A method for characterizing the elliptical operating domain of a flexible power distribution system considering soft switching.
By employing a sub-Bluerstein bar elliptical operating domain characterization method for flexible distribution systems, combined with Wasserstein sub-Bluerstein bar optimization and LSTM models, the quantitative problems of photovoltaic prediction error distribution offset and the impact of flexible soft switching are solved. This improves the accuracy and flexibility of operating domain assessment for distribution networks, reduces risks, and optimizes the operating domain boundary.
Patent Information
- Application Number
- CN202611267147.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing methods for evaluating the operational domain of distribution networks are insufficient to accurately characterize the shift in photovoltaic prediction error distribution and operational risks under limited sample conditions. In particular, under the influence of photovoltaic uncertainties in different weather conditions, the quantitative assessment of the flexible operation capability of distribution networks by flexible soft switching is inadequate.
A method for characterizing the elliptical operating domain of a flexible distribution system considering soft switching is adopted. By combining Wasserstein elliptical bar optimization and root node PQ elliptical operating domain representation with photovoltaic prediction error modeling and flexible resource constraints of the distribution network, a two-stage optimization model is constructed. The LSTM model is used to predict the photovoltaic output error and construct the elliptical operating domain. The error distribution offset is described by Wasserstein fuzzy set, and the operating domain is optimized by column and constraint generation algorithm.
It improves the granularity of operational domain assessment, reduces the risk of overly optimistic or conservative operational domain boundaries, quantifies the impact of flexible soft-switching on operational domain area, voltage level, and branch load, and enhances the flexibility and security of the distribution network.
Smart Images

Figure CN123121792A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of power distribution network operation domain assessment and new energy uncertainty optimization technology, and particularly relates to a method for characterizing the elliptical operation domain of a flexible power distribution system considering soft switching. Background Technology
[0002] As the proportion of distributed photovoltaic (PV) power integrated into distribution networks increases, PV output becomes more random and volatile due to factors such as weather conditions, cloud cover, irradiance variations, and diurnal cycles. PV forecasting errors can alter node power injection, branch power flow, and node voltage distribution, thereby affecting the active and reactive power exchange capacity between the distribution network and the upstream grid.
[0003] Existing power distribution network operation domain assessment methods typically describe photovoltaic uncertainties using typical scenarios, confidence intervals, or deterministic boundaries, making it difficult to characterize the deviation of the true error distribution from the empirical distribution under limited sample conditions. Existing elliptical operation domain methods focus more on boundary calculations under deterministic or limited scenarios, and do not adequately consider the differences in photovoltaic prediction error distribution under different weather conditions and the worst-case operation risk.
[0004] Flexible switching can improve the voltage level, branch load and renewable energy acceptance capacity of the distribution network through active power transfer and reactive power support between feeders, but it is necessary to quantify its impact on the flexible operation capability of the distribution network from the perspective of the root node PQ operating domain under the condition of uncertain photovoltaic distribution. Summary of the Invention
[0005] The purpose of this invention is to address the aforementioned technical problems by providing a method for characterizing the elliptical operating domain of a flexible distribution system that considers soft switching. This method combines photovoltaic prediction error modeling under weather condition differences, Wasserstein elliptical operating domain optimization, root node PQ elliptical operating domain characterization, and flexible resource constraints of the distribution network to characterize the impact of photovoltaic error distribution shift on the operating domain boundary and operating risk under limited sample conditions.
[0006] Technical Solution: To achieve the above-mentioned objectives, this invention proposes a method for characterizing the elliptical operating domain of a flexible power distribution system considering soft switching. This method includes the following steps:
[0007] Step 1: Obtain the network topology parameters, electrical parameters, root node information, photovoltaic unit parameters, energy storage system parameters, distributed power generation parameters, flexible soft switching parameters, historical photovoltaic output data, and historical meteorological data of the target distribution network. The evaluation and scheduling cycle consists of 24 consecutive 1-hour time periods.
[0008] Step 2: Normalize the historical photovoltaic power output data and historical meteorological data, classify the weather conditions as sunny, cloudy and overcast based on the relative irradiance level and its intraday fluctuation characteristics, and construct the corresponding periodic features for hours, days within the year and months.
