form a risk quantile state set containing node quantile characteristics, node coupling characteristics and time evolution characteristics. Based on the state change relationship of each node in the risk quantile state set between consecutive time steps, time-related edges are established, and node-related edges are established based on the node coupling weight matrix, forming a multi-layered network structure. The multi-layered network structure is mapped to the latent space coordinate system to construct the potential space of the energy alliance. In a distributed control-based photovoltaic energy management method, the node position in the potential space of the energy alliance represents the node risk status, and the connection relationship between nodes represents the energy synergy relationship between nodes. Based on the trajectory of node position changes and node connection relationships in the potential space of the energy alliance, potential connection matrices corresponding to multiple candidate structures of the energy alliance are generated.

[0011] Optionally, step five specifically includes: Input the potential connectivity matrix corresponding to each energy alliance candidate structure into the improved DreamerV2 model; The improved DreamerV2 model includes a risk quantile state encoding module, a coalition topology evolution module, a counterfactual coalition splitting module, and a coalition stability feedback module. Input the potential connection matrix corresponding to each energy alliance candidate structure into the risk quantile state encoding module, extract the quantile state features and node coupling weights of each node, generate the alliance potential state vector according to node type and time order, each dimension consists of node quantile offset and node coupling weight, and interpolate or copy the previous valid state for missing or abnormal node states. The generated alliance potential state vector is input into the alliance topology evolution module, the alliance potential state of the current time step is read, the alliance potential connection relationship of the next time step is generated according to the node quantile state change and node coupling weight, and the alliance potential state sequence is updated. At the same time, the newly added or disconnected node connection information is marked. The updated sequence of potential alliance states is input into the counterfactual alliance splitting module, which generates multiple parallel alliance structure branches for the same potential alliance state. The node connection relationship of each branch is adjusted independently according to the quantile state, and the future trajectories of each branch are expanded to form a set of counterfactual trajectories. The counterfactual trajectory set is input into the alliance stability feedback module. The proportion of alliance members and the proportion of node connections maintained in each trajectory are statistically analyzed to construct the alliance state evolution sequence. The alliance stability corresponding to each alliance state evolution sequence is calculated, and the pre-simulated trajectory set is output.

[0012] Optionally, the alliance topology evolution specifically includes: Read the node quantile state characteristics and node coupling weights in the potential state of the alliance at the current time step; For any two nodes, calculate the quantile state difference between the quantile state features of the nodes at the current time step, and generate the node association value by combining the corresponding node coupling weights. Construct a node association sequence based on the association values ​​of each node, and generate the node connection relationship for the next time step based on the node association sequence; Compare the node connection relationships of the next time step with the node connection relationships of the current time step, mark newly formed node connections as new connections, and mark missing node connections as disconnected connections. Update the potential state sequence of the alliance based on the node connection relationship of the next time step, and write the newly added connection marker, disconnection marker and corresponding time step information into the potential state sequence of the alliance. The potential state sequence of the alliance at each time step is continuously updated in chronological order to form the evolution sequence of the alliance topology.

[0013] Optionally, the counterfactual alliance splitting module specifically comprises: Read the potential state sequence of the alliance corresponding to the current time step, and extract the quantile state features and node coupling weights of each node in the alliance. A node association sequence is constructed based on the node quantile state characteristics and node coupling weights. The node association sequence records the association relationships between each node. Multiple node connection relationship combinations are generated according to the node association sequence, and each node connection relationship combination corresponds to a kind of alliance structure branch. Each alliance structure branch is mapped to an independent potential state space, and continuous time step trajectory unfolding is performed according to the alliance potential state transition process; Record the changes in node position status and node connection relationships of each alliance structure branch during the trajectory unfolding process to form a counterfactual trajectory set.

[0014] Optionally, step six specifically includes: Read the node quantile status change information and node connection relationship change information corresponding to each counterfactual trajectory in the pre-simulated trajectory set; The information on changes in node quantile states and changes in node connection relationships are correlated and arranged in chronological order to construct a coalition state evolution sequence. The number of times the node member state changes and the number of times the node connection state changes within each time step in the alliance state evolution sequence are statistically analyzed, and the cumulative change values ​​are used to form the alliance evolution characteristic sequence. The stability of each energy alliance candidate structure is calculated based on the alliance evolution characteristic sequence, and an alliance ranking sequence is formed according to the alliance stability. Read the candidate structure of the energy alliance with the highest sorting position in the alliance sorting sequence to determine the target energy alliance; Extract the node connection relationships and node position status information corresponding to the target energy alliance, and generate the power allocation relationship between nodes; Based on the node allocation status and inter-node connectivity of the target energy alliance, the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the load power of charging nodes are allocated to form a power dispatch result.

[0015] Optionally, step seven specifically includes: Collect the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the actual load power of charging nodes to form an actual execution power data sequence; The actual execution power data sequence is compared with the power scheduling results according to the time series, and the power deviation value of each node is calculated. Based on the power deviation values ​​of each node, the output power of the photovoltaic node, the charging and discharging power of the energy storage node, and the load power of the charging node are corrected for consistency, and the corrected power dispatch results are generated. Distributed control commands are generated based on the corrected power scheduling results and sent to photovoltaic nodes, energy storage nodes, and charging nodes respectively to achieve dynamic execution of energy management.

