A distributed photovoltaic-energy storage collaborative planning bi-level optimization method for voltage deviation suppression

Abstract

The application discloses a kind of distributed photovoltaic-energy storage collaborative planning double-layer optimization methods for voltage offset suppression, belong to the field of electric power system and its automation field, including: collecting distribution network historical source load data, reconstructing node photovoltaic output time series by inverse distance weighted spatial interpolation, combined with thermodynamic free energy criterion cluster generation typical day scene set;Build the upper planning model with annualized investment cost and network voltage offset as double target, decision photovoltaic storage site and capacity;Establish the lower optimal operation model, solve energy storage charging and discharging strategy and feedback voltage performance index;Riemann manifold dissipative structure self-organizing evolution algorithm is used to solve collaboratively, through the Fisher information matrix guide natural gradient update, combined with Helmholtz free energy screening and entropy production rate temperature control mechanism, output pareto optimal planning scheme.The application realizes the complementation of photovoltaic storage function and planning-operation closed loop, effectively suppresses distribution network bidirectional voltage offset.

Description

Technical Field

[0001] This invention belongs to the field of power systems and their automation, and particularly relates to a two-layer optimization method for distributed photovoltaic-energy storage collaborative planning oriented towards voltage offset suppression. Background Technology

[0002] With the construction of new power systems, the penetration rate of distributed photovoltaic (PV) power in medium- and low-voltage distribution networks continues to rise, and traditional passive distribution networks are transforming into active distribution networks. However, the randomness and volatility of PV output and the significant mismatch between PV output and load demand in time and space have led to increasingly prominent voltage quality problems in distribution networks: during peak PV output periods at midday, node voltages rise, even exceeding the safe operating limit; during peak load periods at night or periods of sudden drops in PV output, node voltages fall below the lower limit. This bidirectional voltage deviation directly affects the operational safety and power quality of the distribution network, and urgently needs to be effectively suppressed through the coordinated planning of PV and energy storage.

[0003] Existing technologies still have the following shortcomings in addressing the above problems: In terms of generating typical daily scenarios, mainstream methods are mostly based on temporal clustering, ignoring the differences in the geographical distribution of photovoltaic power output, resulting in deviations in planning results due to insufficient consideration of spatial correlation; In terms of model solving, photovoltaic-storage collaborative planning is essentially a high-dimensional, non-convex, two-level multi-objective optimization problem with mixed integer constraints, and traditional evolutionary algorithms are prone to getting trapped in local optima, making it difficult to guarantee the quality and uniformity of the Pareto solution set; In terms of coupled modeling of planning and operation, some studies simplify power flow constraints and fail to accurately evaluate the voltage performance of the planning scheme at the operational level, resulting in the planning scheme being detached from actual operation and lacking engineering feasibility. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a two-layer optimization method for distributed photovoltaic-energy storage collaborative planning oriented towards voltage offset suppression, comprising: Acquire basic data of the power distribution network, including node topology, branch parameters, historical photovoltaic power output data, load data, and equipment economic and technical parameters; Based on the historical photovoltaic power output data and the load data, the photovoltaic power output time series data of each node in the entire network is reconstructed using the inverse distance weighted spatial interpolation method, and the clustered scenarios are further screened based on the thermodynamic free energy criterion to generate a typical daily scenario set. Based on the typical daily scenario set, an upper-level planning model is constructed with annualized investment cost and grid voltage deviation as dual objectives. The decision variables of the upper-level planning model include the site selection status, power capacity and energy capacity of photovoltaic and energy storage. Based on the planning scheme output by the upper-level planning model, a lower-level optimal operation model based on the DistFlow power flow equation is constructed. The lower-level optimal operation model aims to minimize the daily voltage deviation, solves the optimal charging and discharging scheduling strategy for energy storage under each typical daily scenario, and feeds back the obtained voltage performance index to the upper-level planning model. A self-organizing evolution algorithm based on Riemannian manifold dissipative structure is used to solve the upper-level planning model and the lower-level optimal operation model in a coordinated manner to obtain the Pareto optimal solution set; Planning schemes are selected from the Pareto optimal solution set, and the site selection and capacity determination results of photovoltaic and energy storage, as well as the charging and discharging operation strategies of energy storage under various typical daily scenarios, are output.

