Multi-objective optimization based integrated charging and storage intelligent scheduling method and system

Abstract

The application provides a kind of based on multi-objective optimization's filling storage integrated intelligent scheduling method and system, it is related to power system scheduling technical field, including acquisition operation data, dispatching scene is divided into multiple microgrid units and is distributed virtual energy token, fills and discharge behavior is recorded through blockchain;Build charging scene characteristic spectrum and establish scene probability distribution matrix;Real-time scheduling decision is made using distributed collaborative game method, and finally determine power allocation scheme.The application realizes the automation of distributed scheduling collaboration, improves system operation efficiency, and reduces energy cost.

Description

Technical Field

[0001] This invention relates to the field of power system dispatching technology, and in particular to an integrated intelligent dispatching method and system for charging and energy storage based on multi-objective optimization. Background Technology

[0002] With the rapid development of new energy and electric vehicles, the integrated deployment of charging infrastructure and energy storage systems has become an important way to optimize energy utilization and improve system flexibility. Integrated charging and storage systems, by organically combining charging facilities with energy storage devices, can smooth out fluctuations in new energy power generation, peak shaving and valley filling, participate in grid frequency regulation and peak shaving services, and provide electric vehicles with flexible and diverse charging options. This system utilizes energy storage devices to store electrical energy during off-peak hours and release it during peak hours, which can reduce charging costs, provide ancillary services to the grid, and generate additional revenue.

[0003] In the scheduling and management of traditional integrated charging and energy storage systems, a centralized scheduling strategy is often adopted, using a unified energy management system to schedule and control charging piles and energy storage equipment. With the development of emerging technologies such as blockchain and artificial intelligence, the scheduling methods of integrated charging and energy storage systems are also constantly being innovated, such as intelligent scheduling algorithms based on deep reinforcement learning and robust optimization methods that consider uncertainties, in order to improve the scheduling efficiency and economy of the system.

[0004] However, existing integrated charging and energy storage scheduling technologies still have shortcomings and deficiencies. Existing scheduling methods lack sufficient analysis of charging load characteristics, often relying on simple load forecasting models that fail to fully consider the randomness and volatility of charging behavior. This results in insufficient accuracy and adaptability of scheduling decisions, making it difficult to cope with complex and ever-changing charging demands. Traditional centralized scheduling architectures lack flexibility and scalability. As the scale of charging facilities expands, scheduling complexity increases dramatically, system response speed decreases, and there is a risk of single-point failure, making it difficult to adapt to the development needs of distributed energy systems. Existing scheduling methods are insufficient in multi-objective optimization, often focusing on economic benefits while neglecting comprehensive considerations such as system stability and user experience. They lack effective benefit balancing mechanisms, making it difficult to achieve collaborative win-win results among all participants and limiting the maximization of the social benefits of integrated charging and energy storage systems. Summary of the Invention

[0005] This invention provides an integrated intelligent scheduling method and system for charging and storage based on multi-objective optimization, which can solve the problems in the prior art.

[0006] A first aspect of this invention provides an integrated intelligent scheduling method for charging and storage based on multi-objective optimization, comprising:

[0007] Collect historical and real-time data on the integrated operation of charging and storage systems;

[0008] The integrated charging and storage scheduling scenario is divided into multiple microgrid units. Virtual energy tokens are allocated to each microgrid unit. The charging and discharging behavior of each microgrid unit is recorded based on the blockchain distributed ledger to generate energy transaction records.

[0009] Based on historical and real-time data, load characteristic components are obtained through intrinsic mode functions and white noise-assisted sequences. Hierarchical clustering is performed based on density peaks and inter-layer feature transfer. The charging scenario feature spectrum is constructed by fusing time-domain and frequency-domain features.

[0010] Historical scheduling data is obtained from the blockchain distributed ledger, and a scenario probability distribution matrix is ​​established by combining the charging scenario feature spectrum and the energy transaction records.

[0011] The scheduling constraint boundary between microgrid units is determined by the scenario probability distribution matrix. A distributed collaborative game method is used for real-time scheduling decisions. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function. Through iterative updates and revenue compensation, the power allocation scheme of each device in the microgrid unit is determined.

[0012] Based on the power allocation scheme after game convergence, power control instructions are executed, and the scheduling results are recorded in the blockchain distributed ledger.

[0013] In one optional embodiment, the integrated charging and storage scheduling scenario is divided into multiple microgrid units, and the allocation of virtual energy tokens to each microgrid unit includes:

[0014] The voltage, current and phase angle of each node in the integrated charging and storage scheduling scenario are collected, an impedance matrix is ​​constructed, the electrical distance between nodes is calculated based on the impedance matrix, and when the electrical distance is less than a preset electrical distance threshold, the nodes are divided into the same microgrid unit, and a microgrid unit topology matrix containing the node connection relationship is generated.

[0015] Based on the microgrid unit topology matrix, load power data and equipment distribution data within each microgrid unit are obtained, node voltage margin and power balance values ​​are calculated, upper and lower limits of equipment capacity are determined based on the rated power of each type of equipment, equipment adjustment range is calculated, and the sum of power within the equipment adjustment range is determined as the maximum adjustable capacity of each microgrid unit.

[0016] Based on the maximum adjustable capacity, the historical load curve characteristics of each microgrid unit are extracted, the load fluctuation rate and peak shaving response rate are calculated, and the initial allocation coefficient of virtual energy tokens for each microgrid unit is obtained based on the weighted calculation results of the load fluctuation rate and peak shaving response rate.

[0017] Real-time power data of each microgrid unit is collected, power fluctuation variance and peak-valley difference are calculated, adjustment contribution coefficient is determined, and the adjustment contribution coefficient is weighted with the initial allocation coefficient to dynamically update the virtual energy token allocation weight of each microgrid unit.

[0018] In one optional embodiment, generating energy transaction records based on the charging and discharging behavior of each microgrid unit recorded by a blockchain distributed ledger includes:

[0019] It receives charging and discharging transaction data between microgrid units, extracts the identity identifiers and charging and discharging data of the two parties in the transaction, determines whether the charging and discharging transaction conditions meet the constraints according to the preset verification rules, encrypts the charging and discharging transaction data that meets the constraints, and packages it into blocks.

[0020] Based on the charging and discharging transaction data in the block, historical transaction records of each microgrid unit are extracted, the deviation value between the charging and discharging response time and the agreed time, and the deviation ratio between the actual charging and discharging power and the agreed power are calculated. The deviation value and the deviation ratio are weighted and summed to obtain the node score, and normalized to obtain the node weight. The node with the highest weight is selected as the block producing node, and the block producing node broadcasts the charging and discharging block data to the blockchain network.

[0021] The smart contract is triggered, and the amount of charge / discharge and the corresponding amount of virtual energy tokens are calculated according to the contract rules. The virtual energy token account balances of both parties in the transaction are updated, and the charge / discharge transaction confirmation information is generated and written into the blockchain distributed ledger.

[0022] In one optional embodiment, based on historical and real-time data, load characteristic components are obtained through intrinsic mode functions and white noise-assisted sequences. Hierarchical clustering is performed based on density peaks and inter-layer feature transfer. The charging scenario feature spectrum is constructed by fusing time-domain and frequency-domain features, including:

[0023] Historical load data and real-time load data are extracted, and preprocessed and standardized to obtain the load data to be processed;

[0024] Intrinsic mode decomposition is performed on the load data to be processed based on complementary white noise sequence groups. The decomposition results are screened by mode integrity test, and the endpoint effect and mode mixing are optimized. Load feature components are obtained based on frequency feature fusion.

[0025] The local density values ​​of the load characteristic components are calculated, and the cutoff distance is adaptively determined based on the local density values ​​to identify the density peak points. The load characteristic components are combined with energy proportion and frequency distribution to perform hierarchical division. Each layer is optimized through the inter-layer feature transfer mechanism to obtain the hierarchical clustering results.

[0026] Based on the hierarchical clustering results, time-domain feature parameters and frequency-domain feature indices are calculated, and the time-domain feature parameters and frequency-domain feature indices are fused to construct a charging scene feature spectrum. An incremental learning method is used to adaptively adjust the feature weights in the charging scene feature spectrum and dynamically update the charging scene feature spectrum.

[0027] In one optional embodiment, intrinsic mode decomposition is performed on the load data to be processed based on complementary white noise sequence groups. The decomposition results are screened through mode integrity checks, and optimization is performed to address endpoint effects and mode aliasing. The load feature components obtained based on frequency feature fusion include:

[0028] Multiple complementary white noise sequences are constructed. The amplitude range of the complementary white noise sequences is determined based on the root mean square amplitude of the load data to be processed. The spectral distribution of the complementary white noise sequences is determined according to the sampling frequency of the load data to be processed. The number of complementary white noise sequences is determined based on the orthogonality between the complementary white noise sequences, thus obtaining a complementary noise sequence group.

