Random optimization method for peak shaving of power distribution network considering three-phase imbalance

Abstract

The application provides a power distribution network peak regulation random optimization method considering three-phase imbalance, and belongs to the technical field of power distribution network control. The method comprises the following steps: establishing a new energy power prediction error probability distribution model, generating a power scene for obtaining possible error values of new energy power, and reducing the power scene; establishing a three-phase power distribution network peak regulation optimization model based on feeder load power control; and simplifying and solving the peak regulation optimization model. The application utilizes feeder load power control technology to coordinate various voltage regulation devices to complete feeder load power control, and comprehensively solves the system peak regulation and imbalance treatment problems. The application focuses on the scene of single-phase access of high-proportion new energy, establishes a multi-objective optimization model based on reactive power voltage control, utilizes feeder load power control technology to reduce the system peak-valley difference in peak-valley time periods, utilizes load regulation characteristics and split-phase voltage optimization to relieve system imbalance in non-peak-valley time periods, and reduces the system peak-valley difference and negative sequence voltage.

Description

Technical Field

[0001] This invention belongs to the field of power distribution network control technology, and specifically relates to a stochastic optimization method for peak shaving in power distribution networks that takes into account three-phase imbalance. Background Technology

[0002] With the increasing installation of distributed renewable energy sources in distribution networks, the volatility and randomness of renewable power generation cause large-scale power flow shifts, resulting in significant peak-to-valley differences in power at distribution network points. Distribution networks with a high proportion of renewable energy face severe peak-shaving problems. Distribution network feeder loads exhibit voltage-power coupling characteristics. Within specified limits, operators control feeder load power by adjusting feeder voltage—this is feeder load power control technology. Compared to existing energy storage and power control methods, feeder load power control technology solves the distribution network peak-shaving problem without requiring additional equipment, offering better power control performance and superior economic efficiency.

[0003] Distribution networks contain many single-phase, asymmetrical loads and devices. Especially when a large number of distributed renewable energy sources are connected in single-phase mode, the system imbalance becomes significantly amplified. To ensure power quality for users, distribution network operation control must consider the three-phase imbalance problem. Furthermore, the implementation process and control effect of feeder load power control are closely related to this imbalance; therefore, system imbalance should be considered when using feeder load power control technology to solve peak-shaving problems in distribution networks. Meanwhile, traditional feeder load power control strategies employ a unified three-phase control mode, where the control quantities for the three-phase voltages are identical. However, in actual distribution networks, different phase sequence voltages are inconsistent. Unified three-phase control can result in excessively low voltages for single phase sequences, limiting the flexibility of feeder load power control technology in three-phase distribution networks and reducing the feeder load power control capacity. Summary of the Invention

[0004] The technical problem to be solved by this invention is to provide a stochastic optimization method for peak shaving in distribution networks that takes into account the shortcomings of existing technologies. This method utilizes feeder load power control technology to coordinate multiple voltage regulating devices to complete feeder load power control, thereby comprehensively solving the problems of peak shaving and imbalance management in the system.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] A stochastic optimization method for peak shaving in a distribution network considering three-phase imbalance includes:

[0007] A probability distribution model for predicting new energy power is established to generate power scenarios for obtaining possible error values ​​of new energy power, and the power scenarios are then reduced.

[0008] A peak-shaving optimization model for a three-phase distribution network based on feeder load power control is established to reduce the negative sequence voltage component of the system while performing peak shaving and valley filling.

[0009] The peak-shaving optimization model is simplified and solved.

[0010] Furthermore, the peak-shaving optimization model uses the reduced power scenario as input and employs a phase-by-phase voltage optimization strategy.

[0011] Furthermore, the peak-shaving optimization model is also used to reduce reactive voltage optimization related to feeder load power control during off-peak periods.

[0012] Furthermore, the method for establishing a probability distribution model for new energy power prediction errors includes: sorting the predicted values ​​according to their magnitude and segmenting the predicted value intervals to obtain prediction bins for different value intervals; for each prediction bin, using a nonparametric distribution model with time-varying characteristics to estimate the theoretical distribution of prediction errors.

[0013] Furthermore, the samples in the prediction bin are sorted in ascending order. The nonparametric distribution model is: in Let be the cumulative distribution function of the power prediction error in the prediction box.

