Power distribution network distributed energy scheduling method based on hybrid uncertainty modeling

Abstract

The invention discloses a power distribution network distributed energy scheduling method based on hybrid uncertainty modeling, is used for solving the technical problems that traditional power distribution network scheduling depends on empirical meaning, and operation control is complex due to randomness and intermittency of distributed energy output and load requirements, and belongs to the technical field of distributed energy and power distribution networks. According to the method, a three-layer architecture of data perception, optimization decision and collaborative execution is constructed, and the method comprises the following steps: step 1, a data perception layer; acquiring data in real time, obtaining a standardized feature matrix through a streaming processing engine and a feature extraction module, and transmitting a result to an optimization decision-making layer; 2, optimizing a decision-making layer; s21, generating five types of typical scenes through a hybrid model, wherein the hybrid model uses a multi-source uncertain hybrid modeling technology; s22, constructing a multi-objective-hierarchical distributed optimization framework; 3, a collaborative execution layer; s31, constructing a cross-region mutual aid-reconstruction linkage mechanism; and S32, the demand side responds.

Description

Technical Field

[0001] This invention belongs to the field of distributed energy and distribution network technology, and specifically relates to a distributed energy dispatching method for distribution networks based on hybrid uncertainty modeling. Background Technology

[0002] As the core hub connecting centralized power generation with diverse loads, the power distribution network is facing the challenge of energy structure transformation. With the advancement of the "dual carbon" target, the penetration rate of distributed energy resources (DER) such as wind power and photovoltaics in the power distribution network has exceeded 30%. Although the application of big data technology in the power sector has achieved massive data collection, its value has not yet been fully realized in the dispatching and decision-making process.

[0003] Existing big data applications are mostly limited to data collection and lack deep integration with uncertainty modeling and optimization decision-making. Traditional dispatching relies on fixed operating modes and cannot cope with the randomness of DER output (such as the anti-peak-shaving characteristics of photovoltaics), resulting in annual wind and solar curtailment exceeding 10 billion kWh in some areas. Cross-regional dispatching mainly focuses on the main grid level and lacks a mutual assistance mechanism between distribution networks and DER. When local absorption capacity is insufficient, the incidence of wind and solar curtailment at the distribution network level is as high as 12.7%, which is seriously inconsistent with the requirement of full DER absorption under the dual-carbon target. Conservative dispatching reserves excessive reserves to cope with uncertainties, increasing system costs; aggressive dispatching, on the other hand, causes voltage over-limit accidents (the voltage qualification rate of traditional methods is only 90%). Summary of the Invention

[0004] The purpose of this invention is to provide a distributed energy dispatching method for distribution networks based on mixed uncertainty modeling, which addresses the technical problems of traditional distribution network dispatching relying on empiricism, the complexity of operation and control due to the randomness and intermittency of distributed energy output and load demand, the easy occurrence of wind and solar curtailment due to cross-regional consumption, and the lack of transmission and distribution coordination mechanisms and low solution efficiency, which make it difficult to meet the technical problems of real-time dispatching.

[0005] This invention is achieved using the following technical solution:

[0006] A distributed energy dispatching method for power distribution networks based on hybrid uncertainty modeling constructs a three-layer architecture of "data perception-optimization decision-cooperative execution," forming a closed-loop control system through bidirectional data flow, including the following steps:

[0007] Step 1: Data Awareness Layer;

[0008] Real-time data acquisition is processed by a streaming engine and feature extraction module to obtain a standardized feature matrix, and the results are transmitted to the optimization decision layer.

[0009] Step 2: Optimize the decision-making level;

[0010] S21. Five typical scenarios are generated through a hybrid model, which uses a multi-source uncertain hybrid modeling method.

[0011] S22. Construct a multi-objective, hierarchical, distributed optimization framework;

[0012] After optimization by a multi-objective, hierarchical distributed optimization framework, the output of the hybrid model calculates the scheduling instruction set and the transmission network's direction to the first... Electricity supplied by individual power purchasers To the collaborative execution layer;

[0013] Step 3: Collaborative Execution Layer;

[0014] S31. Constructing cross-regional mutual assistance and reconstructing linkage mechanisms

[0015] Design a mutual aid algorithm based on Nash negotiation, and design a real-time response strategy by reconstructing the linkage mechanism;

[0016] S32, Demand-side response;

[0017] The system issues electricity price signals to guide users to adjust their load and aggregates interruptible loads to participate in system balancing.

[0018] Streaming engine: Cleans, aligns, and timestamps heterogeneous data from multiple sources; Feature extraction module: Dynamically calculates features such as the fluctuation rate of new energy output and the difference between peak and valley loads.

[0019] Nash negotiation is a solution concept for cooperative games, aiming to fairly distribute the cooperative surplus so that all participants obtain better payoffs than when they do not cooperate, and the distribution results satisfy axioms such as Pareto optimality, symmetry, and linear transformation invariance.

[0020] Based on the mutual aid algorithm and Nash negotiation, this system quantifies the incremental benefits and cost savings of each region participating in mutual aid, aiming to maximize the overall cooperative surplus. It determines the optimal power trading direction and transmission power, providing a revenue-oriented decision-making basis for reconfiguration actions. Cross-regional mutual aid involves calculating the optimal trading volume based on Nash negotiation and generating bilateral contracts and settlement prices; real-time detection of line power exceeding limits and triggering of tie-line switches during reconfiguration linkage; and real-time response strategies including issuing electricity price signals to guide users to adjust loads and aggregating interruptible loads to participate in system balancing.

[0021] The data forward propagation path of this method is: data perception layer → optimization decision layer → collaborative execution layer (scheduling instruction flow).

[0022] More preferably, the hybrid model in S21 integrates the interval method and the scenario analysis method to generate five typical scenarios; the five typical scenarios are the basic prediction scenario, the uphill scenario, the downhill scenario, the lower limit of output scenario, and the upper limit of output scenario.