[0029] Complementary noise sequence groups are superimposed onto the load data to be processed to obtain multiple sets of noisy load data. Intrinsic mode decomposition is performed on the noisy load data, and the decomposition rounds are dynamically adjusted based on the decomposition stability index to obtain the initial decomposition results.

[0030] Modal integrity is checked on the initial decomposition results, the degree of residual noise influence of the initial decomposition results is calculated, and decomposition results that do not meet the preset integrity requirements and preset residual noise thresholds are removed to obtain the screened decomposition results;

[0031] Endpoint extension processing and envelope fitting are performed on the screened decomposition results. The frequency separation method is used to identify aliased modes. The aliased modes are eliminated through iteration. The endpoint processing effect evaluation value and mode separation degree are calculated to obtain the optimized decomposition results.

[0032] The fusion weights are determined based on the frequency characteristics of the optimization decomposition results. Multiple sets of optimization decomposition results are then weighted and fused to obtain the load characteristic components.

[0033] In one optional embodiment, the local density values ​​of the load characteristic components are calculated, and the cutoff distance is adaptively determined based on the local density values ​​to identify density peak points. Hierarchical division is performed by combining the energy proportion and frequency distribution of the load characteristic components, and each layer is optimized through an inter-layer feature transfer mechanism to obtain hierarchical clustering results, including:

[0034] Calculate the distance matrix between load feature components, calculate the local density values ​​of the load feature components based on the distance matrix and Gaussian kernel function, determine the initial threshold according to the distribution of the local density values, adaptively optimize the cutoff distance by dynamically adjusting the distance attenuation factor, identify the density peak points based on the cutoff distance, and determine the initial cluster centers;

[0035] Using the initial cluster center as a benchmark, the energy proportion of the load feature components is calculated, the frequency distribution characteristics of the load feature components are extracted, the joint weight is determined based on the energy proportion and the frequency distribution characteristics, the load feature components are ranked according to importance based on the joint weight, and the basis for hierarchical division is determined.

[0036] The load characteristic components are divided into a primary layer and a secondary layer according to the hierarchical division criteria. Intra-layer cluster centers are calculated for the primary layer and the secondary layer respectively. The affiliation relationship of each layer member is updated based on the intra-layer cluster centers. Inter-layer association features are calculated according to the affiliation relationship, and feature correspondence rules from the primary layer to the secondary layer are established.

[0037] Based on the feature correspondence rules, the clustering results of the main layer are passed to the secondary layer, the clustering results of the secondary layer are corrected, the optimization results of the secondary layer are determined, and the optimization results of the secondary layer are fed back to the main layer for cluster center update. After multiple rounds of iteration, the hierarchical clustering results are obtained.

[0038] In one optional embodiment, the scheduling constraint boundary between microgrid units is determined by the scenario probability distribution matrix, and a distributed collaborative game method is used for real-time scheduling decisions. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function, and determines the power allocation scheme of each device within the microgrid unit through iterative updates and revenue compensation, including:

[0039] The power scheduling range of each microgrid unit is determined based on the scenario probability distribution matrix, including the maximum and minimum power values ​​that each microgrid unit can trade, and the scheduling constraint boundary between microgrid units is obtained.

[0040] For each microgrid unit, the revenue of virtual energy tokens is calculated, including revenue from energy trading tokens, scheduling response tokens, and auxiliary service tokens. A cumulative revenue relationship over time is established, and a comprehensive revenue function for virtual energy tokens is constructed. This comprehensive revenue function is then used as the objective function for scheduling optimization.

[0041] Within the scheduling constraint boundary, a distributed cooperative game is performed to determine the initial power allocation value of each microgrid unit. Adjacent microgrid units share their respective power allocation information. Each microgrid unit calculates the optimal response strategy based on the power allocation information received from adjacent units and its own objective function, updates the power allocation value of the microgrid unit, and sends it to the adjacent microgrid units.

[0042] In the distributed collaborative game process, each microgrid unit corrects the power allocation value according to the scheduling constraint boundary, power generation equipment constraint, energy storage equipment constraint and load constraint, calculates the power transaction difference between adjacent microgrid units, sets the compensation coefficient of virtual energy tokens based on the power transaction difference, and adjusts the revenue distribution ratio of each microgrid unit according to the compensation coefficient.

[0043] The distributed collaborative game is repeated. When the difference between the power allocation schemes of adjacent iterations is less than a preset difference threshold, the final power allocation scheme of the devices in each microgrid unit is output.

[0044] A second aspect of this invention provides an integrated intelligent scheduling system for charging and storage based on multi-objective optimization, comprising:

[0045] The first unit is used to collect historical and real-time data on the integrated operation of charging and storage;

[0046] The second unit is used to divide the integrated charging and storage scheduling scenario into multiple microgrid units, allocate virtual energy tokens to each microgrid unit, record the charging and discharging behavior of each microgrid unit based on the blockchain distributed ledger, and generate energy transaction records.

[0047] The third unit is used to obtain load characteristic components based on historical and real-time data through intrinsic mode functions and white noise-assisted sequences, perform hierarchical clustering based on density peaks and inter-layer feature transfer, and construct a charging scenario feature spectrum by fusing time-domain and frequency-domain features.

[0048] The fourth unit is used to obtain historical scheduling data from the blockchain distributed ledger, and to establish a scenario probability distribution matrix by combining the charging scenario feature spectrum and the energy transaction records.

[0049] The fifth unit is used to determine the scheduling constraint boundary between microgrid units based on the scenario probability distribution matrix, and to make real-time scheduling decisions using a distributed collaborative game method. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function, and determines the power allocation scheme of each device within the microgrid unit through iterative updates and revenue compensation.

[0050] The sixth unit is used to execute power control instructions based on the power allocation scheme after game convergence, and record the scheduling results to the blockchain distributed ledger.

[0051] A third aspect of the present invention provides an electronic device, comprising:

[0052] processor;

[0053] Memory used to store processor-executable instructions;

[0054] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0055] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0056] In this embodiment of the invention, a multi-objective optimization-based intelligent scheduling method for integrated charging and storage records the charging and discharging behavior of microgrid units through a blockchain distributed ledger, achieving reliable recording and traceability of energy transactions, ensuring data security and transparency, and effectively solving the single-point failure and data tampering risks of traditional centralized scheduling systems. It combines intrinsic mode function decomposition and white noise-assisted sequences to extract features from load data, achieving accurate scene classification through density peak clustering and inter-layer feature transfer, constructing a charging scene feature spectrum that integrates time-domain and frequency-domain features, improving the ability to identify and predict complex load patterns, and providing reliable data support for scheduling decisions. A distributed collaborative game theory method is used for real-time scheduling decisions, with each microgrid unit making autonomous decisions and collaboratively optimizing based on the revenue function of virtual energy tokens. Through iterative updates and revenue compensation mechanisms, a globally optimal power allocation scheme is achieved, effectively balancing multiple objectives such as system economy, environmental protection, and equipment lifespan, and improving the overall operating efficiency and flexibility of the integrated charging and storage system. Attached Figure Description

[0057] Figure 1 This is a flowchart illustrating the intelligent scheduling method for integrated charging and storage based on multi-objective optimization, as described in an embodiment of the present invention.

[0058] Figure 2 This is a flowchart of the adaptive noise-assisted load decomposition and feature fusion process. Detailed Implementation

[0059] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0060] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0061] Figure 1This is a flowchart illustrating the intelligent scheduling method for integrated charging and storage based on multi-objective optimization according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0062] Collect historical and real-time data on the integrated operation of charging and storage systems;

[0063] The integrated charging and storage scheduling scenario is divided into multiple microgrid units. Virtual energy tokens are allocated to each microgrid unit. The charging and discharging behavior of each microgrid unit is recorded based on the blockchain distributed ledger to generate energy transaction records.

[0064] Based on historical and real-time data, load characteristic components are obtained through intrinsic mode functions and white noise-assisted sequences. Hierarchical clustering is performed based on density peaks and inter-layer feature transfer. The charging scenario feature spectrum is constructed by fusing time-domain and frequency-domain features.

[0065] Historical scheduling data is obtained from the blockchain distributed ledger, and a scenario probability distribution matrix is ​​established by combining the charging scenario feature spectrum and the energy transaction records.

[0066] The scheduling constraint boundary between microgrid units is determined by the scenario probability distribution matrix. A distributed collaborative game method is used for real-time scheduling decisions. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function. Through iterative updates and revenue compensation, the power allocation scheme of each device in the microgrid unit is determined.

[0067] Based on the power allocation scheme after game convergence, power control instructions are executed, and the scheduling results are recorded in the blockchain distributed ledger.