[0005] Optionally, the typical daily scene set can be generated in the following ways: Based on the geographical distance between each photovoltaic observation station and the node to be estimated, the weight coefficient of each observation station to the node to be estimated is calculated by the inverse distance weighting method, and the photovoltaic output time series data of the node to be estimated is reconstructed accordingly. The reconstructed annual photovoltaic power output time series data and load data are organized into feature vectors by day, and then divided into seasonal scene clusters using the K-means algorithm; A thermodynamic free energy function is introduced to perform a secondary screening of the seasonal scene clusters, and typical days that make the scene free energy increment negative are selected as the final scene set.

[0006] Optionally, the upper-level planning model can be constructed in the following ways: The location state variables, power capacity, and energy capacity of photovoltaic and energy storage are used as upper-level decision variables; The annualized investment cost of photovoltaic modules, energy storage converters, and battery packs is taken as the first objective function. The probability-weighted cumulative amount of voltage amplitude deviation from reference voltage at each node at each time under each typical daily scenario is used as the second objective function; Set upper and lower limits for equipment capacity and constraints on energy storage power ratio.

[0007] Optionally, the construction method of the lower-level optimal operating model includes: Given the upper-level planning scheme, with the goal of minimizing the daily voltage deviation, positive and negative deviation auxiliary variables are introduced to eliminate absolute values. Establish branch voltage drop constraints and node power balance constraints based on the DistFlow equation; Establish constraints on the energy storage charge state evolution and daily periodicity; The voltage distribution obtained from solving for each typical daily scenario is fed back into the biobjective function of the upper-level planning model.

[0008] Optionally, the solution method for the Riemannian manifold dissipative structure self-organization evolution algorithm includes: A hybrid encoding strategy is adopted to uniformly represent the addressing state variables and the capacity continuous variables, and the initial population is generated using the Logistic chaotic mapping. By using weighted Chebyshev scalarization, the multi-objective performance of an individual is mapped to a microscopic potential energy, so that the population probability density follows a Boltzmann distribution. The Riemannian metric of the population distribution parameter space is defined using the Fisher information matrix, and natural gradient updates with Levenberg-Marquardt damping are performed.

[0009] Optionally, the natural gradient update can be implemented in the following ways: The population distribution parameter space is regarded as a Riemannian manifold, and its local geometric structure is described by the Fisher information matrix. The population distribution parameters are updated according to the direction of the natural gradient flow, so that the population center moves in the direction of the fastest decrease in the desired potential energy and the distribution shape is adjusted.

[0010] Optionally, it also includes: The solution set is self-organized and filtered using Helmholtz free energy to select a subset that results in a negative increment of the system's free energy. By adaptively adjusting the evolution temperature using a feedback controller based on entropy generation rate, a balance between convergence and diversity is achieved.

[0011] Optionally, the adaptive adjustment method of the evolution temperature includes: When the population converges prematurely, the evolutionary temperature is automatically increased, and the weight of the entropy term in the free energy is increased in order to accept solutions with greater diversity. When the system reaches a steady state, the temperature naturally decays, causing the algorithm to degenerate into a deterministic local search.

[0012] On the other hand, the present invention also provides an electronic device including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0013] On the other hand, the present invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0014] Compared with the prior art, the present invention has the following advantages and technical effects: A scene generation method combining inverse distance weighted spatial interpolation and thermodynamic free energy criterion clustering preserves the spatial decay characteristics and seasonal patterns of photovoltaic power output, avoiding planning deviations caused by ignoring spatial differences. A self-organizing evolution algorithm based on Riemannian manifold dissipative structures is employed, utilizing the geometric structure of the Fisher information matrix to guide population updates on the distributed manifold. Combined with Helmholtz free energy screening and entropy generation rate temperature control mechanisms, this effectively overcomes the problem of easily getting trapped in local optima in high-dimensional non-convex mixed-integer programming, generating a uniformly distributed Pareto front. A closed-loop two-layer optimization framework of upper-level planning and lower-level operation is constructed. Based on the DistFlow power flow equation, voltage distribution and energy storage dynamics are accurately simulated, and the voltage performance of the operation layer is fed back to the planning layer, ensuring the engineering feasibility of the planning scheme and the effectiveness of voltage suppression. Attached Figure Description