[0023] S211. Receive the real-time data collected after the processing in step one;

[0024] S212, Category 5 typical scenarios are represented as follows:

[0025]

[0026]

[0027] parameter:

[0028] Output power of the wind power generation system at time t

[0029] Wind power at time t Mathematical expectation

[0030] Scene number, corresponding to the basic prediction scene.

[0031] Scene number, corresponding to the uphill climbing scene.

[0032] Scene number, corresponding to the downhill climbing scene.

[0033] Scene number, corresponding to the scene with the lower limit of output.

[0034] Scene number, corresponding to the scene with the maximum output.

[0035] 99th percentile of historical wind power ramp-up rate

[0036] The steepest drop in wind power measured during typhoon season

[0037] Time in scene construction

[0038] Lower limit of wind power output

[0039] Upper limit of wind power output

[0040] S213, Construction of Uncertainty Sets:

[0041]

[0042] parameter:

[0043] Wind power capacity lower limit

[0044] Wind power capacity limit

[0045] Wind power ramp-up variation

[0046] Lower limit of climbing constraint

[0047] Upper limit of ramp constraint

[0048] in Indicates the output range. This indicates a climbing constraint.

[0049] Comparative verification was conducted using measured data from a 100MW wind farm in Shanxi Province in 2024 (15-minute intervals, 35,040 data points). Traditional robust optimization achieved a wind curtailment rate of 8.5%, while this model achieved 3.9%. Existing technologies require stochastic programming to handle over 1000 scenarios, taking over 300 seconds. This model uses only 5 typical scenarios, clearing voltage over-limit incidents and improving solution time by 40%.

[0050] Further preferred, S22 specifically involves establishing a three-level optimization model for the transmission network, distribution network, and local resources;

[0051] The power transmission network layer is represented as follows:

[0052] ;

[0053] parameter:

[0054] Total number of time periods within the scheduling period

[0055] The current scheduling period

[0056] Total number of generator units participating in dispatching in the power transmission network

[0057] : The generator set number currently participating in the dispatch

[0058] : No. Unit power generation cost coefficient per generator set, unit: yuan / MWh

[0059] : No. The generator set is Power generation during a given time period, in MW

[0060] Cost coefficient for unit reserve, unit: yuan / MW

[0061] : power transmission network Reserve capacity for a given period, in MW

[0062] The electricity price for the kth electricity purchaser, in yuan / MWh.

[0063] : power transmission network The amount of electricity delivered to the k-th electricity purchaser during the specified time period, in MWh.

[0064] in Indicates the cost of electricity generation. Indicates the cost of reserve. This indicates revenue from electricity sales;

[0065] The following self-regulatory constraints are applied to the distribution network layer:

[0066]

[0067] Key threshold:

[0068]

[0069] The DER self-use rate of the local resource layer is calculated as follows:

[0070]

[0071] parameter:

[0072] Active power interacting between the distribution network and the main grid at time t, unit: MW

[0073] Active power exchanged between the distribution network and the main grid at time t-1, unit: MW

[0074] Maximum permissible rate of change of power between the distribution network and the main grid, unit: MW / min

[0075] Time interval

[0076] : The first in the distribution network at time t Voltage amplitude at each node

[0077] DER self-use rate

[0078] The active power available from distributed energy sources in the distribution network at time t, in MW.

[0079] Active power of the energy storage system at time t, unit: MW

[0080] Active power of demand response at time t, unit: MW

[0081] Total number of time periods in the time interval

[0082] A hierarchical solution using a cascaded objective method is employed: the transmission network layer minimizes generation and reserve costs, while the distribution network layer imposes self-regulatory constraints: power variation rate ≤ 5 MW / min, voltage 0.95-1.05 pu, and local resource layers adjust output in response to electricity price signals. After implementation in Suzhou Industrial Park, the peak-to-valley power difference was reduced from 39.5% to 20.8%, and the main grid frequency regulation cost decreased by 25%. The voltage qualification rate increased from 90% to 98%, and the DER self-use rate reached 87%, a 32 percentage point improvement compared to traditional dispatching.

[0083] like Figure 6 As shown, the hierarchical solution process of the objective cascade method begins with the initialization of boundary variable settings, initial transmission limits, and electricity price signals. It then enters the transmission network layer to optimize and solve the economic dispatch model, generating a boundary power plan which is sent to the distribution network layer. The distribution network layer performs local optimization based on the received plan and verifies self-regulatory constraints. If power or voltage exceeds limits, it invokes local resource adjustments, eliminating the exceedances through network reconfiguration and energy storage adjustments, and then re-optimizes. If the constraints are met, it directly uploads the boundary correction amount to the boundary consistency judgment module. This module evaluates boundary differences; if convergence is not achieved, it returns to the transmission network layer to update the optimization, forming a closed-loop iteration until the boundary power converges and becomes consistent. Finally, it enters the execution correction phase, outputting the optimization schemes of each layer to complete the entire collaborative dispatch process, demonstrating the core characteristics of hierarchical distributed optimization and self-regulatory constraint guarantees.

[0084] Figure 6 It is the theoretical process framework of the Objective Cascade Method (ATC), which defines the core logic of transportation and distribution coordinated optimization; through Figure 7 This paper explains the specific implementation of the framework in distributed computing scenarios, and transforms ATC's "layered optimization + boundary interaction" into executable algorithm steps through technical details such as parallel subproblems, multiplier updates, and convergence judgment.

[0085] Figure 7 The core of the process shown is an iterative loop, the purpose of which is to enable the transmission network (TSO) and multiple distribution networks (DSO) to achieve a global optimal solution by exchanging boundary information while maintaining their independent optimization.

[0086] The process is as follows: After initialization, the system enters a loop, incrementing the iteration counter i. Multiple Designated Sorts (DSOs) optimize their respective subproblems in parallel, which is crucial for hierarchical computation and greatly improves efficiency. Each DSO uploads its optimized boundary conditions to the Transmission Network Solution (TSS). After receiving information from all DSOs, the TSO performs global optimization and calculates new boundary conditions. The TSO then sends the new boundary conditions to each DSO. The system performs a convergence check, verifying whether the boundary conditions of the TSO and DSOs are consistent (whether the error is less than a set threshold). If convergence has not occurred, the multiplier coefficients are updated, and the system proceeds to the next iteration, gradually bringing the results of both sides into agreement. If convergence has occurred, the loop is exited, and the final optimization result is output.