[0068] In one optional implementation, the integrated charging and storage scheduling scenario is divided into multiple microgrid units, and the allocation of virtual energy tokens to each microgrid unit includes:

[0069] The voltage, current and phase angle of each node in the integrated charging and storage scheduling scenario are collected, an impedance matrix is ​​constructed, the electrical distance between nodes is calculated based on the impedance matrix, and when the electrical distance is less than a preset electrical distance threshold, the nodes are divided into the same microgrid unit, and a microgrid unit topology matrix containing the node connection relationship is generated.

[0070] Based on the microgrid unit topology matrix, load power data and equipment distribution data within each microgrid unit are obtained, node voltage margin and power balance values ​​are calculated, upper and lower limits of equipment capacity are determined based on the rated power of each type of equipment, equipment adjustment range is calculated, and the sum of power within the equipment adjustment range is determined as the maximum adjustable capacity of each microgrid unit.

[0071] Based on the maximum adjustable capacity, the historical load curve characteristics of each microgrid unit are extracted, the load fluctuation rate and peak shaving response rate are calculated, and the initial allocation coefficient of virtual energy tokens for each microgrid unit is obtained based on the weighted calculation results of the load fluctuation rate and peak shaving response rate.

[0072] Real-time power data of each microgrid unit is collected, power fluctuation variance and peak-valley difference are calculated, adjustment contribution coefficient is determined, and the adjustment contribution coefficient is weighted with the initial allocation coefficient to dynamically update the virtual energy token allocation weight of each microgrid unit.

[0073] In one specific implementation, voltage, current, and phase angle data of each node in a charging-storage integrated scheduling scenario are collected. For example, in a scheduling scenario with 50 nodes, smart meters or monitoring devices are used to obtain real-time voltage values ​​(e.g., 220V for node 1, phase angle 30°) and current values ​​(e.g., 10A for node 1, phase angle 15°) of each node. Based on the collected voltage and current data, a system impedance matrix is ​​constructed. This matrix reflects the impedance relationship between nodes in the network; for example, the impedance between node 1 and node 2 is 0.5 + j0.3Ω. The electrical distance between nodes is calculated based on the impedance matrix, which can be represented by the magnitude of the impedance value or other electrical characteristic indicators. In this embodiment, an electrical distance threshold of 0.8Ω is set. When the electrical distance between two nodes is less than this threshold, they are divided into the same microgrid unit. By traversing all node pairs, multiple microgrid units are eventually formed, such as dividing 50 nodes into 5 microgrid units, each unit containing 8-12 nodes. Generate a microgrid unit topology matrix. This matrix contains the connection relationships between each node. 0 and 1 represent whether nodes are connected. For example, if node 1 is connected to node 2, the corresponding matrix element is 1.

[0074] Based on the generated microgrid unit topology matrix, load power data and equipment distribution data within each microgrid unit are obtained. For example, microgrid unit 1 contains a total load power of 200kW, including 5 photovoltaic power generation devices (total capacity 100kW), 3 energy storage devices (total capacity 50kWh), and several load points. The voltage margin of each node within the microgrid unit is calculated. For example, if the nominal node voltage is 220V and the actual measured voltage is 218V, with an allowable fluctuation range of ±5%, then the voltage margin is (220×1.05-218) / 220=0.032, or 3.2%. Simultaneously, the power balance value is calculated. For example, if the total power generation of microgrid unit 1 is 90kW and the total load is 85kW, the power balance value is +5kW. Based on equipment parameters, the upper and lower limits of the capacity for each type of equipment are determined. For example, if the rated power of a photovoltaic inverter is 20kW, considering a safety margin, its upper limit is set to 19kW, and its lower limit is 0kW. The energy storage device has a rated power of 10kW, a power range of -10kW to 0 during charging, and 0 to 10kW during discharging. By summarizing the adjustment ranges of all adjustable devices, the maximum adjustable capacity of microgrid unit 1 is calculated to be 75kW.

[0075] Extract the historical load curve characteristics of each microgrid unit and statistically analyze the load data of the past 30 days. Calculate the load fluctuation rate. Taking microgrid unit 1 as an example, its maximum daily load is 120kW, minimum load is 40kW, and average load is 80kW. Then the load fluctuation rate is (120-40) / 80=1.0. Calculate the peak-shaving response rate. Assuming that microgrid unit 1 successfully responded 9 out of the past 10 peak-shaving requests, and the average response power reached 85% of the requested power, then its peak-shaving response rate is 0.9×0.85=0.765. Weight the load fluctuation rate and peak-shaving response rate, setting the weights to 0.4 and 0.6 respectively. Then the initial allocation coefficient of microgrid unit 1 is 0.4×1.0+0.6×0.765=0.859. Similarly, calculate the initial allocation coefficients for other microgrid units.

[0076] Real-time power data of each microgrid unit is collected, and power changes are recorded at 15-minute intervals over 24 hours. The power fluctuation variance is calculated; for example, the 24-hour power data variance for microgrid unit 1 is 225 kW. 2 Calculate the peak-to-valley difference. For example, if the maximum daily power of microgrid unit 1 is 110kW and the minimum daily power is 35kW, the peak-to-valley difference is 75kW. Based on the power fluctuation variance and the peak-to-valley difference, determine the regulation contribution coefficient. For example, set the variance benchmark value to 200kW. 2The peak-valley difference baseline value is 60kW. The power fluctuation variance ratio of microgrid unit 1 is 225 / 200=1.125, and the peak-valley difference ratio is 75 / 60=1.25. Taking weights of 0.5 and 0.5 respectively, the adjustment contribution coefficient is 0.5×1.125+0.5×1.25=1.1875. Weighting the adjustment contribution coefficient with the initial allocation coefficient, and setting the weights to 0.7 and 0.3 respectively, the final virtual energy token allocation weight of microgrid unit 1 is 0.7×0.859+0.3×1.1875=0.9571.

[0077] Based on the aforementioned allocation weights, in a pool of 10,000 virtual energy tokens, microgrid unit 1 receives 10,000 × 0.9571 / ∑(allocation weights of all microgrid units). As the microgrid's operating status changes, the system continuously collects real-time data and updates the allocation weights, achieving dynamic allocation of virtual energy tokens and thus promoting the efficient operation of the integrated charging and storage system.

[0078] In one optional implementation, generating energy transaction records based on the charging and discharging behavior of each microgrid unit using a blockchain distributed ledger includes:

[0079] It receives charging and discharging transaction data between microgrid units, extracts the identity identifiers and charging and discharging data of the two parties in the transaction, determines whether the charging and discharging transaction conditions meet the constraints according to the preset verification rules, encrypts the charging and discharging transaction data that meets the constraints, and packages it into blocks.

[0080] Based on the charging and discharging transaction data in the block, historical transaction records of each microgrid unit are extracted, the deviation value between the charging and discharging response time and the agreed time, and the deviation ratio between the actual charging and discharging power and the agreed power are calculated. The deviation value and the deviation ratio are weighted and summed to obtain the node score, and normalized to obtain the node weight. The node with the highest weight is selected as the block producing node, and the block producing node broadcasts the charging and discharging block data to the blockchain network.

[0081] The smart contract is triggered, and the amount of charge / discharge and the corresponding amount of virtual energy tokens are calculated according to the contract rules. The virtual energy token account balances of both parties in the transaction are updated, and the charge / discharge transaction confirmation information is generated and written into the blockchain distributed ledger.

[0082] In one specific implementation, charging and discharging transaction data between microgrid units is received. This data is collected through a dedicated energy management interface and includes, but is not limited to, information such as transaction initiation time, identification of the trading parties, expected charging / discharging start time, expected charging / discharging end time, expected charging / discharging power, actual charging / discharging start time, actual charging / discharging end time, and actual charging / discharging power. For example, microgrid unit A initiates a charging transaction to microgrid unit B. The transaction data may include: transaction initiation time 2023-05-10 09:00:00, microgrid unit A identifier "MG001", microgrid unit B identifier "MG002", expected charging start time 2023-05-10 10:00:00, expected charging end time 2023-05-10 12:00:00, expected charging power 5kW, actual charging start time 2023-05-10 10:05:00, actual charging end time 2023-05-10 12:10:00, and actual charging power 4.8kW.

[0083] The system extracts the identity identifiers and charging / discharging data of both parties from the received transaction data, and determines whether the charging / discharging transaction conditions meet the constraints according to preset verification rules. Verification rules include, but are not limited to: identity validity verification of both parties, power capacity verification, time window verification, and power limit verification. Identity validity verification ensures that both parties are legitimate microgrid units registered in the system; power capacity verification ensures that the power supplier has sufficient power supply capacity and the power consumer has sufficient power receiving capacity; time window verification ensures that the transaction occurs within the system-allowed time range; and power limit verification ensures that the transaction power does not exceed the system-set safety threshold. For example, in the above transaction, the identities of microgrid units A and B are both valid. A's remaining power supply capacity is 10kW (greater than 5kW), B's remaining power receiving capacity is 8kW (greater than 5kW), the transaction time is within the system-allowed range of 8:00-22:00, and the transaction power of 5kW does not exceed the system-set safety threshold of 7kW. Therefore, the transaction meets all constraints.