[0015] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic diagram illustrating the coordinated operation mechanism of the net load reduction base and the time shift of energy storage in an embodiment of the present invention. Figure 2 This is a two-layer optimization framework diagram for the combined planning of photovoltaic and energy storage according to an embodiment of the present invention; Figure 3 This is a flowchart of the Riemannian manifold self-organizing evolution algorithm according to an embodiment of the present invention; Figure 4 This is a topology diagram of a power distribution network example according to an embodiment of the present invention. Detailed Implementation

[0016] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0017] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0018] Example 1 This embodiment provides a two-layer optimization method for distributed photovoltaic-energy storage collaborative planning oriented towards voltage offset suppression, including: Step 1: Generate a typical daily scene set based on inverse distance weighted spatial interpolation and thermodynamic free energy criterion clustering. Historical photovoltaic power output data and load data of each node within the distribution network coverage area are collected. The inverse distance weighted method is used to assign differentiated weights based on the geographical distance between the observation station and the node to be estimated, reconstructing the time-series data of photovoltaic power output for each node in the entire network. Based on this, K-means clustering is used to divide the annual data into seasonal scene clusters, and the thermodynamic free energy criterion is introduced to perform a secondary screening of the scene set, selecting typical days that result in negative free energy increments as the final scene input. This approach reduces the computational complexity of the planning model while preserving the spatiotemporal correlation of photovoltaic power output.

[0019] Step Two: Construct an upper-level planning model with dual objectives of annualized investment cost and overall network voltage deviation. Upper-level decision variables include the photovoltaic and energy storage site selection state variables, power capacity, and energy capacity of each candidate node in the distribution network. The first objective function is the annualized investment cost calculated from the initial investment costs of photovoltaic modules, energy storage converters, and battery packs; the second objective function is the probability-weighted cumulative amount of voltage amplitude deviation from the reference voltage at each node at each time point in a typical daily scenario. The decision variables must satisfy upper and lower limits for equipment capacity and energy storage power ratio constraints.

[0020] Step 3: Construct a lower-level optimal operation model based on the DistFlow power flow equation. Given the upper-level planning scheme, the lower level aims to minimize the daily voltage deviation. Based on the DistFlow equation, it describes the relationship between branch voltage drop and power flow, establishes active and reactive power balance constraints at nodes, establishes energy storage charge state evolution constraints and daily periodic constraints, solves the optimal energy storage charging and discharging scheduling strategy under various typical daily scenarios, and feeds back the voltage performance indicators at the operation level to the upper level, forming a closed-loop optimization mechanism of planning guiding operation and operation verifying planning.

[0021] Step 4: Solve the upper-level multi-objective programming problem based on the self-organizing evolution algorithm of Riemannian manifold dissipative structures. A hybrid encoding strategy is adopted to uniformly represent the location state variables and capacity continuous variables. The initial population is generated using Logistic chaotic mapping. The multi-objective performance of individuals is mapped to microscopic potential energy through weighted Chebyshev scaling, and the population probability density follows a Boltzmann distribution. The Riemannian metric of the population distribution parameter space is defined using the Fisher information matrix, and natural gradient updates with Levenberg-Marquardt damping are performed. The solution set is self-organized and screened using Helmholtz free energy, and the evolution temperature is adaptively adjusted by a feedback controller based on entropy generation rate to achieve an automatic balance between convergence and diversity.

[0022] Step 5: Select a planning scheme from the converged Pareto optimal solution set, and output the site selection and capacity determination results of photovoltaic and energy storage at each node, as well as the corresponding typical daily energy storage charging and discharging operation strategy.

[0023] The specific process includes: Step 1: Acquisition of basic data and definition of parameters for the power distribution network; Step 11: Define the distribution network topology. Obtain the set of nodes in the distribution network. Set with branch roads Record the resistance of each branch. and reactance and system reference voltage and reference power In the formula, For the set of distribution network nodes, The total number of nodes. For branch road collection, branch road The resistance, branch road Reactance, The system reference voltage, This is the system reference power. All electrical quantities below are expressed in terms of... and The per-unit value is the baseline. Node 1 is the root node of the substation, and power injection from the upper-level power grid only occurs at the root node.