[0087] Further preferred, the real-time data acquisition in step one specifically involves multi-source data acquisition.

[0088] Further preferred sources of multi-source data collection are as follows:

[0089] 1) Meteorological data;

[0090] 2) SCADA system data;

[0091] 3) Electricity market platform data;

[0092] Meteorological data includes: wind speed, irradiance, temperature, and humidity, sourced from weather stations;

[0093] SCADA system data includes: real-time power, voltage, and device switching status;

[0094] The electricity market platform data includes: real-time electricity prices and inter-regional transaction information.

[0095] The further optimized mutual aid algorithm is as follows:

[0096] First, for the two regions involved in the transaction, determine whether their resource output is at its peak state—for solar energy resources.

[0097] For example, if the two regions involved in the transaction are Region A and Region B, and the irradiance of Region A exceeds 700W / m² and the cloud cover is less than 30%, then its solar energy is determined to be at its peak. For wind energy resources, if the wind speed in Region B is between 4 and 25m / s and the turbulence intensity is less than 0.2, then its wind energy is determined to be at its peak.

[0098] If the two regions meet the complementary conditions of peak solar energy and peak wind energy respectively, they enter the mutual assistance potential assessment stage: When calculating the tradable capacity, three constraints need to be considered comprehensively - 80% of the surplus capacity of region A, the deficit capacity of region B, and the remaining capacity of the transmission line between the two regions. The minimum of the three is taken as the actual tradable capacity.

[0099] Based on this capacity, revenue distribution is determined through Nash negotiations: with the two regions as participants, tradable capacity as the resource, and a transaction cost function as the basis, the optimal transaction price and duration are negotiated, and a mutual assistance agreement is signed accordingly, clarifying core terms such as the seller, buyer, transaction volume, price, and duration. Finally, power adjustment operations are executed according to the agreement to achieve cross-regional power mutual assistance. If the two regions do not meet the conditions for spatiotemporal complementarity, there is no need for mutual assistance, and zero adjustment is returned. The revenue distribution process determined through Nash negotiations is automatically calculated using the mutual assistance transaction formula.

[0100] The mutual assistance algorithm in S31 is as follows:

[0101] ;

[0102] ;

[0103] parameter:

[0104] :area Total benefits after participating in cross-regional power exchange

[0105] Unit price coefficient for electricity exports

[0106] : Electricity output from region q to other regions

[0107] Cost savings achieved through cross-regional mutual assistance

[0108] Operating costs during cross-regional mutual assistance

[0109] Total number of regions participating in cross-regional mutual assistance

[0110] Benefits of running region q independently

[0111] : Transmission network to the first The amount of electricity supplied by each electricity purchaser.

[0112] Further optimized, the reconstructed linkage mechanism is as follows:

[0113] Net load power at time t Exceeding the maximum allowable power supply limit When the sum of the load power and the dead zone threshold ε is reached (i.e., the net load power exceeds the upper limit), first determine whether it is possible to access region B:

[0114] If access to Zone B is possible, then perform the "Close Zone B Switch" operation (responding to power limit exceedance by moving in load).

[0115] If it is not possible to connect to area B, then perform the "energy storage charging" operation;

[0116] Net load power at time t Below the minimum allowable power receiving limit When the difference between the load and the operating dead zone threshold ε is greater than the lower limit (i.e., the net load power exceeds the lower limit), the "disconnect zone A switch" operation is executed (responding to the power limit exceedance by removing the load).

[0117] Switch action constraints:

[0118] Maximum number of operations per day ≤ 5 (to extend equipment life).

[0119] Net load power It is a core indicator in the power system used to accurately depict "the power that the system actually needs to obtain from the outside (such as the main grid)". This is the maximum capacity that a system (such as a regional power grid) can safely receive power from the outside (main grid); while net load power is the "actual power required by the system." > When the value is +ε, it indicates that the "actual power demand" has approached / exceeded the "safe power limit." At this point, the system must initiate reconfiguration measures (such as "closing the B switch"), otherwise it will face the risk of overload. Net load power. It is the core trigger condition for the reconfiguration strategy. Only when the net load power exceeds the limit (exceeding the power receiving capacity + action dead zone threshold) is it necessary to enter the "can access area B" judgment and execute the switching operation. It is the key hub between "system status perception" (checking whether the net load exceeds the limit) and "control action execution" (reconfiguration operation).

[0120] Comparative Verification: In the Fujian-Guangdong power grid interconnection project, the traditional manual dispatching method, limited by information processing capacity and coordination efficiency, could only complete an average of two inter-provincial transactions per day. After applying this mechanism, the system can automatically analyze the complementary characteristics of wind and solar power output and generate the optimal transaction plan, resulting in a 40% increase in the average daily transaction volume. In the Yunnan power grid, this mechanism successfully facilitated the inter-regional consumption of over 2 billion kilowatt-hours of surplus hydropower during the 2024 rainy season and reduced network losses by 12% by optimizing power flow distribution.

[0121] This invention drives decision-making through hybrid modeling, integrating scenario analysis and interval methods to generate five typical scenarios, reducing wind curtailment rate to 3.9% and eliminating voltage over-limit accidents. Through a multi-objective, hierarchical distributed optimization framework, it establishes a three-level optimization model for transmission networks, distribution networks, and local resources, increasing the self-consumption rate of renewable energy by 32 percentage points to 87%. Furthermore, through a cross-regional mutual assistance and reconfiguration linkage mechanism, it designs a mutual assistance algorithm based on Nash negotiation, breaking down trading barriers between distribution networks and achieving a measured reduction in power peak-valley difference of 18.7 percentage points. These technological breakthroughs effectively improve the economy, security, and environmental friendliness of distribution networks. Attached Figure Description

[0122] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with the invention and, together with the description, serve to explain the principles of the invention.

[0123] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0124] Figure 1 This is a schematic diagram illustrating the technical process of the present invention.