[0014] Furthermore, methods for reducing the power consumption of the scenario include:

[0015] Scene deletion: Based on the L2 distance, find the two closest scenes i and j, delete scene i, and keep scene j;

[0016] Scene correction: Decrease the number of scenes by 1 and modify the probability of scene j to ensure that the sum of the probabilities of all scenes is still 1;

[0017] Scene determination: Determine whether the number of scenes after reduction is equal to the number of target scenes. If so, end the scene reduction; otherwise, perform scene reduction again.

[0018] Furthermore, the method for simplifying and solving the peak shaving optimization model includes: linearizing the peak shaving optimization model.

[0019] Furthermore, the method for linearizing the peak-shaving optimization model includes:

[0020] Linear transformation of nonlinear power flow is achieved using a linear power flow model for unbalanced distribution networks;

[0021] The bilinear nonlinear term is relaxed using the McCormick envelope.

[0022] The absolute value calculation and the maximum / minimum value calculation are linearized by introducing auxiliary decision variables.

[0023] Furthermore, the decision variables of the peak-shaving optimization model include photovoltaic reactive power, capacitor switching status, and voltage regulator output voltage.

[0024] Furthermore, the constraints of the peak shaving optimization model include: power flow constraints, node voltage constraints, capacitor equipment operation constraints, photovoltaic equipment operation constraints, voltage regulator equipment operation constraints, and load voltage-power coupling characteristic constraints.

[0025] The wiring of a distribution network is radial, with feeders as the basic units. Therefore, direct load control at the feeder level can significantly improve regulation capacity and control effectiveness, avoiding the impact of user behavior on regulation capacity. Furthermore, there is a coupling relationship between feeder load power and voltage; therefore, by utilizing the characteristic that load power responds to voltage changes, load power control can be achieved by adjusting the feeder voltage. Feeder load power control technology solves the peak-shaving problem of distribution networks without requiring the installation of additional devices. Compared with traditional energy storage and power supply control methods, feeder load power control technology is a superior choice.

[0026] Feeder load power control is closely related to the three-phase imbalance of the power grid. It is achieved through reactive power and voltage optimization, and imbalance is one of the key issues in voltage optimization. Furthermore, the main cause of system imbalance is the spatiotemporal unevenness of load distribution. Due to the different load compositions in each phase, their voltage-power coupling characteristics are also different. Feeder load power control produces different power regulation effects in each phase, which may further aggravate load asymmetry. Therefore, the implementation process and control effect of feeder load power control are both related to imbalance. When using feeder load power control technology to solve the peak-shaving problem of the distribution network, the system imbalance problem should be considered. Under three-phase imbalance conditions, feeder load power control cannot adopt the traditional three-phase unified control mode; otherwise, it will cause excessively low voltage in a single phase sequence, limiting the flexibility of feeder load power control technology in three-phase distribution networks and reducing the feeder load power control capacity.

[0027] To address the aforementioned issues, this invention proposes a stochastic optimization method for peak shaving in distribution networks that considers three-phase imbalance. By utilizing feeder load power control technology, it coordinates various voltage regulating devices to complete feeder load power control, thereby comprehensively solving the problems of system peak shaving and imbalance management.

[0028] This invention establishes a probability distribution model for the prediction error of new energy power, obtains the predicted value of new energy power, generates power scenarios based on the predicted power value, and completes the reduction of the number of scenarios.

[0029] This invention establishes a peak-shaving optimization model for a three-phase distribution network based on feeder load power control, achieving peak shaving and valley filling while reducing the negative-sequence voltage component of the system. On one hand, the optimization model uses the reduced power scenario as input to ensure that the optimization results meet operational constraints in every possible scenario. On the other hand, the optimization model employs a phase-by-phase voltage optimization strategy to improve the feeder load power control capacity. The optimization model is a day-ahead optimization with an optimization interval of 15 minutes. Decision variables include the phase switching status of capacitors, the phase taps of transformers, and the three-phase reactive power of photovoltaic and wind power inverters within each time period.

[0030] This invention simplifies and solves a peak-shaving optimization model for a three-phase distribution network based on feeder load power control. The model is relaxed and simplified using linearization methods such as the McCormick envelope, and a commercial solver is used to solve the optimization model, determining the three-phase voltage and feeder load three-phase power at each node during different time periods within the second day.