[0024] Step 12: Define the economic and technical parameters for photovoltaic (PV) and energy storage. The investment cost coefficient per unit capacity of PV is... The investment cost coefficient per unit power of energy storage is The unit energy investment cost coefficient for energy storage is The discount rate is The service life of the equipment is The upper and lower limits of the rated capacity of a photovoltaic inverter are: The upper and lower limits of the rated power capacity of energy storage are The upper and lower limits of energy storage capacity are The upper limit of energy storage power ratio is Energy storage charging efficiency is The discharge efficiency is The permissible range of state of charge is The number of scheduling periods per day is... The scheduling time step is Hours. In the formula, This is the investment cost coefficient per unit capacity of photovoltaic power. This is the investment cost coefficient per unit power of energy storage. This is the unit energy investment cost coefficient for energy storage. For the discount rate, For the service life of the equipment, and These are the lower and upper limits of photovoltaic capacity, respectively. and These represent the lower and upper limits of energy storage power capacity, respectively. and These are the lower and upper limits of energy storage capacity, respectively. This is the energy storage power ratio constraint coefficient. For charging efficiency, For discharge efficiency, and These are the lower and upper limits of the permissible state of charge, respectively. This refers to the number of scheduling periods per day. This is the scheduling time step.

[0025] Step 13: Define the voltage safety operation constraint range. Let the voltage safety operation range of the distribution network nodes be... The reference voltage is In the formula, This is the lower limit for safe operation of node voltage. The upper limit for safe operation of node voltage. This is the reference voltage.

[0026] Step 14: Obtain historical source-load data. Collect data within the distribution network coverage area. Annual power output data of individual photovoltaic monitoring stations ( , (and the hourly load data for each node throughout the year) and In the formula, The number of photovoltaic observation stations, For the first Observation stations at Solar power output at all times For nodes exist Active load at any given time For nodes exist Reactive load at any given moment.

[0027] Step 2: Generating a typical daily scene; Step 21: Reconstruct nodal-level photovoltaic output using inverse distance-weighted spatial interpolation. For any node in the distribution network to be estimated... ,use The photovoltaic power output time series curves of each observation station are reconstructed using the following formula: (1) (2) In the formula, For nodes exist Photovoltaic output is constantly reconstructed through interpolation. For the first Each observation station to node The weighting coefficients, For nodes With the Geographical distance between observation stations This is the distance-attenuation power exponent. The input for this step is the power output data from the observation stations obtained in step 14. Combined with the geographic coordinates of the nodes, the output is the photovoltaic power output reconfiguration sequence for all nodes in the entire network. This is for use in subsequent clustering.

[0028] Step 22: K-means clustering to divide seasonal scene clusters. The data obtained in Step 21 for the entire year... The combined time-series data of the entire network's power generation and load is organized into feature vectors on a daily basis. The hourly sequences of photovoltaic output and load at each node are used as clustering features, and the K-means algorithm is employed to... The heavens are divided into Scene clusters The clustering objective is to minimize the sum of intra-cluster distances. (3) In the formula, The number of clusters. For the first A cluster, For the first The source-load joint eigenvector of the sky, For the first The centroid vector of each cluster.

[0029] Step 23: Secondary screening using the thermodynamic free energy criterion. Based on the clustering results of Step 22, the thermodynamic free energy function is introduced to perform a secondary screening of the scene set to eliminate redundant scenes. Define the scene set. The free energy function is: (4) In the formula, For scene collection Scene freedom energy, For intra-class distance energy, For scene distribution entropy, To filter temperature parameters, select those that increase the scene's free energy. The typical day is used as the final scene input. This step outputs a set of typical day scenes. and their corresponding probability weights ( The probability weight of each typical day is calculated based on the proportion of the actual number of days it represents to the total number of days in the year. (5) In the formula, This is a collection of typical daily scenes after screening. The number of typical daily scenarios, For the scene probability weights, For the scene The actual number of days contained in the corresponding cluster. The total number of days in a year, satisfying .

[0030] Step 3: Construct the upper-level planning model; Step 31: Define upper-level decision variables. Assume the distribution network includes... There are candidate nodes, and the decision vector for the upper-level planning is... Defined as: (6) In the formula, This is the upper-level decision vector. For nodes The location status variable for photovoltaic (1 indicates installation, 0 indicates no installation). For nodes The location state variables of energy storage For nodes At the rated capacity of the photovoltaic inverter, For nodes Rated power capacity of energy storage For nodes Energy storage capacity.