[0016] The beneficial effects of this invention are: This invention constructs a scheduling token system based on quantile risk power intervals, uniformly mapping the operating states of photovoltaic nodes, energy storage nodes, and charging nodes to interactive risk representation data. It also combines a neighborhood token interaction mechanism to construct a risk quantile state set, effectively improving the consistency and temporal correlation of operating state expressions among distributed nodes. In the energy alliance construction phase, it introduces a node coupling weight matrix, a multi-layered relational network structure, and an energy alliance potential space modeling mechanism to uniformly represent node quantile state characteristics, node coupling relationship characteristics, and temporal evolution characteristics, achieving joint modeling of energy synergy relationships and risk evolution processes. In the alliance prediction phase, an improved DreamerV2 model is designed, mapping node risk characteristics to the alliance through a risk quantile state encoding mechanism. The system constructs a potential state space for energy alliances. During state transitions, it introduces an alliance topology evolution mechanism to dynamically model the changes in node connectivity. Combined with a counterfactual alliance splitting mechanism, it generates multiple alliance structural branches and corresponding counterfactual trajectories, enabling parallel pre-simulation of different energy alliance evolution paths. In the trajectory evaluation stage, it constructs an alliance stability feedback mechanism, forming an alliance state evolution sequence through changes in alliance members and node connectivity, and quantitatively evaluates the stability of multiple energy alliance candidate structures. Finally, it generates power allocation relationships between nodes based on the target energy alliance, performs consistency correction based on actual power data, and outputs the corrected power scheduling results and control commands, realizing risk perception, alliance evolution prediction, and dynamic collaborative scheduling of the distributed photovoltaic energy storage and charging system. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is a schematic diagram of the overall process of a photovoltaic energy storage and charging management method based on distributed control proposed in this invention; Figure 2 This is an overall structural diagram of the improved DreamerV2 model in the distributed control-based photovoltaic energy storage and charging energy management method proposed in this invention. Detailed Implementation

[0018] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0019] refer to Figures 1-2 A photovoltaic energy storage and charging management method based on distributed control includes the following steps: Step 1: Collect photovoltaic output data from photovoltaic nodes, energy storage status data from energy storage nodes, and charging load data from charging nodes to construct the local state vector of each node; Step 2: Generate quantile risk power intervals based on local state vectors, and construct scheduling tokens based on quantile risk power intervals; Step 3: Send the scheduling token to the distributed control node in the communication neighborhood, and receive the scheduling token sent by the distributed control node in the communication neighborhood to build a neighborhood token set; Step 4: Construct a risk quantile state set based on the scheduling token and neighborhood token set, and construct the potential space of the energy alliance based on the risk quantile state set; Step 5: Generate multiple candidate structures for energy alliances based on the potential space of energy alliances, input each candidate structure into the improved DreamerV2 model to perform counterfactual trajectory pre-simulation, and generate a set of pre-simulated trajectories; Step 6: Construct the coalition state evolution sequence based on the pre-simulated trajectory set, calculate the coalition stability of each energy coalition candidate structure based on the coalition state evolution sequence, determine the target energy coalition based on the coalition stability, and generate power scheduling results; Step 7: Collect actual power data and perform consistency correction on the power scheduling results, generate corrected power scheduling results, and output control commands.

[0020] In this embodiment, step one specifically includes: Collect photovoltaic output power, photovoltaic array voltage, photovoltaic array current and irradiance data of photovoltaic nodes during continuous sampling periods; Collect state-of-charge data, battery pack terminal voltage data, battery pack current data, and battery pack temperature data of energy storage nodes during continuous sampling periods; Collect charging load data, charging current data, number of charging vehicles connected to the grid data, and number of charging vehicles disconnected from the grid data of charging nodes within a continuous sampling period; The photovoltaic node data, energy storage node data, and charging node data are synchronized and arranged according to timestamps to construct a node operation data sequence; Each time step in the node's running data sequence is used as a state element, and they are combined in chronological order to form a local state vector.

[0021] In this implementation, the node operation data sequence is time-synchronized using a unified clock source, and adjacent sampling periods are kept consistent to ensure that photovoltaic nodes, energy storage nodes, and charging nodes are characterized under the same time reference. The local state vector is organized using a multi-dimensional time-series structure, with each dimension of state data corresponding to a single physical quantity, and each time step state element corresponding to the node operation state at the same sampling time. For missing sampled data, valid data from adjacent time steps is used to fill in the gaps, and for abnormal sampled data, sliding window statistical results are used for correction. The photovoltaic output power, energy storage state of charge, and charging load data retain their original change characteristics and time correlations before entering the local state vector, without undergoing dimensionality reduction processing. The node operation data sequence is stored using a continuous time window method, with the time window length dynamically adjusted according to the actual operation cycle, so that the local state vector simultaneously includes the current operation state and historical evolution characteristics. The constructed local state vector is stored independently according to the node identifier and appended with corresponding timestamp information to form a standardized time-series data structure that can be used for subsequent risk quantile state analysis.