[0084] For charging and discharging transaction data that meets the constraints, an asymmetric encryption algorithm is used for encryption. Specifically, the transaction data is signed using the private key of the transaction initiator, and the transaction content is encrypted using the public key of the recipient, ensuring the security and immutability of the transaction data. The encrypted transaction data is packaged into blocks, each block containing multiple transaction records, a timestamp, the hash value of the previous block, and the hash value of the current block. For example, multiple charging and discharging transactions that meet the constraints (including the transactions from A to B mentioned above) are packaged into one block, which contains the transaction hash value, the timestamp 2023-05-10 12:15:00, the hash value of the previous block, and the hash value of the current block.

[0085] Based on the charging and discharging transaction data in the block, historical transaction records of each microgrid unit are extracted, and the deviation values ​​of the charging and discharging response time and the agreed time, as well as the deviation ratio of the actual charging and discharging power and the agreed power, are calculated. For the transaction from A to B mentioned above, the charging and discharging response time deviation value is 5 minutes (10:05:00 minus 10:00:00), and the deviation ratio of the actual charging and discharging power and the agreed power is 4% ((5-4.8) / 5×100%). These deviation values ​​and deviation ratios are weighted and summed to calculate the score of each node. For example, assuming the weight of the time deviation is 0.6 and the weight of the power deviation is 0.4, then the score of microgrid unit A is 0.6×5+0.4×4=4.6. The scores of all nodes are normalized to obtain the weight of each node. Assuming there are 10 microgrid units, where unit A has a score of 4.6, and the scores of the other 9 microgrid units are 5.2, 6.1, 3.8, 7.0, 5.5, 4.2, 3.9, 5.8, and 6.5 respectively, the normalized weight of unit A is 4.6 / (4.6+5.2+6.1+3.8+7.0+5.5+4.2+3.9+5.8+6.5)=0.087. The node with the highest weight (in this example, the microgrid unit with a score of 7.0) is selected as the block-producing node, and this block-producing node broadcasts charge and discharge block data to the blockchain network.

[0086] Once a new block is confirmed and added to the blockchain, the smart contract is executed. The smart contract calculates the charging / discharging capacity and the corresponding amount of virtual energy tokens according to the contract rules. For example, for the transaction from A to B mentioned above, the actual charging time is 2 hours and 5 minutes (12:10:00 minus 10:05:00), and the actual charging power is 4.8kW. Therefore, the actual charging capacity is 4.8kW × 2.083h = 10kWh. Assuming 1kWh of energy corresponds to 10 virtual energy tokens, the virtual energy token amount for this transaction is 10kWh × 10 = 100 tokens. The smart contract updates the virtual energy token account balances of both parties based on the calculation results, for example, deducting 100 tokens from A's account and adding 100 tokens to B's account. A charging / discharging transaction confirmation is generated, including a unique transaction identifier, transaction time, parties involved, transaction capacity, number of tokens, and transaction status. This confirmation information is written to the blockchain distributed ledger to ensure the immutability and traceability of the transaction record.

[0087] Through the above method, the method of this embodiment realizes the recording of charging and discharging behavior and energy trading management between microgrid units based on blockchain technology, which improves the security, transparency and efficiency of energy trading and provides effective support for the coordinated operation of distributed energy systems.

[0088] In one optional implementation, based on historical and real-time data, load characteristic components are obtained through intrinsic mode functions and white noise-assisted sequences. Hierarchical clustering is performed based on density peaks and inter-layer feature transfer. The charging scenario feature spectrum is constructed by fusing time-domain and frequency-domain features, including:

[0089] Historical load data and real-time load data are extracted, and preprocessed and standardized to obtain the load data to be processed;

[0090] Intrinsic mode decomposition is performed on the load data to be processed based on complementary white noise sequence groups. The decomposition results are screened by mode integrity test, and the endpoint effect and mode mixing are optimized. Load feature components are obtained based on frequency feature fusion.

[0091] The local density values ​​of the load characteristic components are calculated, and the cutoff distance is adaptively determined based on the local density values ​​to identify the density peak points. The load characteristic components are combined with energy proportion and frequency distribution to perform hierarchical division. Each layer is optimized through the inter-layer feature transfer mechanism to obtain the hierarchical clustering results.

[0092] Based on the hierarchical clustering results, time-domain feature parameters and frequency-domain feature indices are calculated, and the time-domain feature parameters and frequency-domain feature indices are fused to construct a charging scene feature spectrum. An incremental learning method is used to adaptively adjust the feature weights in the charging scene feature spectrum and dynamically update the charging scene feature spectrum.

[0093] In one specific implementation, historical load data and real-time load data of electric vehicle charging stations are acquired. Historical load data includes 15-minute interval sampling data from the past 30 days, while real-time load data consists of 1-minute interval sampling data from the current 24 hours. The collected data undergoes preprocessing, including outlier detection and handling (marking data points exceeding three standard deviations of the mean as outliers and replacing them with the average of adjacent time points); missing value imputation (filling missing data using linear interpolation); and smoothing (using a 5-point moving average to reduce random fluctuations). After preprocessing, the data is standardized using Z-score standardization to convert it into a standard normal distribution with a mean of 0 and a standard deviation of 1, yielding the load data to be processed.

[0094] In the load feature component extraction stage, 10 sets of complementary white noise sequences were constructed. Each sequence had the same length as the load data to be processed, and its amplitude was 0.2 times the standard deviation of the load data. These white noise sequences were then added to the load data to generate 10 sets of noisy data. Intrinsic mode decomposition (IMD) was performed on each set of noisy data. The number of iterations for filtering was set to 100, and the stopping condition was that the mean square error between two consecutive filtering results was less than 0.0001. The decomposition results were filtered through a modal integrity test, and the orthogonality index of each component was calculated. The orthogonality index threshold was set to 0.1, and modal components with orthogonality indices less than the threshold were retained. Endpoint effect optimization was performed on the retained modal components using a mirror extension method, extending the original data by 20% at each end to reduce fluctuations at the endpoints. Modal aliasing optimization used the correlation coefficient method, calculating the correlation coefficient between adjacent modal components and merging adjacent modes with a correlation coefficient greater than 0.8. Finally, load characteristic components are obtained based on frequency feature fusion. The average value of the modal components in the same frequency range in the 10 decomposition results is taken as the final load characteristic components.

[0095] In the hierarchical clustering stage, the local density values ​​of the load characteristic components are calculated. For each load characteristic component, its Euclidean distance to other components is calculated, and the distance distribution is statistically analyzed. The 25th quantile of the distance distribution is taken as the initial cutoff distance. The cutoff distance is adaptively determined based on the local density values. If the initial density value distribution is too concentrated, the cutoff distance is reduced to 0.8 times the original value; if the initial density value distribution is too dispersed, the cutoff distance is increased to 1.2 times the original value. Density peak points are identified, and the points with the top 20% local density values ​​and the top 10% relative density (product of local density and distance) are determined as density peak points. The hierarchical division is performed by combining the energy proportion and frequency distribution of the load characteristic components. The energy proportion is calculated as the ratio of the variance of each characteristic component to the total variance, and the frequency distribution is obtained by Fourier transform to obtain the dominant frequency. According to the energy proportion, the characteristic components are divided into three layers: high, medium, and low. The high-energy layer contains components with an energy proportion greater than 30%, the medium-energy layer contains components with an energy proportion between 10% and 30%, and the low-energy layer contains components with an energy proportion less than 10%. The inter-layer feature transfer mechanism is used to optimize each layer. The mutual information of the components between each layer is calculated. A transfer path is established between components with mutual information greater than 0.5. The clustering results of each layer are optimized by transferring feature values ​​with a weight of 0.3.

[0096] When constructing the charging scenario feature spectrum, based on the hierarchical clustering results, time-domain feature parameters are calculated, including maximum, minimum, average, standard deviation, peak-to-valley ratio (ratio of maximum to minimum), impulse factor (ratio of maximum to root mean square), and margin factor (ratio of maximum to average). Frequency-domain feature indicators are calculated, including the energy proportion of the dominant frequency, frequency band energy distribution, energy spectrum entropy (a measure of uncertainty in frequency domain energy distribution), and spectral centroid (energy-frequency weighted average). The time-domain feature parameters and frequency-domain feature indicators are fused, and principal component analysis is used to extract the main features, retaining the principal components with a contribution rate of 85% to construct the charging scenario feature spectrum. An incremental learning method is used to adaptively adjust the feature weights in the charging scenario feature spectrum. After collecting new 24-hour data, the similarity between the new data and the existing feature spectrum is calculated. When the similarity is less than 0.7, the feature weights are updated with a learning rate of 0.2; when the similarity is greater than 0.7, the feature weights are updated with a learning rate of 0.1, dynamically updating the charging scenario feature spectrum.