[0031] Step 32: Construct a dual objective function. The first objective is to minimize the annualized investment cost: (7) In the formula, The annualized investment cost, This is the investment cost coefficient per unit capacity of photovoltaic power. This is the investment cost coefficient per unit power of energy storage. This is the unit energy investment cost coefficient for energy storage. For the discount rate, This refers to the equipment's service life.

[0032] The second objective is to minimize voltage offset: (8) In the formula, This represents the voltage offset across the entire network. This is a collection of typical daily scenes generated in step 2. For the scene probability weights, This refers to the number of scheduling periods per day. For nodes In the scene time voltage amplitude, This is the reference voltage.

[0033] Step 33: Construct planning constraints. The upper-level decision variables must satisfy the following physical and logical constraints: (9) (10) (11) In the formula, and These represent the lower and upper limits of energy storage capacity, respectively. and These are the lower and upper limits of energy storage capacity, respectively. and These are the lower and upper limits of photovoltaic capacity, respectively. This represents the energy storage power ratio constraint coefficient. These constraints ensure that the planning results comply with equipment manufacturing specifications and installation site limitations.

[0034] Step 4: Construct the optimal operating model for the lower layer; Step 40: Determine the input parameters for the lower-level model. For the scene... Each node in The active power output of photovoltaic systems at the operational level is determined jointly by the planned capacity at the upper level and the interpolated data of the scenario: (12) In the formula, For nodes In the scene time The photovoltaic system is contributing its power. For the location state variables of the upper-level decision, The rated capacity of photovoltaic power is determined by higher-level decisions. For the photovoltaic per-unit output in this scenario obtained by interpolation and reconstruction in step 21, For nodes The rated per-unit output of the photovoltaic system. The reactive power output of the photovoltaic system is limited by the apparent power constraint of the inverter. In the formula, For nodes In the scene time The reactive power output of photovoltaics. The following lower-level model formulas are all for a single scenario. For the sake of brevity, scene subscripts are omitted. .

[0035] Step 41: Define the lower-level objective function. Given the upper-level planning scheme... The next step is to find the energy storage charging and discharging control strategy that minimizes the daily voltage deviation. To adapt to convex optimization solutions, auxiliary variables are introduced. and Eliminate absolute values: (13) (14) In the formula, For nodes exist Auxiliary variable indicating the positive deviation of the voltage from the reference value at any given time. The negative deviation auxiliary variable for voltage below the reference value. For nodes exist The voltage amplitude at time t is determined by Recovery from the squared voltage variable For reference voltage, This represents the number of scheduling periods per day.

[0036] Step 42: Establish node power balance constraints. The active and reactive power balance equations for each node in the distribution network are as follows: (15) (16) In the formula, and They are nodes exist The active and reactive power constantly obtained from the upper-level power grid. For nodes The active power output of the photovoltaic system (determined by step 40). Contributing to photovoltaic power generation without contributing its own energy. For the active power of energy storage discharge, Active power for charging energy storage, For energy storage, reactive power is generated. and They are nodes Active and reactive loads, and Branch roads The trend of meritorious and ineffective actions branch road The square variable of the current, For nodes The set of child nodes For nodes The set of parent nodes, and Branch roads The resistance and reactance.

[0037] Step 43: Establish DistFlow voltage drop constraints. The branch voltage drop equation is as follows: (17) In the formula, For nodes exist The squared voltage change at time t. For nodes exist The squared voltage change at time t. and Branch roads Resistance and reactance, and Branch roads The trend of meritorious and ineffective actions branch road The squared variable of the current.

[0038] The relationship between the square of the branch current and the power flow is as follows: (18) The meanings of the variables in the formula are the same as above. This constraint is a non-convex constraint, which can be transformed into a convex constraint through second-order cone relaxation to ensure the solvability of the model.

[0039] Node voltages must meet safety operating constraints: (19) In the formula, and These are the lower and upper limits for safe operation of node voltages as defined in step 13. For nodes exist The squared voltage variable at time t.

[0040] Step 44: Establish energy storage operation constraints. The state of charge evolution of the energy storage system must satisfy cross-time coupling constraints: (20) In the formula, For nodes Energy storage State of charge at time t, The state of charge at the next moment. For charging efficiency, For discharge efficiency, For charging power, For discharge power, This is the scheduling time step.