[0022] In this embodiment, step two specifically involves: Read the state elements of consecutive time steps in the local state vector and construct the state evolution sequence in chronological order; The difference sequence of photovoltaic power, energy storage state of charge and charging load at each time step is calculated for the state evolution sequence. The difference is the difference between the values ​​of adjacent time steps, and the power change sequence is formed in time order. Sort the power change sequence in ascending order of values, and take the values ​​of the tenth, fiftieth, and ninetieth digits of the sorted sequence to form the lower quantile, median, and upper quantile values, thus constituting the quantile risk power interval. By combining the quantile risk power range with the node type identifier, a distributed control-based photovoltaic-storage-charging energy management method is proposed. The node type identifier includes photovoltaic nodes, energy storage nodes, and charging nodes. The method also calculates the values ​​of each field in the node scheduling token, including the lower quantile value, median value, upper quantile value, and the encoding of the node type identifier. The node scheduling tokens are stored in chronological order, associated with their corresponding timestamps and node identifiers, forming a node scheduling token sequence.

[0023] In this implementation, quantile risk power intervals are constructed independently for photovoltaic nodes, energy storage nodes, and charging nodes. The state data corresponding to node types are processed using state evolution sequences of uniform time length to ensure that the risk representation of different nodes has a consistent data scale. During the generation of power change sequences, median filtering is used to smooth data abrupt changes occurring in multiple consecutive time steps to preserve the power change trend and suppress the impact of abnormal fluctuations. During quantile value extraction, the sequence positions corresponding to the lower quantile, median, and upper quantile values ​​are determined using sorted position indices. When the target position is between two adjacent data points, linear interpolation is used to determine the quantile value. The node type identifier encoding adopts a unique numerical encoding rule and is encapsulated together with the quantile risk power intervals into a unified data structure. The node scheduling token sequence is stored in ascending time order, and the scheduling tokens of adjacent time steps maintain a continuous index relationship, thereby forming a time-series token dataset that can characterize the node risk evolution process.

[0024] In this embodiment, step three specifically includes: The scheduling token is encapsulated with the node identifier and timestamp to form a node token data packet; Send node token packets to distributed control nodes within the communication neighborhood, and receive node token packets sent by distributed control nodes within the communication neighborhood; The received node token data packets are classified according to the node identifier to form a photovoltaic node token set, an energy storage node token set, and a charging node token set; The token sets of each node are time-aligned based on the timestamp to form a group of node tokens corresponding to the same time step. The tokens are combined according to the order in which the node identifiers are arranged in the node token group to form a neighborhood token set. The neighborhood token set is stored in chronological order to form a neighborhood token sequence.

[0025] In this implementation, the communication neighborhood is dynamically determined based on the reachable communication links between nodes. Each node maintains a neighborhood node identifier table. Node token data packets are encapsulated in a unified data format, and the data fields are arranged in a fixed order. During time alignment, the sampling period corresponding to the timestamp is used as the alignment benchmark. When there is a deviation in the arrival time of different node token data packets, they are merged according to the sampling period to which the timestamp belongs. Each node identifier in the node token group is represented by a unique encoding method and arranged according to the node type encoding priority and the numerical order of the node identifier. The neighborhood token set adopts a matrix storage structure, with each row corresponding to a time step and each column corresponding to a neighborhood node, thereby maintaining the temporal and spatial correlation between neighborhood nodes. The neighborhood token sequence adopts a cyclic update mechanism, with new node token groups written to the current time step position, and historical node token groups retained in chronological order to form a continuous neighborhood state evolution dataset.

[0026] In this embodiment, step four specifically includes: Align the local scheduling token with the neighborhood token set according to the timestamp to form a node token group containing multiple time steps. Each node token group contains a photovoltaic node scheduling token, an energy storage node scheduling token, and a charging node scheduling token. Extract the lower quantile, median, and upper quantile values ​​from the scheduling tokens of each node, and construct photovoltaic quantile power sequences, energy storage quantile power sequences, and charging quantile power sequences according to node type and time order. For the corresponding quantile values ​​of different nodes within the same time step, calculate the quantile difference value, construct a node coupling relationship table based on each quantile difference value, and generate a node coupling weight matrix according to the degree of association between each node in the node coupling relationship table. The photovoltaic quantile power sequence, energy storage quantile power sequence and charging quantile power sequence corresponding to each time step are mapped to the corresponding node positions in the node coupling weight matrix to form a risk quantile state set containing node quantile characteristics, node coupling characteristics and time evolution characteristics. Based on the state change relationship of each node in the risk quantile state set between consecutive time steps, time-related edges are established, and node-related edges are established based on the node coupling weight matrix, forming a multi-layered network structure. By mapping the multi-layered network structure to the latent space coordinate system, a potential space for energy alliances is constructed. A distributed control-based photovoltaic energy management method for energy storage and charging is proposed. The node positions in the potential space of the energy alliance represent the node risk status, and the connection relationships between nodes represent the energy synergy relationships between nodes. Based on the trajectory of node position changes and node connection relationships in the potential space of the energy alliance, potential connection matrices corresponding to multiple candidate structures of the energy alliance are generated.