[0097] In one optional implementation, intrinsic mode decomposition is performed on the load data to be processed based on complementary white noise sequence groups. The decomposition results are screened through mode integrity checks, and optimization is performed to address endpoint effects and mode aliasing. The load feature components obtained based on frequency feature fusion include:

[0098] Multiple complementary white noise sequences are constructed. The amplitude range of the complementary white noise sequences is determined based on the root mean square amplitude of the load data to be processed. The spectral distribution of the complementary white noise sequences is determined according to the sampling frequency of the load data to be processed. The number of complementary white noise sequences is determined based on the orthogonality between the complementary white noise sequences, thus obtaining a complementary noise sequence group.

[0099] Complementary noise sequence groups are superimposed onto the load data to be processed to obtain multiple sets of noisy load data. Intrinsic mode decomposition is performed on the noisy load data, and the decomposition rounds are dynamically adjusted based on the decomposition stability index to obtain the initial decomposition results.

[0100] Modal integrity is checked on the initial decomposition results, the degree of residual noise influence of the initial decomposition results is calculated, and decomposition results that do not meet the preset integrity requirements and preset residual noise thresholds are removed to obtain the screened decomposition results;

[0101] Endpoint extension processing and envelope fitting are performed on the screened decomposition results. The frequency separation method is used to identify aliased modes. The aliased modes are eliminated through iteration. The endpoint processing effect evaluation value and mode separation degree are calculated to obtain the optimized decomposition results.

[0102] The fusion weights are determined based on the frequency characteristics of the optimization decomposition results. Multiple sets of optimization decomposition results are then weighted and fused to obtain the load characteristic components.

[0103] This invention provides a method for extracting feature components from load data, which achieves effective processing of load data through complementary white noise sequence groups and intrinsic mode decomposition technology.

[0104] In one specific implementation, multiple sets of complementary white noise sequences are constructed. The amplitude range of the complementary white noise sequences is determined based on the root mean square (RMS) of the amplitude of the load data to be processed. Specifically, if the RMS of the amplitude of the load data to be processed is 2.75, the amplitude range of the complementary white noise sequences can be set to 0.02 to 0.05 times the RMS, i.e., 0.055 to 0.1375. The spectral distribution of the complementary white noise sequences is determined according to the sampling frequency of the load data to be processed. For example, for load data with a sampling frequency of 1 Hz, a uniformly distributed white noise sequence with a frequency range of 0-0.5 Hz can be generated. The number of complementary white noise sequences is determined based on the orthogonality between them. This can be controlled below 0.05 by calculating the correlation coefficient between the sequences. In practical applications, 10-20 sets of complementary white noise sequences can be selected to form a complementary noise sequence group.

[0105] The constructed complementary noise sequence groups are superimposed onto the load data to be processed, resulting in multiple sets of noisy load data. Taking 10 sets of complementary white noise sequences as an example, each noise sequence is added to the original load data to obtain 10 sets of noisy load data. Intrinsic mode decomposition (EMD) is performed on the noisy load data, and the decomposition process is executed by controlling the selection function and iteration stopping conditions. During the decomposition process, the number of decomposition rounds is dynamically adjusted based on the decomposition stability index. The similarity between decomposition results of adjacent rounds is used for evaluation. When the similarity exceeds 0.95 and remains stable for three consecutive rounds, it can be determined as the optimal decomposition round. For example, for a certain set of noisy load data, it may be necessary to perform 15 rounds of decomposition to reach a stable state, ultimately obtaining an initial decomposition result containing multiple intrinsic mode functions.

[0106] The initial decomposition results undergo a modal integrity check. Specifically, this check involves calculating the energy entropy and correlation coefficient of each intrinsic mode function (EMF). The correlation coefficient between each EMF and the original signal is calculated, with a threshold of 0.65. EMFs below this value are considered to fail to meet the integrity requirements. Simultaneously, the residual noise impact of the initial decomposition results is calculated, which can be assessed using the energy ratio. A residual noise threshold of 0.2 is set; decomposition results exceeding this value are discarded. In practical applications, if 15 EMFs are initially obtained, after integrity checks and residual noise assessments, 5 EMFs may be discarded, resulting in 10 valid EMFs as the final decomposition results.

[0107] The decomposition results are processed by endpoint extension and envelope fitting. Specifically, a mirror extension method can be used, extending the data length by 20% at each end. For example, for a data sequence of length 1000, 200 points can be extended at each end. Cubic spline interpolation is used for envelope fitting, controlling the fitting error to within 3% of the original signal amplitude. A frequency separation method is used to identify aliased modes. The instantaneous frequency of each intrinsic mode function (IMF) is calculated using Hilbert transform. When the frequency overlap interval of adjacent IMFs exceeds 30%, mode aliasing is identified. Aliased modes are eliminated iteratively, removing the IMF that least conforms to the characteristics of a single mode each time, until the frequency overlap interval of all remaining IMFs is less than 15%. The endpoint processing effect evaluation value is calculated, quantified by the difference in continuity between the endpoints and the overall signal, and controlled within 0.1. The mode separation degree is calculated, quantified by the frequency overlap rate, and controlled below 0.15, ultimately obtaining the optimized decomposition result.

[0108] The fusion weights are determined based on the frequency characteristics of the optimized decomposition results. Specifically, each intrinsic mode function (IMF) is divided into three intervals according to its frequency characteristics: low frequency (0-0.1Hz), mid frequency (0.1-0.3Hz), and high frequency (0.3-0.5Hz), assigned weight coefficients of 0.5, 0.3, and 0.2 respectively. Ten sets of optimized decomposition results are then weighted and fused using an adaptive weight fusion algorithm based on frequency characteristics. For example, for an IMF in a specific frequency interval, if it shows high consistency across multiple decomposition results, it is assigned a higher weight. In practical applications, a weighted average method can be used to fuse the IMFs in each set. For instance, for IMF 1 in the low-frequency interval, the weighted average of IMF 1 from the ten decomposition results can be calculated using a weight coefficient of 0.5. In this way, load characteristic components containing multiple distinct frequency characteristics and good modal separation are ultimately obtained. These characteristic components effectively characterize the essential features of the load data, providing a foundation for subsequent load analysis and forecasting.

[0109] like Figure 2 The diagram shows the flowchart of load decomposition and feature fusion based on adaptive noise assistance.

[0110] In one optional implementation, the local density values ​​of the load characteristic components are calculated, and the cutoff distance is adaptively determined based on the local density values ​​to identify density peak points. Hierarchical division is performed by combining the energy proportion and frequency distribution of the load characteristic components, and each layer is optimized through an inter-layer feature transfer mechanism to obtain hierarchical clustering results, including:

[0111] Calculate the distance matrix between load feature components, calculate the local density values ​​of the load feature components based on the distance matrix and Gaussian kernel function, determine the initial threshold according to the distribution of the local density values, adaptively optimize the cutoff distance by dynamically adjusting the distance attenuation factor, identify the density peak points based on the cutoff distance, and determine the initial cluster centers;

[0112] Using the initial cluster center as a benchmark, the energy proportion of the load feature components is calculated, the frequency distribution characteristics of the load feature components are extracted, the joint weight is determined based on the energy proportion and the frequency distribution characteristics, the load feature components are ranked according to importance based on the joint weight, and the basis for hierarchical division is determined.

[0113] The load characteristic components are divided into a primary layer and a secondary layer according to the hierarchical division criteria. Intra-layer cluster centers are calculated for the primary layer and the secondary layer respectively. The affiliation relationship of each layer member is updated based on the intra-layer cluster centers. Inter-layer association features are calculated according to the affiliation relationship, and feature correspondence rules from the primary layer to the secondary layer are established.

[0114] Based on the feature correspondence rules, the clustering results of the main layer are passed to the secondary layer, the clustering results of the secondary layer are corrected, the optimization results of the secondary layer are determined, and the optimization results of the secondary layer are fed back to the main layer for cluster center update. After multiple rounds of iteration, the hierarchical clustering results are obtained.

[0115] In one specific implementation, a dataset containing multiple load characteristic components is acquired, which includes the time-domain and frequency-domain features of the load curves. For each pair of load characteristic components i and j, the Euclidean distance d(i, j) between them is calculated, and a distance matrix D is constructed. Taking the load data of 10 users in a distribution network as an example, the load curves of each user for 24 hours are extracted to form a dataset containing 10 samples, each sample containing load values ​​at 24 time points. In the calculated distance matrix D, for example, d(1, 2) = 5.6, d(1, 3) = 8.2, etc., the distance relationships between the samples are fully recorded.