[0125] Figure 2 This diagram illustrates the streaming processing architecture of the present invention.

[0126] Figure 3 This diagram illustrates the scenario analysis method of the present invention.

[0127] Figure 4 This diagram illustrates the interval method of the present invention.

[0128] Figure 5 This diagram illustrates the principle of the hybrid model of the present invention.

[0129] Figure 6 This is a flowchart illustrating the Target Cascade Method (ATC) of this invention.

[0130] Figure 7 This is a flowchart illustrating the distributed computing process for power transmission and distribution networks according to the present invention. Detailed Implementation

[0131] To better understand the above-mentioned objectives, features, and advantages of the present invention, the solutions of the present invention will be further described below. It should be noted that, unless otherwise specified, the embodiments of the present invention and the features thereof can be combined with each other.

[0132] Many specific details are set forth in the following description in order to provide a full understanding of the invention, but the invention may also be practiced in other ways different from those described herein; obviously, the embodiments in the specification are only some embodiments of the invention, and not all embodiments.

[0133] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0134] A distributed energy dispatching method for power distribution networks based on hybrid uncertainty modeling constructs a three-layer architecture of "data perception-optimization decision-cooperative execution," forming a closed-loop control system through bidirectional data flow, including the following steps:

[0135] Step 1: Data Awareness Layer;

[0136] Real-time data acquisition is processed by a streaming engine and feature extraction module to obtain a standardized feature matrix, and the results are transmitted to the optimization decision layer.

[0137] Step one involves real-time data acquisition, specifically multi-source data acquisition. The sources of this multi-source data are as follows:

[0138] 1) Meteorological data;

[0139] 2) SCADA system data;

[0140] 3) Electricity market platform data;

[0141] Meteorological data: wind speed (unit: m / s), irradiance (unit: W / m²), temperature (unit: ℃).

[0142] SCADA system data: real-time power (MW), voltage (pu), and device switching status.

[0143] Electricity market platform data: real-time electricity price (yuan / kWh), inter-regional transaction information.

[0144] First, the real-time collected data is extracted and standardized through the feature extraction module to ensure data quality.

[0145] The process of real-time data acquisition, extraction, and standardization is as follows:

[0146] First, multi-source data containing wind speed, irradiance, temperature, power, voltage, equipment on / off status, electricity price, and inter-regional transaction information is obtained from a Kafka real-time stream (JSON format). This data then enters the preprocessing stage—first, eight features are extracted, and missing values ​​are filled using linear interpolation (for categorical features such as equipment on / off status, the mode is used to fill in missing values ​​based on business logic). Then, the multi-source data from 24 hours × 4 points (one point every 15 minutes) is strictly aligned according to timestamps (hourly / daily features are extended to 15-minute granularity through forward padding or amortization). After each feature is independently standardized using StandardScaler, the sliding window method is used to generate the (1,24,8) input format (1 sample, 24 time steps, 8 features) required by LSTM.

[0147] The LSTM prediction model is then defined using a Sequential structure. The first LSTM layer contains 64 units, takes an input shape of (24,8), and returns a sequence with the activation function tanh. The second LSTM layer contains 32 units and outputs only the last time step. The fully connected output layer has 96 units (corresponding to the prediction curve for 24 hours × 4 points) and is compiled using the adam optimizer and mean squared error loss. During prediction, pre-trained weights are loaded, and the pre-processed data is input. The model predicts an output of (1,96), which is then de-standardized to recover a 96-dimensional load prediction vector.

[0148] Finally, a stream processing job was built using Flink: real-time data was consumed from Kafka, a preprocessing function was called via map mapping to generate LSTM input, a prediction function was called via map mapping to obtain the result, and finally the 96-point prediction curve was output to Redis cache to realize the real-time load prediction function.

[0149] Step 2: Optimize the decision-making level;

[0150] S21. Five typical scenarios are generated through a hybrid model, which uses a multi-source uncertain hybrid modeling method.

[0151] The hybrid model described in S21 combines the interval method and the scenario analysis method to generate five typical scenarios: the basic prediction scenario, the uphill scenario, the downhill scenario, the lower limit of output scenario, and the upper limit of output scenario.

[0152] S211. Receive the real-time data collected after the processing in step one;

[0153] S212, Category 5 typical scenarios are represented as follows:

[0154]

[0155]

[0156] parameter:

[0157] Output power of the wind power generation system at time t

[0158] Wind power at time t Mathematical expectation

[0159] Scene number, corresponding to the basic prediction scene.

[0160] Scene number, corresponding to the uphill climbing scene.

[0161] Scene number, corresponding to the downhill climbing scene.

[0162] Scene number, corresponding to the scene with the lower limit of output.

[0163] Scene number, corresponding to the scene with the maximum output.

[0164] 99th percentile of historical wind power ramp-up rate

[0165] The steepest drop in wind power measured during typhoon season

[0166] Time intervals during scene construction

[0167] Lower limit of wind power output

[0168] Upper limit of wind power output

[0169] S213, Construction of Uncertainty Sets:

[0170]

[0171] parameter:

[0172] Wind power capacity lower limit

[0173] Wind power capacity limit

[0174] Wind power ramp-up variation

[0175] Lower limit of climbing constraint

[0176] Upper limit of ramp constraint

[0177] in Indicates the output range. This indicates a climbing constraint.

[0178] The hybrid modeling process is as follows:

[0179] First, historical data from four categories are integrated: meteorological data (wind speed, irradiance, and temperature from weather stations), load data (electricity consumption information collection system), power grid status data (voltage, frequency, and network loss from SCADA), and market data (electricity prices from the power trading platform) to form a multi-dimensional historical dataset (the training set for the hybrid model). Then, a kernel density estimation model (i.e., the hybrid model) is constructed based on this dataset, with a bandwidth parameter set to 0.5 (to control the smoothness of the kernel function; a larger bandwidth results in a smoother density estimate, and vice versa). The hybrid model is trained to learn the latent probability density distribution of the data. Finally, the hybrid model is used to transform the real-time collected data into the final probability density value, achieving probability density estimation of real-time collected data points under the historical data distribution. This can be used for uncertainty analysis, risk scenario generation, and other scenarios.