[0027] In this implementation, the node coupling relationship table is maintained using a time-varying update method, and the degree of association between nodes within the same time step is updated synchronously with the change of quantile power sequence; the node coupling weight matrix is ​​stored using a symmetric matrix structure, with the main diagonal elements recording the node's own state information and the off-diagonal elements recording the association information between nodes; the risk quantile state set is organized using a tensor data structure, with one dimension corresponding to the time step, another dimension corresponding to the node identifier, and the remaining dimensions corresponding to the quantile state characteristics and coupling characteristics; in the multi-layered associated network structure, time-related edges and node-related edges are managed using independent indexes and fused in a unified topology structure; the latent space coordinate system is represented using a continuous vector space, and the mapped coordinate positions of each node maintain time continuity constraints, with the latent space position changes of corresponding nodes in adjacent time steps consistent with the risk quantile state changes, thus forming a potential energy alliance space that can characterize the evolution process of energy synergy relationships.

[0028] In this embodiment, step five specifically includes: Input the potential connectivity matrix corresponding to each energy alliance candidate structure into the improved DreamerV2 model; The improved DreamerV2 model includes a risk quantile state encoding module, a coalition topology evolution module, a counterfactual coalition splitting module, and a coalition stability feedback module. Input the potential connection matrix corresponding to each energy alliance candidate structure into the risk quantile state encoding module, extract the quantile state features and node coupling weights of each node, generate the alliance potential state vector according to node type and time order, each dimension consists of node quantile offset and node coupling weight, and interpolate or copy the previous valid state for missing or abnormal node states. The generated alliance potential state vector is input into the alliance topology evolution module, the alliance potential state of the current time step is read, the alliance potential connection relationship of the next time step is generated according to the node quantile state change and node coupling weight, and the alliance potential state sequence is updated. At the same time, the newly added or disconnected node connection information is marked. The updated sequence of potential alliance states is input into the counterfactual alliance splitting module, which generates multiple parallel alliance structure branches for the same potential alliance state. The node connection relationship of each branch is adjusted independently according to the quantile state, and the future trajectories of each branch are expanded to form a set of counterfactual trajectories. The counterfactual trajectory set is input into the coalition stability feedback module. The coalition member retention rate and node connection retention rate in each trajectory are statistically analyzed to construct the coalition state evolution sequence. The coalition stability corresponding to each coalition state evolution sequence is calculated, and the pre-simulated trajectory set is output.

[0029] In this implementation, after inputting the potential connection matrices corresponding to the candidate structures of each energy alliance into the improved DreamerV2 model, the system first extracts and fuses the quantile state features and node coupling weights of each node through the risk quantile state encoding module to generate the alliance potential state vector, and performs continuity processing on abnormal or missing node states to ensure the integrity of the time series; then, the alliance topology evolution module dynamically generates the node connection relationship of the next time step based on the changes in node quantile states and coupling relationships, while recording the information of newly added and disconnected node connections and updating the alliance potential state sequence; the counterfactual alliance splitting module establishes multiple parallel alliance structure branches based on the generated alliance potential state sequence, expands the future trajectory of each branch to form a counterfactual trajectory set; the alliance stability feedback module counts the proportion of alliance members and the proportion of node connections maintained in each trajectory, constructs the alliance state evolution sequence, and calculates the alliance stability, realizing parallel pre-simulation and stability quantification of different energy alliance candidate structures, providing data support for the selection of target energy alliances, and ensuring the feasibility and reliability of power scheduling schemes under dynamic conditions; The improved DreamerV2 model retains the processing framework of the DreamerV2 model based on latent state space for environmental modeling and trajectory prediction. It uses latent states to represent the current system operating state and performs future state prediction based on the latent state sequence. The improved DreamerV2 model also includes state encoding, state transition and trajectory prediction processing. It uses the latent state space to replace the actual environment space to complete trajectory generation, thereby reducing the computational overhead in the trajectory prediction process. The improved DreamerV2 model modifies the DreamerV2 model's approach from predicting single environmental states to predicting the evolutionary states of energy alliances. In the potential state construction stage, a risk quantile state encoding mechanism is introduced, mapping node quantile state characteristics and node coupling relationships to the alliance's potential state. In the state transition stage, an alliance topology evolution mechanism is introduced, dynamically adjusting the alliance connection structure based on changes in node quantile states and node coupling relationships. In the trajectory pre-simulation stage, a counterfactual alliance splitting mechanism is introduced, generating multiple alliance structure branches for the same potential state and performing trajectory expansion for each. Finally, in the trajectory evaluation stage, an alliance stability feedback mechanism is introduced, quantifying the stability of the alliance evolution process corresponding to each trajectory. The improved DreamerV2 model can simultaneously characterize changes in node risk states and alliance structure during trajectory pre-simulation, forming multiple counterfactual trajectories corresponding to different alliance structures. At the same time, it uses alliance stability to uniformly evaluate different trajectories, enabling the energy alliance selection process to take into account both node collaborative relationships and risk evolution characteristics. This forms an alliance evolution prediction mechanism suitable for distributed photovoltaic-storage-charging scenarios and improves the temporal consistency and structural stability of the energy alliance construction process and the power scheduling process.