[0041] The energy storage state of charge must meet upper and lower limit constraints and daily periodic constraints: (twenty one) (twenty two) In the formula, and These are the lower and upper limits of the permissible state of charge, respectively. For nodes Energy storage capacity and These represent the state of charge at the beginning and end of the scheduling cycle, respectively. The daily periodicity constraint ensures the energy balance during typical daily operation.

[0042] Energy storage charging and discharging power must meet upper and lower limits: (twenty three) In the formula, For charging power, For discharge power, For the rated power capacity of energy storage, This represents the state variable for energy storage location selection in upper-level decision-making. The charging and discharging power cannot be greater than zero simultaneously; this mutual exclusion characteristic can be automatically satisfied through relaxation processing within a convex optimization framework.

[0043] Step 45: The lower-level model calculates the voltage distribution obtained under each typical daily scenario. Feedback to the upper objective function The calculation forms a two-layer closed loop.

[0044] Step 5: Solve using the Riemannian manifold self-organizing evolution algorithm; Step 51: Hybrid Encoding and Chaotic Initialization. Define the first... Hybrid encoding vector of a planning scheme Consistent with the decision vector in step 31. Initial population. The population is generated using a Logistic chaotic mapping. Let the population size be... The maximum number of iterations is The initial evolution temperature was The convergence threshold is For the first 3D decision variables : (twenty four) In the formula, For the first The individual Values ​​of decision variables in dimension These are random numbers generated from a Logistic chaotic sequence. This is the rounding function. and The first Lower and upper bounds for 3D continuous variables Population size. Logistic chaotic sequences are composed of... Iterative generation.

[0045] Step 52: Assessment of the two-layer coupling potential. For any individual in the population... The configuration parameters are passed to the lower-level DistFlow model for solving, and the voltage distribution of each node is obtained. Then calculate the objective function. and We use weighted Chebyshev scalarization to map multi-objective performance into microscopic potential energy: (25) In the formula, For individuals microscopic potential energy, For the first The weight of each objective, For individuals The The objective function value, For the first The ideal point for a goal For the first The Nadir point for each target. Weight vector. ( By uniformly sampling within the interval [0, 1] Each value is generated, and different weights are selected in each iteration to scan the Pareto front. In the formula, This represents the number of weight vectors.

[0046] The probability density function of the population in the solution space follows a Boltzmann distribution: (26) In the formula, For planning scheme In distribution parameters The probability density under, For the partition function, This refers to the evolution temperature.

[0047] Step 53: Fisher Information Matrix and Natural Gradient Update. The population distribution parameter space is considered as a Riemannian manifold, and its local geometry is determined by the Fisher information matrix. definition: (27) In the formula, For Fisher's information matrix, For population distribution parameters, To gradient operator, Let be the mathematical expectation.

[0048] The update of population distribution parameters follows the natural gradient flow: (28) In the formula, For the first The distribution parameters of the generation, For the first The distribution parameters of the generation, To learn step length, For the current generation of Fisher information matrix, The Levenberg-Marquardt damping coefficient is... It is the identity matrix. Let be the desired potential energy. The update formula moves the population center and adjusts the distribution shape along the direction that causes the desired cost and voltage offset to decrease the fastest on the probability distribution manifold.

[0049] Step 54: Helmholtz Free Energy Screening and Temperature Adaptive Control. The Helmholtz free energy of the population system is defined as: (29) In the formula, For the first Generation population The Helmholtz free energy, The population intrinsic energy characterizes the average quality of the current set of proposals. Population entropy, representing the spatial distribution diversity of the scheme. For the first The evolutionary temperature of the generation. Environmental selection strategies screen out those that increase the system's free energy. A subset of that, automatically achieving a balance between convergence and diversity.

[0050] The evolutionary temperature is regulated by a feedback controller based on the entropy production rate: (30) In the formula, For the next generation of evolutionary temperature, The current evolutionary temperature, This is the proportional gain coefficient. For the target entropy generation rate, This represents the current actual entropy generation rate. When the population converges prematurely, When the temperature rises automatically, the weight of the entropy term in the free energy increases, forcing the system to accept solutions with greater diversity in order to escape local optima; when the system enters steady state, the temperature naturally decays, and the algorithm degenerates into a deterministic local search to approximate the Pareto extreme point more precisely.