[0180] S22. Construct a multi-objective, hierarchical, distributed optimization framework;

[0181] After optimization by a multi-objective, hierarchical distributed optimization framework, the output of the hybrid model calculates the scheduling instruction set and the transmission network's direction to the first... Electricity supplied by individual power purchasers To the collaborative execution layer;

[0182] S22 specifically involves establishing a three-level optimization model for transmission network, distribution network, and local resources.

[0183] The power transmission network layer is represented as follows:

[0184]

[0185] parameter:

[0186] Total number of time periods within the scheduling period

[0187] The current scheduling period

[0188] Total number of generator units participating in dispatching in the power transmission network

[0189] : The generator set number currently participating in the dispatch

[0190] : No. Unit power generation cost coefficient per generator set, unit: yuan / MWh

[0191] : No. The generator set is Power generation during a given time period, in MW

[0192] Cost coefficient for unit reserve, unit: yuan / MW

[0193] : power transmission network Reserve capacity for a given period, in MW

[0194] The electricity price for the kth electricity purchaser, in yuan / MWh.

[0195] : power transmission network The amount of electricity delivered to the k-th electricity purchaser during the specified time period, in MWh.

[0196] in Indicates the cost of electricity generation. Indicates the cost of reserve. This indicates revenue from electricity sales;

[0197] The following self-regulatory constraints are applied to the distribution network layer:

[0198]

[0199] Key threshold:

[0200]

[0201] parameter:

[0202] Active power interacting between the distribution network and the main grid at time t, unit: MW

[0203] Active power exchanged between the distribution network and the main grid at time t-1, unit: MW

[0204] Maximum permissible rate of change of power between the distribution network and the main grid, unit: MW / min

[0205] Time interval

[0206] : The first in the distribution network at time t Voltage amplitude at each node

[0207] The objective function for the local resource layer is as follows:

[0208]

[0209]

[0210] parameter:

[0211] The number of time steps in the scheduling cycle

[0212] Time Index

[0213] Uncertainty scenarios

[0214] Total number of scenes

[0215] Scene probability of occurrence

[0216] Adjusted electricity price at time t (unit: yuan / MWh)

[0217] Optimized trading power at time t

[0218] Interaction power at time t in scenario s

[0219] Time-period fluctuation penalty coefficient (unit: yuan / (MW²))

[0220] Plan deviation penalty coefficient (unit: yuan / MW)

[0221] in Represents revenue from electricity sales. Indicates time-period fluctuation penalty. This indicates a penalty for deviation from the plan.

[0222] The DER self-use rate of the local resource layer is calculated as follows:

[0223]

[0224] parameter:

[0225] DER self-use rate

[0226] Time interval

[0227] The active power available from distributed energy sources in the distribution network at time t, in MW.

[0228] Active power of the energy storage system at time t, unit: MW

[0229] Active power of demand response at time t, unit: MW

[0230] Total number of time periods in the time interval

[0231] The DER self-consumption rate function is an indicator that quantifies the degree to which distributed energy is directly consumed in a local area, reflecting the local energy self-governance capability. The local resource layer is the physical / functional layer that carries distributed energy and realizes local consumption. The self-consumption rate is the core basis for evaluating the operating efficiency of the local resource layer and guiding resource allocation.

[0232] The objective cascade method is used to solve the problem in layers: the transmission network layer minimizes the generation cost and reserve cost, the distribution network layer applies self-regulatory constraints: power change rate ≤ 5MW / min, voltage 0.95-1.05pu, and the local resource layer adjusts the output in response to the electricity price signal.

[0233] The specific calculation process is as follows, and a parallel solution framework is constructed:

[0234] The optimized decision-making layer decomposes the global problem into a main problem (decision variables such as transmission power and power generation plan) and parallel sub-problems (decision variables such as load response and energy storage action). By processing the main problem and sub-problems simultaneously through parallel computing, the solution time is significantly reduced (from 300 seconds to 180 seconds in specific implementation), directly meeting the real-time scheduling requirements.

[0235] Main problem decision variables:

[0236]

[0237] Decision variables for parallel subproblems:

[0238]

[0239] parameter:

[0240] Linear cost term It is a cost coefficient vector. It is a vector of decision variables. This represents the "linear combination cost of decision variables," and the optimization objective is to minimize the total cost.

[0241] Non-linear risk / penalty term

[0242] Uncertainty scenarios

[0243] Scene Disturbance / response variable

[0244] Scene constant vector

[0245] Scene coefficient matrix

[0246] Power curtailment

[0247] Acceleration strategy:

[0248] Column Constraint Generation (C&CG) Cuts Invalid Scenarios: The column constraint generation algorithm is an exact algorithm for solving two-stage robust optimization problems. Compared to traditional Benders decomposition, C&CG adds not only a Benders cut (similar to a "column") to the main problem in each iteration, but also variables and constraints related to the worst-case scenario (i.e., "constraint generation"). C&CG can handle both optimality cuts and feasible cuts simultaneously, simplifying the process. Because scenario-related variables and constraints are explicitly introduced into the main problem, the solution to the main problem can better approximate the original problem, thus usually obtaining a tighter lower bound than traditional Benders decomposition and faster convergence. It is particularly suitable for two-stage robust optimization problems where the uncertainty set is discrete and finite, or where the worst-case scenario can be generated through the solution process.

[0249] The Adaptive Alternating Direction Multiplier Method (ADMM) decomposes coupled constraints: First, the global constraints in transmission and distribution coordinated scheduling are decomposed into subproblems that can be solved in parallel, such as regional generation planning and load response. Then, an augmented Lagrangian function is constructed, and a dual variable y and an adaptive penalty factor are introduced. Next, the decision variables are updated alternately in the iteration: first, the power generation output x is optimized, then the energy storage action z is optimized, and the dual variable is updated to penalize the constraint violation Ax + Bz − c; at the same time, the decision variables are dynamically adjusted based on the original residual r and the dual residual s. (like If it is too large, it will increase. Accelerate convergence, if If it is too large, then decrease. (To avoid oscillations); finally, convergence occurs when the residuals drop to the threshold, achieving efficient collaborative solution of complex constraints.