[0009] Step 3: Using the historical sequence of the past 72 hours as the input window for the LSTM photovoltaic power prediction model in each prediction period, perform hourly rolling predictions for 24 consecutive 1-hour periods within the scheduling cycle to be evaluated; compare the predicted output of each period with the corresponding actual output, combine the 24 consecutive periods to form a 24-dimensional photovoltaic prediction error scenario vector, and construct photovoltaic prediction error scenario sets corresponding to sunny, cloudy, and overcast days according to the weather conditions obtained in Step 2;
[0010] Step 4: For the projection of the high-dimensional feasible set of the distribution network onto the active-reactive power plane of the root node, the PQ operating domain of the root node is parameterized by an ellipse formed by translation, rotation and scaling of the unit circle. The center of the ellipse and the positive semi-definite matrix are used as the operating domain parameters to be optimized. Further, the vertex is constructed by constructing a finite external approximation polygon containing the unit circle, and the finite vertices on the PQ plane of the root node are obtained by mapping the ellipse center and the shape matrix, which provides input for the establishment of the elliptical approximation operating domain model of the distribution network in Step 5.
[0011] Step 5: Based on the vertices of the approximate polygon outside the ellipse formed in Step 4, establish a distribution network scenario operation model. The model consists of distribution network operation equations, network security constraints, and flexible resource operation constraints. The distribution network operation equations include node active power balance, node reactive power balance, node net injected power composition, and linearized branch voltage drop equations. Network security constraints include node voltage upper and lower limits and branch capacity constraints. Flexible resource operation constraints include photovoltaic, energy storage systems, distributed power sources, flexible soft switching, and load shedding constraints.
[0012] Step 6: Based on the 24-dimensional photovoltaic prediction error scenario set obtained in Step 3, construct the empirical distribution, scenario distance, Wasserstein radius, and Wasserstein fuzzy set. Combined with the candidate elliptical operating domain established in Step 4 and the distribution network elliptical approximate operating domain model established in Step 5, construct a two-stage Wasserstein sparse bar elliptical operating domain optimization model. In the first stage, determine the ellipse center and ellipse shape matrix for each time period. In the second stage, adjust the distribution network operation for different photovoltaic prediction error scenarios to expand the elliptical operating domain of the root node PQ and reduce the operating risk corresponding to the worst probability distribution within the Wasserstein fuzzy set.
[0013] Step 7: Use the column and constraint generation algorithm to solve the two-stage Wasserstein sub-Brubar elliptical running domain optimization model established in Step 6; determine the candidate elliptical running domain and sub-Brubar risk upper bound under the current set of active error scenarios through the main problem; then check all candidate error scenarios through the second-stage running sub-problems and identify the unfavorable scenario that generates the maximum risk constraint violation; iteratively add the scenario to the main problem until the maximum violation does not exceed the preset convergence threshold; and output the root node PQ sub-Brubar elliptical running domain result.
[0014] Further, step 2 includes: performing maximum and minimum normalization on photovoltaic power, total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure; this implementation method, based on the existing weather classification concept of "relative irradiance level + fluctuation degree," uses the 90th percentile irradiance of the effective sunshine sample of the same month and hour as the data-driven reference irradiance, and then calculates the daily relative total solar irradiance mean and standard deviation for classification. Based on this, weather conditions are divided into sunny, cloudy, and overcast; and sinusoidal and cosine periodic characteristics are constructed for hours, months, and days within the year.
[0015] Furthermore, in step 2, the maximum-minimum normalization process and the construction of the time period features are performed according to the following formula:
[0016]
[0017]
[0018] In the formula, This represents the normalized variable value; This represents the original variable value before normalization; This represents the minimum value of the variable in the training samples; This represents the maximum value of the variable in the training samples; Indicates time period Time-period feature vector; Indicates time period The corresponding hour number; Indicates time period The corresponding day number within the year; Indicates time period The corresponding month number.