[0030] In this embodiment, the alliance topology evolution is specifically as follows: Read the node quantile state characteristics and node coupling weights in the potential state of the alliance at the current time step; For any two nodes, calculate the quantile state difference between the quantile state features of the nodes at the current time step, and generate the node association value by combining the corresponding node coupling weights. Construct a node association sequence based on the association values ​​of each node, and generate the node connection relationship for the next time step based on the node association sequence; Compare the node connection relationships of the next time step with the node connection relationships of the current time step, mark newly formed node connections as new connections, and mark missing node connections as disconnected connections. Update the potential state sequence of the alliance based on the node connection relationship of the next time step, and write the newly added connection marker, disconnection marker and corresponding time step information into the potential state sequence of the alliance. The potential state sequence of the alliance at each time step is continuously updated in chronological order to form the evolution sequence of the alliance topology.

[0031] In this embodiment, the counterfactual alliance splitting module is specifically as follows: Read the potential state sequence of the alliance corresponding to the current time step, and extract the quantile state features and node coupling weights of each node in the alliance. A node association sequence is constructed based on the node quantile state characteristics and node coupling weights. The node association sequence records the association relationships between each node. Multiple node connection relationship combinations are generated according to the node association sequence, and each node connection relationship combination corresponds to a kind of alliance structure branch. Each alliance structure branch is mapped to an independent potential state space, and continuous time step trajectory unfolding is performed according to the alliance potential state transition process; Record the changes in node position status and node connection relationships of each alliance structure branch during the trajectory unfolding process to form a counterfactual trajectory set.

[0032] In this implementation, the counterfactual alliance splitting module analyzes the quantile state characteristics and node coupling weights of each node in the potential state sequence of the alliance, calculates the differences in quantile states between nodes, and generates a node risk ranking sequence accordingly. Based on this, the node connection relationships are adjusted according to the ranking results, disconnecting connections between high-risk nodes and establishing connections between nodes with low differences, thereby forming multiple different alliance structure branches. Each branch independently unfolds its future trajectory within a continuous time step, recording changes in node quantile states and node connections. At the same time, it maintains temporal continuity by interpolating or copying the previous valid state for missing or abnormal node states. By statistically analyzing the changes in alliance members and connections in each branch trajectory, an alliance state evolution sequence is constructed, providing input for subsequent calculations of alliance stability. The generated counterfactual trajectory sets can fully characterize the dynamic changes of the potential alliance structure under various risk evolution conditions, providing a comprehensive and multi-branch predictive basis for energy alliance selection and power scheduling.

[0033] In this embodiment, step six specifically includes: Read the node quantile status change information and node connection relationship change information corresponding to each counterfactual trajectory in the pre-simulated trajectory set; The information on changes in node quantile states and changes in node connection relationships are correlated and arranged in chronological order to construct a coalition state evolution sequence. The number of times the node member state changes and the number of times the node connection state changes within each time step in the alliance state evolution sequence are statistically analyzed, and the cumulative change values ​​are used to form the alliance evolution characteristic sequence. The stability of each energy alliance candidate structure is calculated based on the alliance evolution characteristic sequence, and an alliance ranking sequence is formed according to the alliance stability. Read the candidate structure of the energy alliance with the highest sorting position in the alliance sorting sequence to determine the target energy alliance; Extract the node connection relationships and node position status information corresponding to the target energy alliance, and generate the power allocation relationship between nodes; Based on the node allocation status and inter-node connectivity of the target energy alliance, the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the load power of charging nodes are allocated to form a power dispatch result.

[0034] In this implementation, the alliance state evolution sequence is formed by sequentially arranging the node quantile state changes and node connection relationship changes of each counterfactual trajectory within continuous time steps, ensuring that the quantile state and connection state of each node are completely recorded in chronological order. When statistically analyzing the alliance evolution feature sequence, the state changes and connection changes of each node at each time step are cumulatively counted, while retaining the node type and spatial location identifier to describe the internal structural evolution of the alliance. During the alliance stability calculation, the retention status of each node member and the retention status of node connections in the alliance evolution feature sequence are comprehensively mapped into a stability index, and the target energy alliance is identified by sorting the sequence. The power allocation relationship between nodes is generated based on the node quantile state and connection relationship of the target alliance. Photovoltaic output power, energy storage charging and discharging power, and charging load are allocated according to the cooperative relationship between nodes, ensuring that the power scheduling result is consistent with the alliance state evolution sequence in time series, thereby realizing the dynamic coordination and stable operation of distributed photovoltaic-storage-charging energy management.