[0031] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0032] This invention focuses on scenarios with a high proportion of single-phase renewable energy access. It establishes a multi-objective optimization model based on reactive power and voltage control. On one hand, during peak-valley periods, it utilizes feeder load power control technology to reduce the system peak-valley difference. On the other hand, during off-peak periods, it leverages load regulation characteristics and phase-specific voltage optimization to alleviate system imbalance, thereby reducing the system peak-valley difference and negative sequence voltage. Simultaneously, this invention employs a feeder load power control strategy based on phase-specific voltage optimization, decoupling the power control constraints of each phase load and improving the feeder load power control capacity and flexibility. Furthermore, it uses finite scenarios to model the uncertainty of renewable energy prediction errors, using the expected system peak-valley difference corresponding to the most likely scenario as the optimization objective, thus improving the robustness of the peak-shaving strategy.

[0033] This invention addresses three-phase unbalanced distribution networks with single-phase access to renewable energy sources. It constructs a peak-shaving optimization model for these networks based on feeder load power control. The model uses peak-valley difference and negative-sequence voltage as optimization objectives, organically combining feeder load power control technology with imbalance mitigation. This effectively reduces system negative-sequence voltage while achieving peak shaving and valley filling. Furthermore, it proposes a feeder load power control strategy based on phase-by-phase voltage optimization. This phase-by-phase optimization decouples three-phase voltage constraints, effectively improving feeder load power control capacity and flexibility. Finally, the optimization model employs a scenario-based strategy to construct a deterministic optimization model that considers the stochastic prediction errors of renewable energy sources, enhancing the robustness of the peak-shaving optimization strategy to these prediction errors.

[0034] This invention addresses the peak-shaving problem in three-phase unbalanced distribution networks based on feeder load power control technology, and constructs a peak-shaving optimization method for distribution networks that considers three-phase imbalance. It has the following advantages:

[0035] 1. Feeder load power control and system imbalance are closely related in both implementation and control effect. The invented optimization method constructs a multi-objective model of the unbalanced distribution network based on reactive voltage optimization, and uses feeder load power control to take into account both peak shaving and valley filling and imbalance management. 2. The invented optimization method adopts a feeder load power control strategy based on phase voltage optimization. By decoupling the voltage constraints of each phase, it enhances the flexibility and capacity of load power control. 3. The invented optimization method uses a finite but comprehensive power scenario to characterize the uncertainty of new energy sources, transforming the stochastic optimization model into a deterministic model, reducing the model solution complexity, and improving the robustness of the peak shaving strategy. Attached Figure Description

[0036] The present invention will now be described in further detail with reference to the accompanying drawings.

[0037] Figure 1 : Flowchart of the optimization method of this invention;

[0038] Figure 2 : A flowchart of the error modeling steps for new energy power prediction in this invention;

[0039] Figure 3 : A schematic diagram of photovoltaic prediction error modeling using a prediction box in this invention;

[0040] Figure 4 : A schematic diagram of the process for generating new energy power prediction error scenarios in this invention. Detailed Implementation

[0041] To better understand the present invention, the content of the invention is further clearly illustrated below with reference to embodiments and accompanying drawings. However, the scope of protection of the present invention is not limited to the embodiments described below. Numerous specific details are set forth in the following description to provide a more thorough understanding of the invention. However, it will be apparent to those skilled in the art that the present invention can be practiced without one or more of these details.

[0042] See Figures 1-4 The purpose of this embodiment is to provide a stochastic optimization method for peak shaving in distribution networks that takes into account three-phase imbalance.

[0043] like Figure 1 As shown, the optimization method includes the following steps:

[0044] S1: Modeling the error in predicting new energy power.

[0045] This step mainly involves obtaining the predicted power value of new energy sources, generating power scenarios based on the predicted power value, and reducing the number of scenarios.

[0046] Whether using direct forecasting methods based on historical data and numerical weather prediction, or indirect forecasting methods based on irradiance and wind speed predictions, the day-ahead power forecasting of new energy sources generally suffers from errors. Peak-shaving optimization methods utilize new energy inverters for reactive power and voltage optimization. The reactive power of new energy inverters is affected by the predicted active power of the inverters. To improve the feasibility of the optimization model, the impact of new energy power forecasting errors should be considered.