[0116] The local density value ρ for each load characteristic component is calculated based on the distance matrix D and the Gaussian kernel function. Specifically, for characteristic component i, its local density value ρ(i) is equal to the sum of the Gaussian kernel function values ​​for all other characteristic components j to i. The Gaussian kernel function takes the form exp(-d(i, j)). 2 / 2σ 2 ), where σ is the scale parameter. In actual calculations, for the above 10 samples, the initial σ is set to 3.0, and the resulting local density values ​​are ρ(1)=2.8, ρ(2)=3.1, ρ(3)=1.7, etc.

[0117] Based on the distribution characteristics of the local density values, an initial threshold τ is determined. The average value ρ_avg and standard deviation ρ_std of all local density values ​​are calculated, and the initial threshold τ is set to ρ_avg + 0.5 × ρ_std. For the example above, ρ_avg = 2.5 and ρ_std = 0.8, therefore the initial threshold τ = 2.9.

[0118] The cutoff distance is adaptively optimized by dynamically adjusting the distance decay factor. Initially, the distance decay factor α = 0.1, and the cutoff distance dc = τ / α. During iteration, the value of α is adjusted based on the number of density peaks identified: if there are too many peaks, the value of α is increased; if there are too few peaks, the value of α is decreased. In this example, the initial α = 0.1, dc = 29, and 5 density peaks were identified, exceeding the expected number of clusters (3). Therefore, α was adjusted to 0.15, and dc = 19.3 was recalculated. At this point, 3 density peaks were identified, reaching the expected number of clusters, and the final cutoff distance dc = 19.3 was determined.

[0119] Based on the final determined cutoff distance dc, density peak points are identified as initial cluster centers. Specifically, for each feature component i, if its local density value ρ(i) is greater than the local density values ​​of all its neighbors whose distance is less than dc, then i is identified as a density peak point. In this example, samples 2, 5, and 8 are identified as density peak points and used as initial cluster centers.

[0120] Using the initial cluster centers as a reference, the energy proportion of each load characteristic component is calculated. Wavelet transform is performed on each characteristic component to obtain the energy distribution of different frequency bands. The proportion of energy in each frequency band to the total energy is calculated to form the energy proportion vector E. In this example, the energy proportion vector E(1) = [0.45, 0.30, 0.15, 0.10] of sample 1 indicates that the energy proportions of the four frequency bands are 45%, 30%, 15%, and 10%, respectively.

[0121] Extract the frequency distribution characteristics of the load characteristic components. Perform a Fourier transform on each characteristic component to analyze its distribution characteristics in the frequency domain, calculate the distribution of the main frequency components, and form a frequency distribution feature vector F. In this example, the frequency distribution feature vector F(1) = [0.5, 0.3, 0.2] of sample 1 represents the weights of the three main frequency components.

[0122] The joint weight W is determined based on the energy proportion E and the frequency distribution feature F. The joint weight is calculated as W = 0.6 × E + 0.4 × F, which means that the energy proportion weight is 0.6 and the frequency distribution feature weight is 0.4. In this example, the joint weight W(1) of sample 1 is W(1) = 0.6 × [0.45, 0.30, 0.15, 0.10] + 0.4 × [0.5, 0.3, 0.2] = [0.47, 0.30, 0.17, 0.06].

[0123] The load feature components are ranked by importance based on their joint weights to determine the basis for hierarchical division. The average value of the joint weights of each feature component is calculated and used as the importance score. In this example, the importance scores of the samples are S(1)=0.25, S(2)=0.32, S(3)=0.18, etc. An importance threshold S_th=0.23 is set, and samples with scores higher than the threshold are assigned to the main layer, while the remaining samples are assigned to the secondary layer.

[0124] The load characteristic components are divided into a primary layer and a secondary layer according to the hierarchical partitioning criteria. In this example, samples 1, 2, 5, 8, and 9 are assigned to the primary layer, and samples 3, 4, 6, 7, and 10 are assigned to the secondary layer. Density peak clustering is applied to both the primary and secondary layers to calculate their respective intra-layer cluster centers. In the primary layer, samples 2 and 8 are determined as cluster centers; in the secondary layer, samples 4 and 7 are determined as cluster centers.

[0125] The affiliation of members in each layer is updated based on the cluster centers within each layer. For each sample in the main layer, the distance to each cluster center is calculated, and it is assigned to the nearest cluster center. In this example, samples 1, 2, and 9 belong to the cluster centered on sample 2, and samples 5 and 8 belong to the cluster centered on sample 8. The same method is used to process samples in the sub-layers.

[0126] Based on the affiliation relationship, the inter-layer association features are calculated, and the feature correspondence rules from the main layer to the sub-layer are established. For each cluster in the main layer, the distribution characteristics of its members in the feature space are analyzed, and representative features are extracted. In this example, the association features of the main layer cluster centered on sample 2 are [0.42, 0.31, 0.27], and the association features of the main layer cluster centered on sample 8 are [0.38, 0.35, 0.27].

[0127] Based on feature correspondence rules, the primary clustering results are passed to the secondary clustering layers to correct the secondary clustering results. The similarity between the secondary samples and the associated features of the primary clusters is calculated. If the similarity is higher than a threshold, the secondary sample is assigned to the corresponding primary cluster. In this example, the similarity between sample 3 and the primary cluster centered on sample 2 is 0.85, which is higher than the threshold of 0.8. Therefore, sample 3 is assigned to that primary cluster.

[0128] The optimization results of the secondary layer are fed back to the primary layer to update the cluster centers. Based on the adjusted affiliation relationships, the centers of each cluster in the primary layer are recalculated. After multiple iterations, the final hierarchical clustering results are obtained: the first cluster contains samples 1, 2, 3, and 9; the second cluster contains samples 5 and 8; and the third cluster contains samples 4, 6, 7, and 10.

[0129] In one optional implementation, the scheduling constraint boundary between microgrid units is determined by the scenario probability distribution matrix, and a distributed collaborative game method is used for real-time scheduling decisions. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function, and determines the power allocation scheme of each device within the microgrid unit through iterative updates and revenue compensation, including:

[0130] The power scheduling range of each microgrid unit is determined based on the scenario probability distribution matrix, including the maximum and minimum power values ​​that each microgrid unit can trade, and the scheduling constraint boundary between microgrid units is obtained.

[0131] For each microgrid unit, the revenue of virtual energy tokens is calculated, including revenue from energy trading tokens, scheduling response tokens, and auxiliary service tokens. A cumulative revenue relationship over time is established, and a comprehensive revenue function for virtual energy tokens is constructed. This comprehensive revenue function is then used as the objective function for scheduling optimization.

[0132] Within the scheduling constraint boundary, a distributed cooperative game is performed to determine the initial power allocation value of each microgrid unit. Adjacent microgrid units share their respective power allocation information. Each microgrid unit calculates the optimal response strategy based on the power allocation information received from adjacent units and its own objective function, updates the power allocation value of the microgrid unit, and sends it to the adjacent microgrid units.

[0133] In the distributed collaborative game process, each microgrid unit corrects the power allocation value according to the scheduling constraint boundary, power generation equipment constraint, energy storage equipment constraint and load constraint, calculates the power transaction difference between adjacent microgrid units, sets the compensation coefficient of virtual energy tokens based on the power transaction difference, and adjusts the revenue distribution ratio of each microgrid unit according to the compensation coefficient.

[0134] The distributed collaborative game is repeated. When the difference between the power allocation schemes of adjacent iterations is less than a preset difference threshold, the final power allocation scheme of the devices in each microgrid unit is output.

[0135] In one specific implementation, the scheduling constraint boundaries between microgrid units are determined based on a scenario probability distribution matrix. Specifically, for each microgrid unit i, a scenario probability distribution matrix Si is established based on historical data. This matrix contains the probability distribution of different time periods t under different weather conditions, load demand, and electricity price levels. For example, during the 14:00-15:00 period on a summer weekday, under sunny conditions, the probability of high load demand and high electricity price is 0.65; under cloudy conditions, the probability of medium load demand and medium electricity price is 0.25; and under overcast conditions, the probability of low load demand and low electricity price is 0.1. Based on this probability distribution matrix, the maximum tradable power value Pi_max(t) and the minimum tradable power value Pi_min(t) of microgrid unit i in time period t are calculated. For example, the maximum tradable power value of microgrid unit 1 in a certain time period is 50kW, and the minimum power value is -30kW (negative values ​​indicate purchased power). These power boundary values ​​constitute the scheduling constraint boundaries between microgrid units.