[0035] In this embodiment, step seven specifically includes: Collect the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the actual load power of charging nodes to form an actual execution power data sequence; The actual power data sequence and the power scheduling results are compared in a time series manner, and the power deviation value of each node is calculated. Based on the power deviation values ​​of each node, the output power of the photovoltaic node, the charging and discharging power of the energy storage node, and the load power of the charging node are corrected for consistency, and the corrected power dispatch results are generated. Distributed control commands are generated based on the corrected power scheduling results and sent to photovoltaic nodes, energy storage nodes, and charging nodes respectively to achieve dynamic execution of energy management.

[0036] In this implementation, the actual power data sequence is formed by collecting real-time power information from each photovoltaic node, energy storage node, and charging node, and arranged in chronological order to ensure the correspondence between node power status and power scheduling results. During the consistency correction process, the power deviation of each node in continuous time steps is calculated, and the node output power is adjusted according to the deviation magnitude, while maintaining the power distribution ratio between nodes consistent with the internal topology of the energy alliance. The corrected power scheduling results are converted into control signals that can be sent to each node through the distributed control command generation module. Each command includes node identifier, target power, and execution time step information to ensure that photovoltaic power generation, energy storage charging and discharging, and charging load are consistent with the scheduling results in actual operation, forming a continuous, traceable, and dynamically responsive energy management and control chain.

[0037] Example 1: To verify the feasibility of this invention in practical applications, it was applied to a photovoltaic-storage-charging cluster system in a highway integrated energy service area. The service area has 8 distributed photovoltaic power generation nodes, 4 energy storage nodes, and 36 DC fast charging terminals, with a total photovoltaic installed capacity of 3.2MW and a total energy storage capacity of 5.6MWh. During holidays, the concentrated entry of new energy vehicles into the service area causes a significant surge in charging load, with frequent short-term rapid load increases during midday. Simultaneously, cloud cover causes fluctuations in photovoltaic output, making it difficult for traditional energy dispatching methods based on fixed topology to respond promptly to load changes. This can easily lead to situations where some charging areas have insufficient power while other areas have idle energy storage.

[0038] During actual operation, the system continuously collects data on the output power of photovoltaic nodes, the state of charge of energy storage nodes, and the load of charging terminals to construct local state vectors for each node and generate corresponding scheduling tokens. Each node exchanges scheduling tokens through its communication neighborhood, forming a neighborhood token set. The system further constructs a risk quantile state set and a potential energy alliance space to dynamically characterize the collaborative relationships between different nodes. As the number of charging vehicles changes, multiple candidate energy alliance structures emerge among different photovoltaic nodes, energy storage nodes, and charging nodes.

[0039] During the energy alliance construction phase, this invention employs an improved DreamerV2 model to perform counterfactual trajectory simulations of various candidate energy alliance structures. For energy alliance structures formed at the same time, the system simultaneously generates multiple alliance structure branches and simulates the alliance evolution process under different load change scenarios within the next 30 minutes. The alliance topology evolution mechanism continuously adjusts the connection relationships between nodes, the counterfactual alliance splitting mechanism synchronously generates multiple alliance evolution trajectories, and the alliance stability feedback mechanism evaluates each trajectory, ultimately determining the target energy alliance and generating corresponding power scheduling results.

[0040] This invention can maintain the stable operation of the energy alliance structure under conditions of rapid fluctuations in charging load, and significantly improve the utilization level of photovoltaics and the synergistic efficiency of energy storage. At the same time, due to the introduction of a counterfactual alliance splitting mechanism, the system can identify potential alliance instability states in advance and complete the alliance structure reconstruction before instability occurs, thereby reducing power dispatch deviations.

[0041] To verify the actual effect of the present invention, three consecutive days of operating data were used as test samples and compared with traditional distributed power allocation methods. The experimental results are shown in Table 1.

[0042] Table 1. Comparison of Operational Performance of Photovoltaic-Storage-Charging Clusters in Highway Service Areas

[0043] As shown in Table 1, this invention outperforms traditional distributed scheduling methods in all key indicators. Photovoltaic utilization rate increased from 73.8% to 89.6%, indicating that this invention can improve the absorption capacity of photovoltaic power generation resources and reduce curtailment. Energy storage utilization rate increased from 76.4% to 91.8%, demonstrating a significant improvement in the participation of energy storage nodes in energy collaborative scheduling. Charging load satisfaction rate reached 97.9%, while charging waiting time decreased from 18.4 min to 6.2 min, indicating that this invention can more effectively coordinate the matching relationship between charging demand and energy supply capacity. Alliance stability score increased from 0.71 to 0.93, and alliance member retention rate and node connection retention rate increased to 95.7% and 94.1%, respectively. This demonstrates that through energy alliance potential space modeling and alliance topology evolution mechanism, the continuous stability of the energy alliance structure can be maintained, verifying that the improved DreamerV2 model can accurately predict different alliance evolution paths and complete alliance structure optimization. Furthermore, the power scheduling deviation was reduced to 2.6%, and the control command execution consistency rate was increased to 98.4%, indicating that the consistency correction mechanism can effectively reduce the deviation between the actual execution results and the scheduling results. Comprehensive analysis shows that this invention can realize the dynamic construction of energy alliances, prediction of alliance evolution, and coordinated power scheduling, thereby improving the overall operating efficiency and stability of photovoltaic-storage-charging systems.