[0047] Specifically, this step includes:

[0048] S11, Establish a probability distribution model for the prediction error of new energy power.

[0049] Power prediction for new energy sources, such as photovoltaics, is often done through point prediction. Using a deterministic point prediction method, when the amount of prediction data is large enough, in addition to the statistical characteristics of photovoltaic power, the prediction error will also follow a certain probability distribution. If the error distribution of each point prediction value can be obtained, the conditional probability distribution of the possible photovoltaic power can be obtained given that the point prediction value is known.

[0050] However, calculating the error distribution of each predicted value individually is quite difficult. To reduce the computational burden while ensuring accuracy, a "prediction bin" is used to classify the predicted values. First, the predicted values ​​are sorted according to their magnitude, and the predicted value intervals are segmented to obtain prediction bins for different value intervals. For example, when the value interval is 0.02 pu, 50 different prediction bins can be obtained. Based on the magnitude of the predicted values, the corresponding [predicted value, error value] data are classified into different prediction bins. The predicted values ​​within the same prediction bin are very similar, but the error values ​​vary greatly. By statistically analyzing the probability distribution of the error values ​​within each prediction bin, the probability distribution of the possible prediction error of photovoltaic power within that prediction interval can be obtained.

[0051] For the probability distribution model of the error value within each prediction bin, a certain distribution model can be used for fitting. However, the use of a deterministic distribution model has limitations. For example, the normal distribution fits well in summer, while the beta distribution may be better in winter. Therefore, this invention uses a nonparametric distribution model with time-varying characteristics to estimate the theoretical distribution of the prediction error. The nonparametric distribution model is based on Grievance's theorem, that is, when the sample size is large, the empirical distribution of the sample approaches the population distribution, as shown in Equation (1).

[0052]

[0053] In the formula, P(x) is the probability of x; F(x) is the empirical distribution function; F(x) is the actual distribution function.

[0054] For the prediction error data in the prediction bin, arrange them in monotonically increasing order to obtain the sample. Define the following nonparametric empirical distribution model:

[0055]

[0056]

[0057] In the formula, Let be the cumulative distribution function of the power prediction error in the prediction box.

[0058] For each prediction box, using the prediction error sequence, the nonparametric empirical distribution model of the prediction error interval can be quickly obtained through equations (2) and (3). When the sample size N approaches infinity, the empirical distribution function uniformly approaches the actual distribution function with probability 1. The photovoltaic prediction error model using prediction boxes is as follows: Figure 3 As shown.

[0059] S12, Generation of new energy power prediction error scenario.

[0060] Non-parametric empirical distribution models cannot directly participate in peak shaving optimization models and need to obtain the possible error values ​​of new energy power through scenario generation. Combining inverse transform sampling and normal distribution function, corresponding new energy power prediction error scenarios are generated for different prediction boxes. Inverse transform sampling technology belongs to Monte Carlo sampling, which is simple to calculate and has high sampling efficiency, and is widely used for stochastic modeling of photovoltaic and wind power. The process of inverse transform sampling is as follows: generate a random number z in the range of [0,1], and assume that z is the cumulative distribution function value of photovoltaic prediction error distribution. Using the inverse function of the cumulative distribution function of prediction error, as shown in Equation (4), the possible error of the corresponding photovoltaic power prediction value is obtained according to z, and a single scenario generation is completed.

[0061]

[0062] In the formula, z is a random number in the range [0,1]; P is the photovoltaic prediction error scenario value corresponding to the random number z.

[0063] Choosing the right random number z is crucial for scene quality. To ensure scene comprehensiveness, the random number z should uniformly cover the numerical range [0,1]. Random numbers y are generated using a random variable Y that follows a standard normal distribution, as shown in equation (5), and the cumulative distribution function value corresponding to the random number y is used as the random number z, as shown in equation (6). When a large number of random numbers y are generated, uniform random numbers z can be obtained, thus yielding a series of photovoltaic power scenes.