[0019] Furthermore, in step 3, the LSTM photovoltaic power prediction model uses its previous 72-hour historical sequence as the input window for each prediction period, where 72 hours represents the historical review length of a single prediction; the evaluation scheduling period consists of 24 consecutive 1-hour periods, where 24 hours represents the time dimension of the error scenario vector. Predictions are performed sequentially for periods 1 to 24 within the evaluation period, and the input window is shifted forward by 72 hours hourly with each prediction time, thus forming a photovoltaic prediction sequence of 24 consecutive periods; the historical sequence includes at least the normalized photovoltaic output, irradiance, ambient temperature, air pressure, and time period characteristics of the previous hour; the LSTM photovoltaic power prediction model includes a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a ReLU fully connected layer, and a linear output layer connected in sequence, and the normalized photovoltaic prediction value is obtained based on the LSTM gated state update; the 24-hour photovoltaic prediction error scenario set consists of multiple error scenario vectors, each of which includes the photovoltaic prediction error of 24 consecutive periods.
[0020] 3.1 Further, in step 3, the LSTM photovoltaic power prediction, inverse normalization, prediction error, error scenario vector, and available photovoltaic output are determined by the following formula:
[0021]
[0022]
[0023]
[0024]
[0025]
[0026] In the formula, This indicates a scenario of photovoltaic prediction error. This represents the set of candidate photovoltaic prediction error scenarios; This represents the normalized photovoltaic prediction value output by the LSTM. This represents a nonlinear mapping constructed from an LSTM network. This represents the trainable parameters of the LSTM model; This represents the historical sequence features input to the LSTM; Indicates time period The projected output of photovoltaic power; This represents the maximum photovoltaic output in the training samples; This represents the minimum photovoltaic output in the training samples; Indicates time period The photovoltaic prediction error; Indicates time period The actual output of photovoltaic power; Indicates the first A 24-hour photovoltaic forecast error scenario vector; superscript Indicates vector transpose; Indicates the first The scene in time period The photovoltaic prediction error; Indicates the first One error scenario in the time period The photovoltaic power available for use; This indicates the apparent power capacity of photovoltaics.
[0027] 3.2 The historical sequence input, input gate, forget gate, and output gate calculation of the LSTM photovoltaic power prediction model, candidate cell state and cell state update, Dropout random inactivation, ReLU fully connected layer, linear output layer, training loss, and continuous multi-time recursive input relationship are expanded by the following formula:
[0028]
[0029]
[0030]
[0031]
[0032]
[0033]
[0034]
[0035]
[0036]
[0037]
[0038]
[0039]
[0040]
[0041]
[0042]
[0043] In the formula, Indicates the length of the history window; Indicates time period Single-time feature vector, superscript This indicates that the corresponding variable has undergone normalization. This represents the normalized photovoltaic output for the previous hour. Indicates time period Normalized actual photovoltaic output This represents a normalized meteorological characteristic vector, which includes at least total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure; when meteorological data for future periods are unavailable. It can be set to an empty vector or replaced by weather state forecasts. Indicates the LSTM layer number. Corresponding to the first LSTM layer, Corresponding to the second LSTM layer; Indicates the first Layer in time period Input, This indicates a hidden state. Indicates cell state; , , These represent the input gate, forget gate, and output gate, respectively. Indicates the state of candidate cells; and This represents the input weight matrix and the cyclic weight matrix used in the corresponding calculations. Denotes the corresponding calculated bias vector, where, Pick , , and The time parameters correspond to the calculation parameters of the input gate, forget gate, output gate, and candidate cell state, respectively; where the cell state is the internal memory variable used by LSTM to store and transmit historical information in the time series. This represents the Sigmoid function. Represents the hyperbolic tangent function. Represents the Hadamard product; and This represents the Dropout operator. and Indicates the discard rate. This indicates that the first layer of LSTM is in the time period The output hidden state is discarded at a rate of The feature vector obtained after processing by the first Dropout layer is used as the input of the second LSTM layer; This indicates that the second layer of LSTM is in the time period The output hidden state is discarded at a rate of The feature vector obtained after processing by the second Dropout layer is used as the input to the subsequent ReLU fully connected layer; This represents a linear rectified function; random deactivation is performed during the training phase and disabled during the prediction phase. This represents the hidden state of the second LSTM layer output after Dropout; This indicates the output of the ReLU fully connected layer. , This represents the weights and biases of the ReLU fully connected layer. , This represents the weights and biases of the linear output layer; Indicates time periods in the training samples Normalized actual photovoltaic output Indicates time period The corresponding Huber loss, This represents the Huber threshold. This represents the training loss of LSTM. This represents the set of target time period indices corresponding to the training samples of the LSTM model; This indicates the number of training samples in this set; The prediction step number is indicated by the continuous rolling prediction. The Adam optimizer is used to train the parameters in the LSTM layer, the fully connected layer and the output layer. Formula (22) represents the recursive input relationship during continuous prediction, that is, the normalized prediction value of the previous time period is used as the photovoltaic power output feature of the previous hour in the next time period.