[0250] In the coordinated optimization of transportation and distribution, the acceleration strategy employs the Objective Concatenation (ATC) method to coordinate boundary conditions at the optimization decision level. The Adaptive Alternating Direction Multiplier (ADMM) method is used to decompose coupled constraints, achieving distributed and efficient convergence through iterative updates of dual variables and penalty factors. Simultaneously, the Column Constraint Generation (C&CG) algorithm cuts out invalid solutions in uncertain scenarios, ensuring robustness and preventing optimization decisions from becoming unstable due to random fluctuations.

[0251] Dynamic scene filtering:

[0252]

[0253] parameter:

[0254] : Wind power time series under scenario s.

[0255] Initial reference power time series.

[0256] L2 norm (Euclidean distance, calculated as the square root of the sum of squares of the power differences at each time point) of two sequences measures the overall deviation between scene s and the initial scene.

[0257] Initial reference power sequence The maximum value in (used to normalize the bias and eliminate the influence of power dimensions).

[0258] Screening threshold: 5% (retaining scenarios with significant fluctuations), derived from the tolerance limit for grid connection fluctuations of distributed power sources in the IEEE 1547 standard.

[0259] Dynamic scenario screening explained in detail: The optimization decision layer needs to make decisions based on 5 typical scenarios generated by the hybrid model, set a 5% fluctuation threshold (derived from the IEEE 1547 standard), and dynamically screen scenarios with significant fluctuations, thereby increasing the scenario processing capacity from 1000+ to 5000+, making the optimization decision more focused on key risks and reducing computational redundancy.

[0260] Comparative verification: Compared with traditional centralized solutions, in a provincial power grid case, the computation time was reduced from 300 seconds to 180 seconds, and the memory usage decreased from 32GB to 16GB. In a large system containing 50 distribution networks, the scenario processing capacity increased from 1000+ to 5000+, with no convergence failures observed.

[0261] The algorithm is embedded in the decision-making process. In the distributed computing process of power transmission and distribution networks, the optimization decision-making layer relies on the ADMM algorithm to achieve hierarchical collaboration. For example... Figure 7 As shown, after independent optimization of the transmission network (TSO) and distribution network (DSO), they converge iteratively by exchanging boundary information. The dynamic penalty factor adjustment mechanism of ADMM (such as adaptive adjustment ρ based on residuals) is the core of achieving efficient collaboration.

[0262] The dynamic scene selection algorithm prioritizes extracting high-probability scenes (such as scenes with a wind curtailment rate > 5%) and feeds them into the optimization model to ensure accurate and efficient decision-making.

[0263] Step 3: Collaborative Execution Layer;

[0264] The collaborative execution layer is primarily responsible for translating optimization decisions into practical operations, ensuring the economy, safety, and real-time performance of the power grid.

[0265] The final output includes: execution instructions and correction operations: including topology reconfiguration (such as switching actions), energy storage charging and discharging plans, power adjustment instructions, and real-time mutual aid transaction execution: outputting power export / import adjustments in accordance with the Nash negotiation agreement.

[0266] In actual implementation, it is preferable to add an online rolling correction step, that is, to directly receive the scheduling plan issued by the optimization decision-making level, generate the corrected execution instructions to guide the execution of cross-regional transactions, and feed back the mutual assistance transaction results to the online rolling correction to refresh the boundary conditions of the next round of detection.

[0267] The two core mechanisms of online rolling correction are the over-limit detection algorithm and the dynamic refresh mechanism;

[0268] Exceedance detection algorithm:

[0269] The system periodically (every 15 minutes) monitors the grid status. First, it checks if node voltages exceed limits. If a node voltage deviates from its rated value (baseline 1.0 per unit) by more than 5% (threshold 0.05 per unit), a grid reconfiguration operation is triggered to restore voltage stability. If the voltage does not exceed limits, the system further checks for abnormal power change rates. When the absolute value of the power change rate over time exceeds a preset maximum allowable change rate threshold, the smoothing function of the energy storage system is activated. This suppresses sudden power fluctuations through energy storage charging and discharging, ensuring system dynamic stability. This entire process repeats continuously, achieving real-time over-limit monitoring and closed-loop correction of the grid's operating status.

[0270] Dynamic refresh mechanism:

[0271] The dynamic refresh mechanism achieves dynamic optimization through the fusion of real-time and predictive data in a dual-channel manner and a rolling time window update strategy. Real-time measurement data (SCADA millisecond level) and predictive data (LSTM 15-minute level) are adaptively mixed with weights according to the operating status (α=0.9 in emergency state, emphasizing real-time measurement; α=0.6 in stable state, balancing historical predictions). A fixed 6-hour window length is used for rolling updates to ensure data timeliness of less than 15 minutes. The mechanism drives the core content update through dual logics: periodic triggering (forced refresh of measurement data / topology status / standby capacity every 900 seconds) and event triggering (responding to weather warnings / load surges >10% / equipment failures / electricity price fluctuations). Validation in a park in Jiangsu Province shows that this mechanism reduces wind and solar prediction errors by 42% to 4.7%, increases the effective scenario ratio to 95%, speeds up voltage over-limit detection by 80% to 8.7 seconds, and supports a stable renewable energy absorption rate of 95%±0.8%, reducing fluctuations by 65% ​​compared to static models.

[0272] S31. Constructing cross-regional mutual assistance and reconstructing linkage mechanisms

[0273] Design a mutual aid algorithm based on Nash negotiation, and design a real-time response strategy by reconstructing the linkage mechanism;

[0274] The mutual assistance algorithm is as follows:

[0275] ;

[0276] ;

[0277] parameter:

[0278] :area Total benefits after participating in cross-regional power exchange

[0279] Unit price coefficient for electricity exports

[0280] : Electricity output from region q to other regions

[0281] Cost savings achieved through cross-regional mutual assistance

[0282] Operating costs during cross-regional mutual assistance

[0283] Total number of regions participating in cross-regional mutual assistance

[0284] Benefits of running region q independently

[0285] : Transmission network to the first The amount of electricity supplied by each electricity purchaser.