[0064]

[0065]

[0066] Scene generation strategies based on inverse transform sampling and normal distribution function, such as... Figure 4 As shown. For example, first, a random number is obtained using the standard normal distribution, such as 0. Then, its corresponding cumulative distribution function value is calculated to be 0.5 using equation (5). Figure 4 As shown in (b); then 0.5 is taken as input and set as the cumulative distribution function value of the prediction error, as follows. Figure 4 As shown in (a), the error value of the corresponding predicted power is 0.21pu obtained by equation (6).

[0067] S13, Reduction of prediction error scenarios for new energy power.

[0068] To fully describe the possible scenarios of new energy power, the number of prediction error scenarios is often large, which significantly increases the complexity of solving the optimization problem. Furthermore, a large number of similar samples exist within these massive scenarios, resulting in high redundancy. Therefore, scenario reduction is necessary, using a smaller number of scenarios to approximate the statistical distribution of the original large number of scenarios. The classic synchronous back-substitution elimination method is employed for scenario reduction. This method uses the L2 norm to quantitatively describe the distance between two scenarios. Each iteration finds two scenarios with low probability of occurrence and close distance, merging them to complete scenario reduction. This process is iterated until the required number of scenarios is reached.

[0069] Assume there are initially N scenes {P1, P2, ..., P...} N Each initial scenario has the same probability of occurrence, π = 1 / N, and satisfies ∑ i∈N π i =1, the target number of scenes is M. The scene reduction process is as follows:

[0070] S131, Scene deletion: Based on the L2 distance, use Equation (7) to find the two closest scenes i and j, delete scene i, and keep scene j.

[0071]

[0072] Scene deletion only removes scenes most closely related to others, while scenes significantly different from others are retained, ensuring that the final scene retains more important information. In addition to scene distance, scene deletion also considers the probability of scene occurrence, prioritizing the deletion of scenes with low probability of occurrence and those that are not representative.

[0073] S132, Scene Correction: Decrease the number of scenes by 1, N←N-1; Then, modify the probability of scene j occurring, π j ←π j +π i This ensures that the sum of the probabilities of all scenarios is still 1.

[0074] S133, Scene Judgment: Determine whether the number of scenes N after reduction is equal to M. If yes, end the scene reduction program; otherwise, return to step S131 and continue the scene reduction.

[0075] S2: Construct an optimization model for peak shaving in an unbalanced distribution network.

[0076] This step establishes a three-phase distribution network peak-shaving optimization model based on feeder load power control, reducing the negative sequence voltage component of the system while performing peak shaving and valley filling. On the one hand, the optimization model uses the reduced power scenario as input to ensure that the optimization results meet the operational constraints in every possible scenario; on the other hand, the optimization model adopts a phase-by-phase voltage optimization strategy to improve the feeder load power control capacity.

[0077] The objective function of the peak shaving optimization model is shown in equation (8).

[0078]

[0079] In the formula, π s Let s be the probability of scenario s occurring. For scenario s, the peak power of the distribution network points under the next day. For scenario s, the valley power of the distribution network points in the next day. Let e ​​be the active power of the distribution network points within time period t under scenario s, and be the sum of the three-phase power; i,t,s,- with f i,t,s,- These are the real and imaginary parts of the negative sequence voltage of node i in scenario s during time period t, respectively. For the photovoltaic at node i within time period t Phase reactive power; For the capacitor at node i during time interval t Phase switching state; V0 is the rated voltage; For the i-node voltage regulator within the time period t Phase output voltage; β1-β5 are weighting coefficients. S is the photovoltaic power scenario set; T is the time set; N PV N is a set containing photovoltaic nodes; CB Let N be the set of nodes containing capacitors; VT It is a set of nodes containing voltage regulators.

[0080] Considering the impact of prediction errors, the objective function aims to minimize the system peak-to-valley difference corresponding to the most likely scenario (the first term of the objective function). The total peak-to-valley difference, after probability weighting, represents the expected system peak-to-valley difference under the condition of errors in renewable energy prediction, which can improve the accuracy of the optimization strategy. Simultaneously, since feeder load power control and phase voltage optimization may exacerbate system imbalance, the second term of the objective function reduces the system negative-sequence voltage to ensure power quality for users. Furthermore, to reduce the potential impact of feeder load power control on users, the last three terms of the objective function are used to reduce reactive voltage optimization related to feeder load power control during off-peak periods. Since the effect of feeder load power control is the same in all scenarios, the operation of reactive power regulating equipment is independent of the scenario. For simplicity, this invention only considers photovoltaic power, but renewable energy sources such as wind turbines can also be included in this optimization model. The decision variables of the optimization model include photovoltaic reactive power Q. PV Capacitor switching state X CB and the voltage regulator output voltage V out .