[0044] Furthermore, step 4 includes:
[0045] 4.1 For each scheduling period within the 24-hour evaluation scheduling cycle Using the active and reactive power of the root node to form a two-dimensional power vector, the candidate elliptical operating domains of the root node PQ for each time period are constructed according to equation (23):
[0046]
[0047] Among them, the two-dimensional symmetric shape matrix Represented as:
[0048]
[0049] In the formula, The root node PQ represents the elliptical running domain of a continuous time period; Represents the Cartesian product of the elliptical running domain of the root node PQ at each time period; , They represent time periods respectively. Active and reactive power at the center of the ellipse; A positive semi-definite two-dimensional matrix representing the shape of an ellipse; Indicates the set of evaluation periods; Indicates the scheduling period; , They represent time periods respectively. The active and reactive power of the root node; , , and Representing shape matrices respectively The corresponding matrix elements; Represents a two-dimensional reference vector within the unit circle; The second norm of a vector;
[0050] 4.2 Construct the finite number of vertices of the outer approximation polygon of the unit circle, setting the total number of vertices of the outer approximation polygon to be... , No. The vertices of the approximate polygon are denoted as _____. And determine its two coordinate components according to equation (25):
[0051]
[0052] In the formula, Indicates the first Two-dimensional coordinate vectors of the vertices of an approximate polygon; and They represent the first The coordinate components of the vertices of an external approximate polygon in the directions of the first and second coordinate axes; Indicates the vertex number of the external approximate polygon.
[0053] 4.3 Map the vertices of the outer approximation polygon to the root node PQ plane, for each scheduling period. and the vertices of each external approximate polygon Using the ellipse center and shape matrix determined in step 4.1, the reference vertex obtained in step 4.2 is... Mapping to the root node PQ plane, the corresponding root node PQ verification vertex is obtained according to equation (26):
[0054]
[0055] In the formula, , They represent time periods respectively. Next The active and reactive power of the root node are obtained by mapping the vertices of an external approximation polygon. The active and reactive power generated by each scheduling period... Each root node PQ verification vertex serves as the root node power input for establishing the distribution network scenario operation model in step 5. Distribution network operation equations, network security constraints, and photovoltaic, energy storage system, distributed power source, flexible soft switch and load shedding operation constraints are applied to each verification vertex.
[0056] Furthermore, in step 5, a power distribution network scenario operation model is established for the vertices of the approximate polygon outside the ellipse. This model consists of three parts: "power distribution network operation equations," "network security constraints," and "flexible resource operation constraints." The power distribution network operation equations describe the physical relationships between nodes and branches; network security constraints limit node voltage and branch capacity; and flexible resource operation constraints define the adjustable range of photovoltaic systems, energy storage systems, distributed power sources, flexible soft switching, and load shedding.