[0051] Step 55: Determine the convergence condition. When the iterative algebra... Reaching the maximum number of iterations Or the Pareto front change over multiple consecutive generations is less than the convergence threshold. When the iteration terminates, the Pareto optimal solution set is output. In the formula, The maximum number of iterations is set for step 51. This is the convergence threshold.

[0052] Step 6: Output of planning results and evaluation of solutions; Step 61: From the converged Pareto optimal solution set, select the compromise optimal solution according to the decision-maker's preference. .

[0053] Step 62: Output Corresponding photovoltaic site selection and capacity determination results Comparison with energy storage site selection and capacity determination results .

[0054] Step 63: Output the optimal charging and discharging operation strategy of the energy storage system under various typical daily scenarios. and voltage distribution at each node .

[0055] As an example, a standard IEEE-33 node distribution system can be selected for verification. The system's reference voltage is 12.66 kV, and the voltage safety operating constraint range is set at 0.93–1.07 pu. The photovoltaic candidate access nodes are determined based on the voltage sensitivity of the end branches in the network structure. Planning economic parameters are set based on market data, and lithium iron phosphate batteries are used for the energy storage system, with a SOC allowable range of [0.1, 0.9]. Algorithm parameters include population size. Maximum number of iterations Initial evolution temperature Through the complete process of steps 1 to 6 above, the site selection and capacity setting schemes for photovoltaics and energy storage at each node, as well as the energy storage operation strategy, can be obtained, verifying the effectiveness of the proposed method in suppressing voltage deviation, improving power flow distribution, and enhancing planning economics.

[0056] A schematic diagram illustrating the synergistic operation mechanism of photovoltaic net load reduction and energy storage time shift is shown below. Figure 1 As shown in the diagram, the left side represents a scenario with only energy storage configured, where the original load curve is generally high, and the energy storage bears all the energy shift pressure; the right side represents a scenario with photovoltaic and energy storage working together, where the photovoltaic output significantly reduces the net load base, and the energy storage only needs to undertake the time-series peak shaving and valley filling tasks.

[0057] The dual-layer optimization framework diagram of the photovoltaic-storage joint planning is shown below. Figure 2As shown. The upper part is the scene preprocessing flow: raw data → IDW spatial interpolation → K-means clustering → free energy criterion screening → typical daily scene set; the middle part is the upper planning layer: generating / updating planning schemes → calculating bi-objectives → Riemannian manifold self-organizing evolution calculation; the lower part is the lower operation layer: receiving planning schemes → traversing typical days → solving the optimal operation model → outputting voltage indicators → feeding back to the upper layer; the bottom outputs the Pareto optimal solution set.

[0058] The flowchart of the Riemannian manifold self-organizing evolution algorithm is as follows: Figure 3 As shown, starting from the chaotic initialization population, the process proceeds through a two-layer model interface and Riemannian manifold evolution: Fisher information matrix calculation → natural gradient update → Helmholtz free energy screening → entropy generation rate temperature control feedback and self-organizing selection, until the Pareto optimal frontier is output.

[0059] The topology diagram of the distribution network example is as follows: Figure 4 As shown, this displays the node numbers and branch connections of a radial distribution network, and marks the locations of potential photovoltaic (PV) grid connection nodes.

[0060] On the other hand, this embodiment also provides an electronic device, including a memory, a processor, and a computing program stored in the memory and executable on the processor, wherein the processor implements the method when executing the computing program.

[0061] On the other hand, this embodiment also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the method.