[0286] A real-time response strategy is designed by reconstructing the linkage mechanism, as follows:

[0287] When the net load power PDt at time t exceeds the sum of the allowable upper limit of power received PG+ and the dead zone threshold ε (i.e., the net load power exceeds the upper limit), first determine whether it is possible to connect to area B:

[0288] If access to Zone B is possible, then perform the "Close Zone B Switch" operation (responding to power limit exceedance by moving in load).

[0289] If it is not possible to connect to area B, then perform the "energy storage charging" operation;

[0290] When the net load power PDt at time t is lower than the difference between the allowable lower limit of the power received PG− and the dead zone threshold ε (i.e. the net load power exceeds the lower limit), the "disconnect zone A switch" operation is executed (responding to the power limit exceedance by removing the load).

[0291] Switch action constraints:

[0292] Maximum number of operations per day ≤ 5 (to extend equipment life).

[0293] S32, Demand-side response;

[0294] The system issues electricity price signals to guide users to adjust their load and aggregates interruptible loads to participate in system balancing.

[0295] The power grid drives users to adjust their electricity consumption behavior by setting "pricing tools" (i.e., "price signals") such as time-of-use pricing, peak pricing, and real-time pricing: higher prices during peak hours lead users to reduce unnecessary electricity consumption and avoid high electricity bills; lower prices during off-peak hours encourage users to consume electricity at these times, saving costs and filling the "power gap" in the grid during off-peak hours. This approach shifts users from "passive electricity consumption" to "active cooperation," alleviating the power supply pressure on the grid during peak periods and improving the power absorption capacity during off-peak periods, achieving "peak shaving and valley filling." "Interruptible loads" refer to users agreeing in advance with the grid to voluntarily suspend some non-critical electrical equipment when the grid faces emergencies (such as overload or a sudden drop in renewable energy generation) or needs peak shaving (the grid will provide economic compensation to offset the user's losses). "Aggregation" refers to integrating "interruptible loads" scattered across different users to form an adjustable resource pool similar to a "virtual power plant," which participates in the grid's peak shaving, frequency regulation, and emergency support operations.

[0296] The working principle is as follows:

[0297] Figure 3 In the diagram, s1, s2, and s3 represent wind power scenarios. The optimization objective of the scenario analysis method is to optimize the expected value. As the number of scenarios increases, the computational efficiency gradually decreases.

[0298] Figure 4 This is a schematic diagram of the interval method, where the interval boundaries are defined by... Figure 3The maximum and minimum values ​​for all scenarios at each time point are obtained (the confidence interval can also be obtained using the interval prediction method). Here, s1 constitutes the lower boundary for times t=1-4 and t=6, and the lower boundary for time t=5 is obtained from s2; s3 constitutes the upper boundary for times t=1-2 and t=4-6, and the upper boundary for t=3 is obtained from scenario s2. Climbing 1 and Climbing 2 represent the extreme climbing scenarios for wind power between two adjacent time points. The interval method describes the uncertainty of wind power with one prediction scenario and two extreme climbing scenarios. The optimization objective is to ensure that the system has sufficient backup and climbing capacity to cope with extreme climbing scenarios. The interval method has high computational efficiency, but the results are relatively conservative.

[0299] Figure 5 This is a schematic diagram of the hybrid model principle. The formation of interval boundaries is the same as in the interval method, except that the extreme climbing constraint is replaced by four climbing scenarios, and the slope of each climbing scenario is determined by... Figure 4 The maximum uphill / downhill ramp rates were obtained at various time points. Between time points 1 and 2, the maximum uphill ramp rate was 10 MW / h, and the maximum downhill ramp rate was 0 MW / h. Therefore, in Figure 5 In the above, the upper boundary at t=2 marks the end of the uphill scene with a slope of 10, meaning the starting point is at t=1 with a power of 40MW, as shown by the red dotted line. The end of the downhill scene is at the lower boundary at t=2 with a slope of 0, and the starting point is at t=1 with a power of 30MW, as shown by the blue dashed line. Between times 2 and 3, the maximum uphill rate is 20MW / h, and the downhill rate is 0MW / h. Therefore, the end of the uphill scene is at the upper boundary at t=3 with a slope of 20, and the starting point is at t=2 with a power of 40MW, as shown by the black dotted line. The downhill scene has a slope of 0, as shown by the black dashed line. The generation method for the uphill scenes at other times is the same.

[0300] The hybrid model replaces the extreme climbing scenario in the interval method with a climbing scenario, which reduces conservatism compared to the interval method. Compared to the scenario analysis method, it replaces a large number of wind power scenarios with 5 typical scenarios, which improves computational efficiency without significantly deteriorating the objective function.

[0301] The data forward transmission path of this method is: data perception layer → optimization decision layer → collaborative execution layer (scheduling instruction flow); the data feedback adjustment path is: collaborative execution layer → data perception layer (real-time feedback of voltage over-limit / load response data).

[0302] This method accurately characterizes uncertainty through big data analytics, integrating scenario analysis and interval methods. It generates five typical scenarios using a hybrid model to precisely quantify the randomness and anti-peak-shaving characteristics of distributed energy output, ensuring model reliability. Through a multi-objective, hierarchical distributed optimization framework, it establishes a three-level optimization model for transmission networks, distribution networks, and local resources. At the transmission network level, it optimizes generation and reserve costs, while at the distribution network level, it implements self-regulatory constraints, achieving a dynamic balance between economy and security. A cross-regional mutual assistance and reconfiguration linkage mechanism promotes resource allocation. A mutual assistance algorithm based on Nash negotiation is designed to drive energy exchange between distribution networks, and topology reconfiguration logic is integrated to respond to load fluctuations in real time, eliminating regional barriers. This method, through the aforementioned closed-loop technology, directly addresses the pain points of traditional dispatching, achieving end-to-end optimization from data-driven to decision-making execution.