[0044] The above description is only a preferred embodiment 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 photovoltaic energy storage and charging management method based on distributed control, characterized in that, Includes the following steps: Step 1: Collect photovoltaic output data from photovoltaic nodes, energy storage status data from energy storage nodes, and charging load data from charging nodes to construct the local state vector of each node; Step 2: Generate quantile risk power intervals based on the local state vector, and construct scheduling tokens based on the quantile risk power intervals; Step 3: Send the scheduling token to the distributed control node in the communication neighborhood, and receive the scheduling token sent by the distributed control node in the communication neighborhood to construct a neighborhood token set; Step 4: Construct a risk quantile state set based on the scheduling token and neighborhood token set, and construct an energy alliance potential space based on the risk quantile state set; Step 5: Generate multiple energy alliance candidate structures based on the energy alliance potential space, input each energy alliance candidate structure into the improved DreamerV2 model to perform counterfactual trajectory pre-simulation, and generate a set of pre-simulated trajectories; Step 6: Construct a coalition state evolution sequence based on the pre-simulated trajectory set, calculate the coalition stability of each energy coalition candidate structure based on the coalition state evolution sequence, determine the target energy coalition based on the coalition stability, and generate power scheduling results; Step 7: Collect actual power data and perform consistency correction on the power scheduling results to generate corrected power scheduling results and output control commands.

2. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step one specifically involves: Collect photovoltaic output power, photovoltaic array voltage, photovoltaic array current and irradiance data of photovoltaic nodes during continuous sampling periods; Collect state-of-charge data, battery pack terminal voltage data, battery pack current data, and battery pack temperature data of the energy storage node during the continuous sampling period; The charging node collects charging load data, charging current data, number of charging vehicles connected to the grid data, and number of charging vehicles disconnected from the grid data during the continuous sampling period. The photovoltaic node data, energy storage node data, and charging node data are synchronized and arranged according to timestamps to construct a node operation data sequence; Each time step of the node's running data sequence is used as a state element, and the data are combined in chronological order to form a local state vector.

3. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step two specifically involves: Read the state elements of consecutive time steps in the local state vector and construct the state evolution sequence in chronological order; The difference sequence of photovoltaic power, energy storage state of charge and charging load at each time step is calculated for the state evolution sequence. The difference is the difference between adjacent time step values, and a power change sequence is formed in time order. The power change sequence is sorted in ascending order of numerical values. The values ​​of the tenth, fiftieth, and ninetieth percentiles of the sorted sequence are taken to form the lower percentile, median, and upper percentile values, which constitute the percentile risk power interval. The quantile risk power range is combined with the node type identifier, which includes photovoltaic nodes, energy storage nodes, and charging nodes. The values ​​of each field in the node scheduling token are calculated, including the lower quantile value, median value, upper quantile value, and the encoding of the node type identifier. The node scheduling tokens are stored in chronological order, associated with their corresponding timestamps and node identifiers, to form a node scheduling token sequence.

4. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step three specifically involves: The scheduling token is encapsulated with the node identifier and timestamp to form a node token data packet; The node token data packet is sent to the distributed control node in the communication neighborhood, and the node token data packet sent by the distributed control node in the communication neighborhood is received. The received node token data packets are classified according to the node identifier to form a photovoltaic node token set, an energy storage node token set, and a charging node token set; The token sets of each node are time-aligned based on the timestamp to form a group of node tokens corresponding to the same time step. The tokens are combined according to the order in which the node identifiers are arranged in the node token group to form a neighborhood token set. The neighborhood token set is stored in chronological order to form a neighborhood token sequence.

5. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step four specifically involves: Align the local scheduling token with the neighborhood token set according to the timestamp to form a node token group containing multiple time steps. Each node token group contains a photovoltaic node scheduling token, an energy storage node scheduling token, and a charging node scheduling token. Extract the lower quantile, median, and upper quantile values ​​from the scheduling tokens of each node, and construct photovoltaic quantile power sequences, energy storage quantile power sequences, and charging quantile power sequences according to node type and time order. For the corresponding quantile values ​​of different nodes within the same time step, calculate the quantile difference value, construct a node coupling relationship table based on each quantile difference value, and generate a node coupling weight matrix according to the degree of association between each node in the node coupling relationship table. The photovoltaic quantile power sequence, energy storage quantile power sequence and charging quantile power sequence corresponding to each time step are mapped to the corresponding node positions in the node coupling weight matrix to form a risk quantile state set containing node quantile characteristics, node coupling characteristics and time evolution characteristics. Based on the state change relationship of each node in the risk quantile state set between consecutive time steps, time-related edges are established, and node-related edges are established based on the node coupling weight matrix, forming a multi-layered network structure. The multi-layered network structure is mapped to the latent space coordinate system to construct the potential space of the energy alliance. The node position in the potential space of the energy alliance represents the node risk status, and the connection relationship between nodes represents the energy synergy relationship between nodes. Based on the trajectory of node position changes and node connection relationships in the potential space of the energy alliance, potential connection matrices corresponding to multiple candidate structures of the energy alliance are generated.

6. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step five specifically involves: Input the potential connectivity matrix corresponding to each energy alliance candidate structure into the improved DreamerV2 model; The improved DreamerV2 model includes a risk quantile state encoding module, a coalition topology evolution module, a counterfactual coalition splitting module, and a coalition stability feedback module. Input the potential connection matrix corresponding to each energy alliance candidate structure into the risk quantile state encoding module, extract the quantile state features and node coupling weights of each node, generate the alliance potential state vector according to node type and time order, each dimension consists of node quantile offset and node coupling weight, and interpolate or copy the previous valid state for missing or abnormal node states. The generated alliance potential state vector is input into the alliance topology evolution module, the alliance potential state of the current time step is read, the alliance potential connection relationship of the next time step is generated according to the node quantile state change and node coupling weight, and the alliance potential state sequence is updated. At the same time, the newly added or disconnected node connection information is marked. The updated sequence of potential alliance states is input into the counterfactual alliance splitting module, which generates multiple parallel alliance structure branches for the same potential alliance state. The node connection relationship of each branch is adjusted independently according to the quantile state, and the future trajectories of each branch are expanded to form a set of counterfactual trajectories. The counterfactual trajectory set is input into the alliance stability feedback module. The retention rates of alliance members and node connections in each trajectory are statistically analyzed to construct an alliance state evolution sequence. The alliance stability corresponding to each alliance state evolution sequence is calculated, and the pre-simulated trajectory set is output.

7. The photovoltaic energy storage and charging management method based on distributed control according to claim 6, characterized in that, The alliance topology evolution module is specifically as follows: Read the node quantile state characteristics and node coupling weights in the potential state of the alliance at the current time step; For any two nodes, calculate the quantile state difference between the quantile state features of the nodes at the current time step, and generate the node association value by combining the corresponding node coupling weights. Construct a node association sequence based on the association values ​​of each node, and generate the node connection relationship for the next time step based on the node association sequence; Compare the node connection relationships of the next time step with the node connection relationships of the current time step, mark newly formed node connections as new connections, and mark missing node connections as disconnected connections. Update the potential state sequence of the alliance based on the node connection relationship of the next time step, and write the newly added connection marker, disconnection marker and corresponding time step information into the potential state sequence of the alliance. The potential state sequence of the alliance at each time step is continuously updated in chronological order to form the evolution sequence of the alliance topology.

8. The photovoltaic energy storage and charging management method based on distributed control according to claim 6, characterized in that, The specific antifactual alliance split module is as follows: Read the potential state sequence of the alliance corresponding to the current time step, and extract the quantile state features and node coupling weights of each node in the alliance. A node association sequence is constructed based on the node quantile state characteristics and node coupling weights. The node association sequence records the association relationships between each node. Multiple node connection relationship combinations are generated according to the node association sequence, and each node connection relationship combination corresponds to a kind of alliance structure branch. Each alliance structure branch is mapped to an independent potential state space, and continuous time step trajectory unfolding is performed according to the alliance potential state transition process; Record the changes in node position status and node connection relationships of each alliance structure branch during the trajectory unfolding process to form a counterfactual trajectory set.

9. The photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step six specifically involves: Read the node quantile status change information and node connection relationship change information corresponding to each counterfactual trajectory in the pre-simulated trajectory set; The information on changes in node quantile states and changes in node connection relationships are correlated and arranged in chronological order to construct a coalition state evolution sequence. The number of times the node member state changes and the number of times the node connection state changes within each time step in the alliance state evolution sequence are statistically analyzed, and the cumulative change values ​​are used to form the alliance evolution characteristic sequence. The stability of each energy alliance candidate structure is calculated based on the alliance evolution characteristic sequence, and an alliance ranking sequence is formed according to the alliance stability. Read the candidate structure of the energy alliance with the highest sorting position in the alliance sorting sequence to determine the target energy alliance; Extract the node connection relationships and node position status information corresponding to the target energy alliance, and generate the power allocation relationship between nodes; Based on the node allocation status and inter-node connectivity of the target energy alliance, the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the load power of charging nodes are allocated to form a power dispatch result.

10. A photovoltaic energy storage and charging management method based on distributed control according to claim 1, characterized in that, Step seven specifically involves: Collect the output power of photovoltaic nodes, the charging and discharging power of energy storage nodes, and the actual load power of charging nodes to form an actual execution power data sequence; The actual execution power data sequence is compared with the power scheduling results according to the time series, and the power deviation value of each node is calculated. Based on the power deviation values ​​of each node, the output power of the photovoltaic node, the charging and discharging power of the energy storage node, and the load power of the charging node are corrected for consistency, and the corrected power dispatch results are generated. Distributed control commands are generated based on the corrected power scheduling results and sent to photovoltaic nodes, energy storage nodes, and charging nodes respectively to achieve dynamic execution of energy management.

Images