[0081] The power flow constraints of the peak shaving optimization model are shown in equations (9) and (10).

[0082]

[0083]

[0084] In the formula, and For each distribution network substation in scenario s within time period t, the substations are respectively... Active and reactive power are ignored when the node is not a downstream point. For the photovoltaic at node i in scenario s within time period t Phase active power, For the photovoltaic at node i within time period t Phase reactive power; this variable is ignored when there is no photovoltaic power at the node. For the capacitor at node i during time interval t Phase reactive power, this variable is ignored when there are no capacitors at the nodes; and For each node i, the load is s within the time period t. Active power and reactive power; and For each node i in the time period t, the scenario s is as follows: Real and imaginary parts of phase voltage; and These are the nodes i and j respectively. The real and imaginary parts of the phase line admittance.

[0085] The node voltage constraints of the peak shaving optimization model are shown in Equation (11).

[0086]

[0087] In the formula, For node i in scenario s within time period t Phase voltage; V max With V min These represent the maximum and minimum voltage values, respectively. According to national standards, V in a medium-voltage distribution network... max With V min They are 0.93 pu and 1.07 pu respectively.

[0088] The operating constraints of the capacitor bank in the peak-shaving optimization model are shown in equations (12) and (13). The reactive power of each phase of the capacitor is related to its switching state. At the same time, it is necessary to limit the maximum number of switching changes per day for each phase of the capacitor.

[0089]

[0090]

[0091] In the formula, For the capacitor at node i during time interval t Phase reactive power; For the i-node capacitor Phase rated reactive power capacity; N is the threshold for the maximum number of daily switching changes for capacitors; CB This is the set of capacitor nodes.

[0092] The photovoltaic equipment operation constraints of the peak-shaving optimization model are shown in equations (14)-(17). For PQ control mode photovoltaics, it is necessary to constrain the output power of each phase; for PV control mode photovoltaics, in addition to the output power, it is also necessary to constrain the grid connection point voltage. At the same time, the active power of photovoltaics in each time period is equal to the scenario power.

[0093]

[0094]

[0095]

[0096]

[0097] In the formula, For node i The reactive power of the photovoltaic system during time period t; and For each node i Maximum and minimum values ​​of photovoltaic reactive power; For node i Phase photovoltaic rated capacity; For the photovoltaic system at node i in scenario s within time period t. Phase active power; For the photovoltaic system at node i in scenario s within time period t. Phase prediction power; For the photovoltaic at node i within time period t Phase parallel grid point voltage; V PV,max With V PV,min These represent the maximum and minimum voltage values ​​at the photovoltaic grid connection point, respectively; N PV It is a set of photovoltaic nodes.

[0098] The operating constraints of the voltage regulator equipment in the peak-shaving optimization model are shown in equations (18)-(20). Transformer taps are used for feeder voltage control. Transformer taps support phase-by-phase control, but the maximum number of daily operations per phase is limited. It should be noted that other equipment such as dynamic voltage restorers and intelligent transformers can replace transformer taps for feeder voltage control, but their respective operating constraints must be met.

[0099]

[0100]

[0101]

[0102] In the formula, and The tap changers of the transformer at node i within time period t are respectively Phase input voltage and output voltage; For the tap changer of the i-node transformer within the time period t Phase stop; V tap This refers to the voltage variation at each tap of the transformer. and These represent the maximum and minimum values ​​for the tap position of the transformer at node i; N represents the threshold for the maximum number of tap changes per day for a transformer. OLTC This is the set of transformer tap changer nodes.

[0103] The load voltage-power coupling characteristic constraints of the peak shaving optimization model are shown in equations (21) and (22).