[0303] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the present invention. Although detailed descriptions have been provided 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 therein; 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, and they should all be covered within the protection scope of the claims.

Claims

1. A method for dispatching distributed energy resources in a distribution network based on hybrid uncertainty modeling, characterized in that: A three-tier architecture of data perception, optimization decision-making, and collaborative execution was constructed, forming a closed-loop control system through bidirectional data flow, including the following steps: Step 1: Data Awareness Layer; Real-time data acquisition is processed by a streaming engine and feature extraction module to obtain a standardized feature matrix, and the results are transmitted to the optimization decision layer. Step 2: Optimize the decision-making level; S21. Five typical scenarios are generated through a hybrid model, which uses a multi-source uncertain hybrid modeling method. S22. Construct a multi-objective, hierarchical, distributed optimization framework; After optimization by a multi-objective, hierarchical distributed optimization framework, the output of the hybrid model calculates the scheduling instruction set and the transmission network's direction to the first... Electricity supplied by individual power purchasers To the collaborative execution layer; Step 3: Collaborative Execution Layer; S31. Establish cross-regional mutual assistance and restructure linkage mechanisms; Design a mutual aid algorithm based on Nash negotiation, and design a real-time response strategy by reconstructing the linkage mechanism; S32, Demand-side response; The system issues electricity price signals to guide users to adjust their load and aggregates interruptible loads to participate in system balancing.

2. The method for dispatching distributed energy resources in a distribution network based on hybrid uncertainty modeling as described in claim 1, characterized in that: The hybrid model described in S21 combines the interval method and the scenario analysis method to generate five typical scenarios: the basic prediction scenario, the uphill scenario, the downhill scenario, the lower limit of output scenario, and the upper limit of output scenario. S211. Receive the real-time data collected after the processing in step one; S212, Category 5 typical scenarios are represented as follows: ； ； parameter: : Output power of the wind power generation system at time t; Wind power at time t The mathematical expectation; Scene number, corresponding to the basic prediction scene; Scene number, corresponding to the uphill climbing scene; Scene number, corresponding to the downhill climbing scene; Scene number, corresponding to the scene with the lower limit of output; Scene number, corresponding to the scene with the maximum output capacity; : 99th percentile of historical wind power ramp-up rate; The steepest drop in wind power output measured during typhoon season; The time interval during scene construction; : The lower limit of wind power output; The upper limit of wind power output; S213, Construction of Uncertainty Sets: ； parameter: Lower limit of wind power output; Wind power capacity limit; : Changes in wind power output during ramp-up; : Lower limit of the ramp constraint; : Upper limit of ramp constraint; in Indicates the output range. Indicates a ramp constraint; The uncertain set output by the hybrid model is input into a multi-objective hierarchical distributed optimization framework for optimization.

3. The method for distributed energy dispatching in a distribution network based on hybrid uncertainty modeling as described in claim 1, characterized in that: S22 specifically involves establishing a three-level optimization model for transmission network, distribution network, and local resources. The power transmission network layer is represented as follows: ； parameter: The total number of time periods within the scheduling cycle; The current time period being scheduled; The total number of generator units participating in dispatching in the power transmission network; : The generator set number currently participating in the dispatch; : Unit power generation cost coefficient of the i-th generator set, unit: yuan / MWh; The i-th generator set is Power generation during a given time period, in MW; Cost coefficient for standby units, unit: yuan / MW; : power transmission network Reserve capacity for a given time period, in MW; The electricity price for the kth electricity purchaser, in yuan / MWh; : power transmission network The amount of electricity delivered to the k-th electricity purchaser during the specified time period, in MWh; in Indicates the cost of electricity generation. Indicates the cost of reserve. This indicates revenue from electricity sales; The following self-regulatory constraints are applied to the distribution network layer: ； Key threshold: ； The DER self-use rate of the local resource layer is calculated as follows: ； parameter: Active power interacting between the distribution network and the main grid at time t, in MW; Active power interacting between the distribution network and the main grid at time t-1, unit: MW; : Maximum permissible rate of change of power between the distribution network and the main grid, unit: MW / min; Time interval; : The first in the distribution network at time t Voltage amplitude at each node; DER self-use rate; The active power available from distributed energy sources in the distribution network at time t, in MW; Active power of the energy storage system at time t, in MW; Active power of demand response at time t, in MW; : Total number of time periods within the time interval; The objective cascade method is used to solve the problem in layers: the transmission network layer minimizes the generation cost and reserve cost, the distribution network layer applies self-regulatory constraints: power change rate ≤ 5MW / min, voltage 0.95-1.05pu, and the local resource layer adjusts the output in response to the electricity price signal.

4. The method for distributed energy dispatching in a distribution network based on hybrid uncertainty modeling as described in claim 1, characterized in that: The real-time data collection in step one specifically involves multi-source data collection.

5. The method for distributed energy dispatching in a distribution network based on hybrid uncertainty modeling as described in claim 4, characterized in that: The sources of multi-source data collection are as follows: 1) Meteorological data; 2) SCADA system data; 3) Electricity market platform data.

6. The method for dispatching distributed energy resources in a distribution network based on hybrid uncertainty modeling as described in claim 5, characterized in that: Meteorological data includes: wind speed, irradiance, temperature, and humidity, sourced from weather stations; SCADA system data includes: real-time power, voltage, and device switching status; The electricity market platform data includes: real-time electricity prices and inter-regional transaction information.

7. The method for dispatching distributed energy resources in a distribution network based on hybrid uncertainty modeling as described in claim 1, characterized in that: The mutual assistance algorithm in S31 is as follows: ； ； parameter: :area Total benefits after participating in cross-regional power exchange; : Unit price coefficient for electricity exports; : The electrical power exported from region q to other regions; Cost savings achieved through cross-regional mutual assistance; Operating costs during cross-regional mutual assistance processes; The total number of regions participating in cross-regional mutual assistance; : The benefits of running region q independently; : Transmission network to the first The amount of electricity supplied by each electricity purchaser.

Images