[0057] The distribution network operation equations described in 5.1 consist of node power balance, node net injection, and linearized branch voltage drop:
[0058]
[0059]
[0060]
[0061]
[0062]
[0063] 5.2 The network security constraints described herein consist of node voltage upper and lower limits and branch apparent power capacity constraints:
[0064]
[0065]
[0066] In the formula, Represents the set of nodes in the distribution network. Indicates the node index; Represents the set of branches in the distribution network. Indicates the connection node and nodes The side road; , Representing nodes respectively Net active power and net reactive power injection; , Representing branch roads The active and reactive power flowing through; , Representing nodes respectively During the period Ellipse boundary points and error scenarios The photovoltaic system outputs both active and reactive power. , These represent the energy storage discharge and charging power, respectively. , These represent the active and reactive power outputs of the distributed power source, respectively. , These represent the flexible soft-switching injection nodes. Active and reactive power; , These represent the active and reactive load shedding quantities at the node, respectively. , These represent the active and reactive loads of the node, respectively. Represents a node Voltage squared, , Representing branch roads Resistance and reactance, , These represent the lower and upper limits of the square of the node voltage, respectively. Indicates the maximum capacity of the branch;
[0067] 5.3 Photovoltaic operation constraints consist of available active power, available output after prediction error correction, inverter capacity, and reactive power regulation range:
[0068]
[0069]
[0070]
[0071]
[0072] 5.4 The operational constraints of energy storage systems consist of upper and lower limits for charge and discharge power, limitations on charge and discharge modes, energy state updates, and limitations on remaining energy margin.
[0073]
[0074]
[0075]
[0076]
[0077] 5.5 Distributed power generation operation constraints consist of upper and lower limits for active and reactive power output:
[0078]
[0079]
[0080] 5.6 The operational constraints of flexible soft switching consist of two-port capacity constraints and active power balance:
[0081]
[0082]
[0083]
[0084] 5.7 The load shedding constraint is that neither the active nor reactive load shedding amount shall exceed the corresponding load demand:
[0085]
[0086]
[0087] In the formula, This indicates the usable active power output of photovoltaic systems after accounting for prediction errors. Indicates the first Apparent power capacity of Taiwan's solar panels This indicates the maximum reactive power regulation capability of photovoltaic systems. Indicates the projected output of photovoltaic power; and These represent the maximum charging power and the maximum discharging power, respectively. and These represent the charging state and the discharging state, respectively. Formula (39) is used to avoid simultaneous charging and discharging at the same time. Indicates the energy storage status. Indicates the energy storage self-sufficiency coefficient. and These represent charging efficiency and discharging efficiency, respectively. Indicates the scheduling time interval. Indicates energy storage capacity, and These represent the lower and upper limits of the energy storage state of charge ratio, respectively. , These represent the lower and upper limits of the active power output of distributed generation, respectively. , These represent the lower and upper limits of reactive power output of distributed generation, respectively. This represents the set of flexible soft-switching connection ports. and These represent the flexible soft switches at the ports. ,port The rated capacity at the point; Formula (44) represents the active power balance between the two ports of the flexible soft switch when active power loss is ignored; Formulas (47) and (48) are used to constrain the load shedding amount to not exceed the corresponding load demand; Through Formulas (27)-(48), for any given photovoltaic prediction error scenario and any node PQ operating point, it can be determined whether there are internal operating variables that satisfy all distribution network operating equations, network security constraints and flexible resource operating constraints.
[0088] Furthermore, step 6 includes the following steps:
[0089] 6.1 Construct a Wasserstein fuzzy set. Based on the 24-dimensional photovoltaic prediction error scenario set formed in step 3, construct an empirical distribution and assign corresponding empirical probabilities to each historical error scenario. Use the norm distance between any two 24-dimensional photovoltaic prediction error scenario vectors as the scenario distance. Determine the Wasserstein radius based on the error scenario sample mean, sample bias statistic, number of error scenarios, and confidence level parameter. Construct a first-order Wasserstein fuzzy set with the empirical distribution as the center and the Wasserstein radius as the upper limit of the distance. This set is used to describe the possible offset range of the true photovoltaic prediction error probability distribution relative to the empirical distribution under limited historical sample conditions.