[0104]

[0105]

[0106] In the formula, and For each node i, the load is s within the time period t. Active power and reactive power; and For each node i The initial active and reactive power of the phase load during time period t; and For node i in scenario s within time period t, respectively Phase voltage and initial voltage; and The load of node i within time period t is respectively Phase voltage-active power factor and voltage-reactive power factor.

[0107] S3: Simplify and solve the peak-shaving optimization model.

[0108] This step simplifies and solves the peak-shaving optimization model of a three-phase distribution network based on feeder load power control. The peak-shaving optimization model for an unbalanced distribution network is a mixed-integer nonlinear programming problem, requiring linearization. The peak-shaving optimization model involves nonlinear terms including power flow, bilinear nonlinear terms, absolute value calculations, and extremum calculations.

[0109] For nonlinear power flow, a linear transformation can be achieved using an unbalanced distribution network linear power flow model, such as the Distflow power flow model.

[0110] For bilinear nonlinear terms caused by optimizing transformer tap changer output voltage and inverter reactive power, as well as quadratic terms such as the real and imaginary parts of negative sequence voltage in the objective function, convex relaxation can be performed using the McCormick envelope.

[0111] For absolute value and maximum / minimum value calculations, linearization can be achieved by introducing auxiliary decision variables.

[0112] Through the above simplification and relaxation, the original mixed-integer nonlinear optimization model can be transformed into a mixed-integer linearization problem, which can be solved directly and quickly using existing commercial solvers, such as the CPLEX solver.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.

Claims

1. A stochastic optimization method for peak shaving in a distribution network considering three-phase imbalance, characterized in that: include: The predicted values ​​of each point of new energy power are sorted and the predicted value intervals are segmented to obtain prediction bins for different value intervals; for each prediction bin, a nonparametric distribution model with time-varying characteristics is used to estimate the theoretical distribution of prediction error and establish a probability distribution model of new energy power prediction error. By combining inverse transform sampling and normal distribution function, power scenarios are generated for different prediction boxes to obtain possible errors in the predicted value of new energy power. Reduce the power in the scenario described; A peak-shaving optimization model for a three-phase distribution network based on feeder load power control is established to reduce the negative sequence voltage component of the system while performing peak shaving and valley filling. The objective function of the peak-shaving optimization model takes minimizing the peak-valley difference of the system corresponding to the most likely scenario as the optimization objective. The objective function also includes a negative sequence voltage term to reduce the negative sequence voltage of the system, as well as a photovoltaic reactive power term, a capacitor switching status term, and a voltage regulator output voltage term to reduce the reactive voltage related to feeder load power control during non-peak and valley periods. The peak-shaving optimization model is simplified and solved.

2. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 1, characterized in that: The peak-shaving optimization model uses the reduced power scenario as input and employs a phase-by-phase voltage optimization strategy.

3. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 1, characterized in that: The samples in the prediction bin are sorted in ascending order. The nonparametric distribution model is: , ;in Let N be the cumulative distribution function of the power prediction error in the prediction box, and N be the sample size. Used to determine whether the target value x is greater than or equal to the sample value. .

4. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 1, characterized in that: Methods for reducing the power of the scenario include: Scene deletion: Based on the L2 distance, find the two closest scenes i and j, delete scene i, and keep scene j; Scene correction: Decrease the number of scenes by 1 and modify the probability of scene j to ensure that the sum of the probabilities of all scenes is still 1; Scene determination: Determine whether the number of scenes after reduction is equal to the number of target scenes. If so, end the scene reduction; otherwise, perform scene reduction again.

5. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 1, characterized in that: The method for simplifying and solving the peak shaving optimization model includes: linearizing the peak shaving optimization model.

6. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 5, characterized in that: The method for linearizing the peak-shaving optimization model includes: Linear transformation of nonlinear power flow is achieved using a linear power flow model for unbalanced distribution networks; The bilinear nonlinear term is relaxed using the McCormick envelope. The absolute value calculation and the maximum / minimum value calculation are linearized by introducing auxiliary decision variables.

7. The stochastic optimization method for peak shaving in distribution networks considering three-phase imbalance as described in claim 1, characterized in that: The constraints of the peak shaving optimization model include: power flow constraints, node voltage constraints, capacitor bank operation constraints, photovoltaic bank operation constraints, voltage regulator operation constraints, and load voltage-power coupling characteristic constraints.

Images