[0136] A comprehensive revenue function for virtual energy tokens is constructed for each microgrid unit. The virtual energy token revenue consists of three parts: energy trading token revenue RTi(t), dispatch response token revenue RDi(t), and ancillary service token revenue RSi(t). The energy trading token revenue RTi(t) is calculated based on the power trading volume Pi(t) of microgrid unit i in time period t and the energy trading price λT(t). For example, when Pi(t) = 20kW and λT(t) = 0.8 yuan / kWh, RTi(t) = 16 yuan / h. The dispatch response token revenue RDi(t) is determined based on the microgrid unit's responsiveness to system dispatch commands. For example, when the response command deviation is 5%, RDi(t) = 5 yuan / h. The ancillary service token revenue RSi(t) is calculated based on the amount and corresponding price of ancillary services such as frequency regulation and backup provided by the microgrid unit. For example, when providing 10kW frequency regulation service at a price of 0.5 yuan / kWh, RSi(t) = 5 yuan / h. The comprehensive return function Fi(t) is the weighted sum of the three parts of the return, i.e., Fi(t) = w1·RTi(t) + w2·RDi(t) + w3·RSi(t), where w1, w2, and w3 are weight coefficients, set to 0.6, 0.3, and 0.1 respectively.

[0137] Distributed collaborative game theory is performed within the scheduling constraints. At game initialization, each microgrid unit i determines its initial power allocation value Pi(0) based on its own load forecast and generation forecast. For example, microgrid unit 2's initial allocation value is 25kW (a positive value indicates power supply to external systems). Each microgrid unit sends its own power allocation information to its neighboring units; for example, microgrid unit 2 sends its 25kW information to its connected microgrid units 1 and 3. After receiving the power allocation information from its neighboring units, each microgrid unit calculates its optimal response strategy based on its own comprehensive revenue function and updates its power allocation value. For example, after receiving the 25kW information from microgrid unit 2, microgrid unit 1 calculates its optimal power allocation value as -15kW (indicating the purchase of 15kW of electricity from microgrid 2). The updated power allocation value is then sent to neighboring microgrid units again, entering the next round of the game.

[0138] During the game, each microgrid unit needs to adjust the power allocation value according to the constraints. These constraints include: scheduling boundary constraints (such as Pi_max(t) and Pi_min(t) mentioned above), power generation equipment constraints (such as the maximum output of photovoltaic power generation equipment being 30kW), energy storage equipment constraints (such as the state of charge (SOC) of energy storage equipment must be maintained between 20% and 85%, and the charge / discharge rate must not exceed 0.5C), and load constraints (such as critical loads must be 100% satisfied, and adjustable loads can be adjusted within the range of 80%-100% of rated power). When the calculated power allocation value exceeds the constraints, the system adjusts it to the nearest feasible value.

[0139] To address the imbalance in power trading between adjacent microgrid units, a power trading difference compensation mechanism is introduced. The power trading difference between adjacent microgrid units i and j is calculated as ΔPij(t) = Pi→j(t) - Pj→i(t), where Pi→j(t) represents the power transmitted from microgrid unit i to j. For example, if microgrid unit 1 plans to purchase 15kW of electricity from microgrid unit 2, while microgrid unit 2 plans to supply 25kW of electricity to microgrid unit 1, then ΔP12(t) = 10kW. Based on this difference, a compensation coefficient αij(t) = f(ΔPij(t)) is set for the virtual energy token. For example, when ΔPij(t) = 10kW, αij(t) = 0.85. The revenue distribution ratio of each microgrid unit is adjusted according to the compensation coefficient, ensuring that the microgrid unit with a larger power trading volume receives more revenue compensation.

[0140] The distributed collaborative game is repeatedly executed. When the difference |Pi(k+1)-Pi(k)| between the power allocation schemes of adjacent iterations k and k+1 is less than a preset difference threshold ε (e.g., ε=0.5kW), the game converges, and the final power allocation scheme of the devices in each microgrid unit is output. The final scheme specifies in detail the power values ​​of each power generation device, energy storage device, and adjustable load in each microgrid unit. For example, in microgrid unit 1, the photovoltaic power generation is 25kW, the energy storage device discharge power is 10kW, the flexible load adjustment is reduced by 5kW, and the total external output power is 15kW.

[0141] The method in this embodiment enables coordinated and optimized operation of each microgrid unit in a microgrid cluster, improving energy utilization efficiency, reducing operating costs, and enhancing system stability and flexibility.

[0142] The integrated intelligent scheduling system for charging and storage based on multi-objective optimization according to embodiments of the present invention includes:

[0143] The first unit is used to collect historical and real-time data on the integrated operation of charging and storage;

[0144] The second unit is used to divide the integrated charging and storage scheduling scenario into multiple microgrid units, allocate virtual energy tokens to each microgrid unit, record the charging and discharging behavior of each microgrid unit based on the blockchain distributed ledger, and generate energy transaction records.

[0145] The third unit is used to obtain load characteristic components based on historical and real-time data through intrinsic mode functions and white noise-assisted sequences, perform hierarchical clustering based on density peaks and inter-layer feature transfer, and construct a charging scenario feature spectrum by fusing time-domain and frequency-domain features.

[0146] The fourth unit is used to obtain historical scheduling data from the blockchain distributed ledger, and to establish a scenario probability distribution matrix by combining the charging scenario feature spectrum and the energy transaction records.

[0147] The fifth unit is used to determine the scheduling constraint boundary between microgrid units based on the scenario probability distribution matrix, and to make real-time scheduling decisions using a distributed collaborative game method. Each microgrid unit uses the comprehensive revenue function of virtual energy tokens as the objective function, and determines the power allocation scheme of each device within the microgrid unit through iterative updates and revenue compensation.

[0148] The sixth unit is used to execute power control instructions based on the power allocation scheme after game convergence, and record the scheduling results to the blockchain distributed ledger.

[0149] A third aspect of the present invention provides an electronic device, comprising:

[0150] processor;

[0151] Memory used to store processor-executable instructions;

[0152] The processor is configured to invoke instructions stored in the memory to execute the aforementioned method.

[0153] A fourth aspect of the present invention provides a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, implement the aforementioned method.

[0154] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A multi-objective optimization-based integrated charging and storage intelligent scheduling method, characterized in that, The method comprises the following steps: Collecting historical data and real-time data of integrated charging and storage operation; Dividing the integrated charging and storage scheduling scene into multiple micro-grid units, assigning virtual energy tokens to each micro-grid unit, recording the charging and discharging behavior of each micro-grid unit based on a blockchain distributed ledger, and generating energy transaction records; Based on the historical data and real-time data, the load characteristic components are obtained through the intrinsic mode function and white noise auxiliary sequence, the hierarchical clustering is carried out based on the density peak value and interlayer feature transmission, and the charging scene feature spectrum is constructed by fusing time domain and frequency domain features; Obtaining historical scheduling data from the blockchain distributed ledger, combining the charging scene feature spectrum and the energy transaction records, and establishing a scene probability distribution matrix; Determine the scheduling constraint boundary between micro-grid units based on the scene probability distribution matrix, use a distributed collaborative game method for real-time scheduling decision, and take the comprehensive income function of virtual energy tokens as the objective function of each micro-grid unit, and determine the power allocation scheme of each device in the micro-grid unit through iterative update and income compensation; According to the power allocation scheme after the game converges, execute the power control instruction, and record the scheduling result to the blockchain distributed ledger.

2. The method of claim 1, wherein, The method for dividing the integrated charging and storage scheduling scene into multiple micro-grid units and assigning virtual energy tokens to each micro-grid unit comprises the following steps: Collecting the voltage, current and phase angle of each node in the integrated charging and storage scheduling scene, constructing an impedance matrix, calculating the electrical distance between nodes according to the impedance matrix, and dividing the nodes into the same micro-grid unit when the electrical distance is less than a preset electrical distance threshold, and generating a micro-grid unit topology structure matrix containing node connection relationship; According to the micro-grid unit topology structure matrix, obtain the load power data and device distribution data in each micro-grid unit, calculate the node voltage margin and power balance value, determine the upper and lower limits of device capacity according to the rated power of each type of device, calculate the device adjustment range, and determine the sum of power in the device adjustment range as the maximum adjustable capacity of each micro-grid unit; According to the maximum adjustable capacity, extract the historical load curve characteristics of each micro-grid unit, calculate the load fluctuation rate and peak regulation response rate, and obtain the initial allocation coefficient of the virtual energy tokens of each micro-grid unit according to the weighted calculation result of the load fluctuation rate and the peak regulation response rate; Collecting real-time power data of each micro-grid unit, calculating power fluctuation variance and peak-valley difference, determining adjustment contribution coefficient, and weighting the adjustment contribution coefficient and the initial allocation coefficient to dynamically update the virtual energy token allocation weight of each micro-grid unit.