[0062] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A two-layer optimization method for distributed photovoltaic-energy storage collaborative planning oriented towards voltage offset suppression, characterized in that, include: Acquire basic data of the power distribution network, including node topology, branch parameters, historical photovoltaic power output data, load data, and equipment economic and technical parameters; Based on the historical photovoltaic power output data and the load data, the photovoltaic power output time series data of each node in the entire network is reconstructed using the inverse distance weighted spatial interpolation method, and the clustered scenarios are further screened based on the thermodynamic free energy criterion to generate a typical daily scenario set. Based on the typical daily scenario set, an upper-level planning model is constructed with annualized investment cost and grid voltage deviation as dual objectives. The decision variables of the upper-level planning model include the site selection status, power capacity and energy capacity of photovoltaic and energy storage. Based on the planning scheme output by the upper-level planning model, a lower-level optimal operation model based on the DistFlow power flow equation is constructed. The lower-level optimal operation model aims to minimize the daily voltage deviation, solves the optimal charging and discharging scheduling strategy for energy storage under each typical daily scenario, and feeds back the obtained voltage performance index to the upper-level planning model. A self-organizing evolution algorithm based on Riemannian manifold dissipative structure is used to solve the upper-level planning model and the lower-level optimal operation model in a coordinated manner to obtain the Pareto optimal solution set; Planning schemes are selected from the Pareto optimal solution set, and the site selection and capacity determination results of photovoltaic and energy storage, as well as the charging and discharging operation strategies of energy storage under various typical daily scenarios, are output.

2. The method according to claim 1, characterized in that, The generation methods for the typical daily scene set include: Based on the geographical distance between each photovoltaic observation station and the node to be estimated, the weight coefficient of each observation station to the node to be estimated is calculated by the inverse distance weighting method, and the photovoltaic output time series data of the node to be estimated is reconstructed accordingly. The reconstructed annual photovoltaic power output time series data and load data are organized into feature vectors by day, and then divided into seasonal scene clusters using the K-means algorithm; A thermodynamic free energy function is introduced to perform a secondary screening of the seasonal scene clusters, and typical days that make the scene free energy increment negative are selected as the final scene set.

3. The method according to claim 1, characterized in that, The construction methods of the upper-level planning model include: The location state variables, power capacity, and energy capacity of photovoltaic and energy storage are used as upper-level decision variables; The annualized investment cost of photovoltaic modules, energy storage converters, and battery packs is taken as the first objective function. The probability-weighted cumulative amount of voltage amplitude deviation from reference voltage at each node at each time under each typical daily scenario is used as the second objective function; Set upper and lower limits for equipment capacity and constraints on energy storage power ratio.

4. The method according to claim 1, characterized in that, The construction methods of the lower-level optimal operating model include: Given the upper-level planning scheme, with the goal of minimizing the daily voltage deviation, positive and negative deviation auxiliary variables are introduced to eliminate absolute values. Establish branch voltage drop constraints and node power balance constraints based on the DistFlow equation; Establish constraints on the energy storage charge state evolution and daily periodicity; The voltage distribution obtained from solving for each typical daily scenario is fed back into the biobjective function of the upper-level planning model.

5. The method according to claim 1, characterized in that, The solution methods for the Riemannian manifold dissipative structure self-organizing evolution algorithm include: A hybrid encoding strategy is adopted to uniformly represent the addressing state variables and the capacity continuous variables, and the initial population is generated using the Logistic chaotic mapping. By using weighted Chebyshev scalarization, the multi-objective performance of an individual is mapped to a microscopic potential energy, so that the population probability density follows a Boltzmann distribution. The Riemannian metric of the population distribution parameter space is defined using the Fisher information matrix, and natural gradient updates with Levenberg-Marquardt damping are performed.

6. The method according to claim 5, characterized in that, The implementation methods of the natural gradient update include: The population distribution parameter space is regarded as a Riemannian manifold, and its local geometric structure is described by the Fisher information matrix. The population distribution parameters are updated according to the direction of the natural gradient flow, so that the population center moves in the direction of the fastest decrease in the desired potential energy and the distribution shape is adjusted.

7. The method according to claim 1, characterized in that, Also includes: The solution set is self-organized and filtered using Helmholtz free energy to select a subset that results in a negative increment of the system's free energy. By adaptively adjusting the evolution temperature using a feedback controller based on entropy generation rate, a balance between convergence and diversity is achieved.

8. The method according to claim 7, characterized in that, The adaptive adjustment method for the evolution temperature includes: When the population converges prematurely, the evolutionary temperature is automatically increased, and the weight of the entropy term in the free energy is increased in order to accept solutions with greater diversity. When the system reaches a steady state, the temperature naturally decays, causing the algorithm to degenerate into a deterministic local search.

9. An electronic device comprising a memory, a processor, and a computing program stored in the memory and executable on the processor, characterized in that, When the processor executes the computing program, it implements the method of any one of claims 1-8.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1-8.

Images