3. The method of claim 1, wherein, Recording the charging and discharging behavior of each micro-grid unit based on the blockchain distributed ledger, and generating energy transaction records comprises the following steps: Receive the charging and discharging transaction data between each micro-grid unit, extract the identity and charging and discharging data of both parties, judge whether the charging and discharging transaction condition meets the constraint according to the preset verification rule, encrypt the charging and discharging transaction data that meets the constraint, and pack it into a block. According to the charging and discharging transaction data in the block, the historical transaction records of each micro-grid unit are extracted, the deviation value of the charging and discharging response time and the agreed time, and the deviation proportion of the actual charging and discharging power and the agreed power are calculated, the deviation value and the deviation proportion are weighted and summed to obtain a node score, and the node weight is obtained by normalization processing, and the node with the highest weight is selected as the block node, and the charging and discharging block data is broadcasted to the blockchain network by the block node; Triggering the smart contract, calculating the charging and discharging electric quantity and the corresponding virtual energy token amount according to the contract rules, updating the virtual energy token account balance of the transaction parties, generating charging and discharging transaction confirmation information and writing into the blockchain distributed ledger.

4. The method of claim 1, wherein, Based on historical data and real-time data, load characteristic components are obtained through intrinsic mode function and white noise auxiliary sequence, hierarchical clustering is performed based on density peak value and inter-layer feature transmission, and charging scene feature spectrum is constructed by fusing time domain and frequency domain features including: Extracting historical load data and real-time load data, and performing preprocessing and standardization processing to obtain the load data to be processed; Based on the complementary white noise sequence group, the load data to be processed is decomposed by intrinsic mode, the decomposition results are screened through modal integrity test, the end effect and modal aliasing are optimized, and the load characteristic components are obtained based on frequency feature fusion; The local density value of the load characteristic component is calculated, the local density value is adaptively determined to identify the density peak point, the hierarchical division is performed in combination with the energy proportion and frequency distribution of the load characteristic component, the layers are optimized through the inter-layer feature transmission mechanism, and the hierarchical clustering result is obtained; Based on the hierarchical clustering result, time domain feature parameters and frequency domain feature indexes are calculated, the time domain feature parameters and the frequency domain feature indexes are fused, and a charging scene feature spectrum is constructed; an incremental learning method is used to adaptively adjust the feature weight in the charging scene feature spectrum, and the charging scene feature spectrum is dynamically updated.

5. The method of claim 4, wherein, Based on the complementary white noise sequence group, the load data to be processed is decomposed by intrinsic mode, the decomposition results are screened through modal integrity test, the end effect and modal aliasing are optimized, and the load characteristic components are obtained based on frequency feature fusion including: A plurality of complementary white noise sequences are constructed, the amplitude range of the complementary white noise sequence is determined based on the root mean square of the amplitude of the load data to be processed, the frequency spectrum distribution of the complementary white noise sequence is determined according to the sampling frequency of the load data to be processed, and the number of complementary white noise sequences is determined based on the orthogonality between the complementary white noise sequences, thereby obtaining a complementary noise sequence group; The complementary noise sequence group is superimposed on the load data to be processed respectively to obtain a plurality of noisy load data, the noisy load data is decomposed by intrinsic mode, and the decomposition round is dynamically adjusted based on the decomposition stability index to obtain an initial decomposition result; The initial decomposition result is subjected to modal integrity test, the residual noise influence degree of the initial decomposition result is calculated, the decomposition results that do not meet the preset integrity requirement and the preset residual noise threshold are removed, and a screened decomposition result is obtained. The endpoint extension processing and envelope fitting are performed on the screening decomposition result, the frequency separation method is used to identify the mixed modal, the mixed modal is removed through iteration, and the endpoint processing effect evaluation value and the modal separation degree are calculated to obtain the optimized decomposition result; The fusion weight is determined based on the frequency characteristics of the optimized decomposition result, and the multiple sets of optimized decomposition results are weighted and fused to obtain the load characteristic component.

6. The method of claim 4, wherein, The local density value of the load characteristic component is calculated, the density peak value point is adaptively determined based on the local density value, the energy proportion and frequency distribution of the load characteristic component are combined for hierarchical division, each layer is optimized through the interlayer feature transmission mechanism to obtain the hierarchical clustering result including: The distance matrix between the load characteristic components is calculated, the local density value of the load characteristic component is calculated based on the distance matrix and the Gaussian kernel function, the initial threshold is determined according to the distribution of the local density value, the distance attenuation factor is dynamically adjusted to adaptively optimize the cutoff distance, the density peak value point is identified based on the cutoff distance, and the initial clustering center is determined; The energy proportion of the load characteristic component is calculated with the initial clustering center as the reference, the frequency distribution characteristics of the load characteristic component are extracted, the joint weight is determined based on the energy proportion and the frequency distribution characteristics, the importance of the load characteristic component is sorted according to the joint weight, and the hierarchical division basis is determined; The load characteristic component is divided into a main layer and a secondary layer according to the hierarchical division basis, the intra-layer clustering center is calculated for the main layer and the secondary layer, and the membership relationship of each layer member is updated based on the intra-layer clustering center; the interlayer correlation feature is calculated according to the membership relationship, and the feature correspondence rule from the main layer to the secondary layer is established; The main layer clustering result is transmitted to the secondary layer based on the feature correspondence rule, the secondary layer clustering result is corrected, the secondary layer optimization result is determined, the secondary layer optimization result is fed back to the main layer for clustering center updating, and after multiple iterations, the hierarchical clustering result is obtained.

7. The method of claim 1, wherein, The scheduling constraint boundary between the microgrid units is determined based on the scenario probability distribution matrix, the real-time scheduling decision is made by using the distributed cooperative game method, each microgrid unit takes the comprehensive income function of the virtual energy token as the objective function, and the power allocation scheme of each device in the microgrid unit is determined through iterative updating and income compensation including: The power scheduling range of each microgrid unit is determined based on the scenario probability distribution matrix, including the maximum power value and the minimum power value that can be traded by each microgrid unit, to obtain the scheduling constraint boundary between the microgrid units; The income of the virtual energy token is calculated for each microgrid unit, including the energy transaction token income, the scheduling response token income and the auxiliary service token income, the time dimension income accumulation relationship is established, and the comprehensive income function of the virtual energy token is constructed, which is taken as the objective function of the scheduling optimization; The distributed cooperative game is executed within the scheduling constraint boundary to determine the initial power allocation value of each microgrid unit, the power allocation information of each microgrid unit is shared with adjacent microgrid units, each microgrid unit calculates the optimal response strategy according to the received power allocation information of adjacent units and its own objective function, updates the power allocation value of the microgrid unit, and sends it to adjacent microgrid units; In the distributed cooperative game process, each micro-grid unit corrects the power distribution value according to the scheduling constraint boundary, the power generation equipment constraint condition, the energy storage equipment constraint condition and the load constraint condition, calculates the power transaction difference between adjacent micro-grid units, sets the compensation coefficient of the virtual energy token based on the power transaction difference, and adjusts the income distribution proportion of each micro-grid unit according to the compensation coefficient; The distributed cooperative game is repeatedly executed, and when the difference between the power distribution schemes of adjacent iteration rounds is less than a preset difference threshold, the final power distribution scheme of the equipment in each micro-grid unit is output.

8. A charging and storage integrated intelligent dispatching system based on multi-objective optimization, for implementing the method of any one of the preceding claims 1-7, characterized in that, Comprise: A first unit for collecting historical data and real-time data of integrated charging and storage operation; A second unit for dividing the integrated charging and storage scheduling scene into a plurality of micro-grid units, allocating virtual energy tokens to each micro-grid unit, recording the charging and discharging behavior of each micro-grid unit based on a blockchain distributed ledger, and generating an energy transaction record; A third unit for obtaining load feature components based on historical data and real-time data through an intrinsic mode function and a white noise auxiliary sequence, performing hierarchical clustering based on density peaks and inter-layer feature transmission, and constructing a charging scene feature spectrum by fusing time domain and frequency domain features; A fourth unit for obtaining historical scheduling data from the blockchain distributed ledger, combining the charging scene feature spectrum and the energy transaction record, and establishing a scene probability distribution matrix; A fifth unit for determining the scheduling constraint boundary between micro-grid units based on the scene probability distribution matrix, making real-time scheduling decisions using a distributed cooperative game method, taking the comprehensive income function of virtual energy tokens as the objective function, and determining the power distribution scheme of each device in the micro-grid unit through iterative updating and income compensation; A sixth unit for executing power control instructions according to the power distribution scheme after game convergence, and recording the scheduling result to the blockchain distributed ledger.

9. An electronic device, comprising: Comprise: A processor; A memory for storing processor-executable instructions; Wherein the processor is configured to call the instructions stored in the memory to execute the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon computer program instructions, wherein, The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 7. The computer program instructions are executed by the processor to implement the method of any one of claims 1 to 7.

Images