[0090] The empirical distribution, scene distance, Wasserstein fuzzy set, and Wasserstein radius are determined by the following formula:
[0091]
[0092]
[0093]
[0094]
[0095] In the formula, Indicates by The empirical distribution formed by candidate photovoltaic prediction error scenarios; and Indicates the error scenario index; This indicates that the focus is on the error scene vector. Dirac measure at the location; Represents the error scenario vector and The distance between them; Denotes the norm 1; Represents the Wasserstein fuzzy set; This represents the candidate true distribution in the Wasserstein fuzzy set; Represents the first-order Wasserstein distance; Represents the Wasserstein radius; This represents the radius statistic determined by the dispersion of the error samples; Indicates the confidence level parameter;
[0096] 6.2 Construct a two-stage Wasserstein fuzzy set elliptical operating domain optimization model. Couple the Wasserstein fuzzy set constructed in step 6.1 with the candidate ellipse of root node PQ established in step 4 and the distribution network scenario operation model established in step 5 to establish a two-stage Wasserstein fuzzy set elliptical operating domain optimization model. The first-stage decision variables include the ellipse center and ellipse shape matrix for each time period, which are used to determine the candidate elliptical operating domain of root node PQ before the realization of photovoltaic prediction error. The second-stage decision variables include the operating variables of photovoltaic, energy storage system, distributed power source, flexible soft switch, branch power flow, node voltage and load shedding, which are adjusted to meet the distribution network operation equation, network security constraints and flexible resource operation constraints after the first-stage elliptical operating domain and photovoltaic prediction error scenario.
[0097] The objective function of the two-stage Wasserstein sub-Bruker elliptical running domain optimization model includes a running domain extension term determined by the determinant of the elliptical shape matrix and a sub-Bruker running risk term under the Wasserstein fuzzy set; wherein, the sub-Bruker running risk term is the maximum value of the expected value of the second-stage running cost among all candidate error probability distributions allowed by the Wasserstein fuzzy set. The objective function is determined by the following formula:
[0098]
[0099] In the formula, Represents the determinant of a matrix; This represents the risk weighting coefficient for the operation of the sub-bar. Represent Wasserstein's dual variable; Indicates the first The auxiliary upper bound variable corresponding to each empirical sample.
[0100] 6.3 Determine the minimum operating cost of the second stage and establish Wasserstein dual constraints. For a given first-stage ellipse parameter and any photovoltaic prediction error scenario, the second-stage operation adjustment ensures that each ellipse checkpoint meets the requirements for safe operation of the distribution network. The weighted sum of the active power load shedding cost, reactive power load shedding cost, and curtailment cost is defined as the second-stage operating cost corresponding to the error scenario, and its minimum value is taken as the minimum operating cost of the second stage for that scenario. Among all error probability distributions allowed by the Wasserstein fuzzy set, the maximum expected value of the minimum operating cost of the second stage is taken as the sub-Bruker operating risk. The sub-Bruker operating risk is transformed into a solvable dual constraint through Wasserstein dual variables and empirical sample auxiliary upper bound variables.
[0101] The minimum operating cost of the second stage and the Wasserstein duality constraint are determined by the following formula:
[0102]
[0103]
[0104] In the formula, This represents the minimum running cost of the second stage under given ellipse parameters and error scenarios; This represents the set of variables composed of the centers of the ellipse at each time interval; Representing a scene The second stage of operation involves regulating variables; Represents the elliptical shape matrix of each time period A set of variables; This represents the set of feasible operations under given ellipse parameters and error scenarios; This represents the active power reduction cost coefficient; This represents the reactive power reduction cost coefficient; This represents the cost coefficient for wasted light. Representing a scene The amount of active power reduction below; Representing a scene The amount of reactive power reduction; Representing a scene The amount of light discarded.
[0105] Furthermore, step 7 includes:
[0106] 7.1 Construct the main problem and solve the second-stage running subproblems. Select an initial activation error scenario from the photovoltaic prediction error scenario set formed in step 3, and establish an activation scenario set based on the initial activation error scenario; select the photovoltaic prediction error scenario with the smallest 1-norm distance to the zero error vector as the initial activation error scenario, and the initial activation scenario is determined by the following formula:
[0107]
[0108] In the formula, The initial activation scenario number is represented. A main problem is constructed and solved based on the current set of activation error scenarios. The main problem is to solve the two-stage Wasserstein sub-Bruker elliptical running domain optimization model described in step 6 within the range of currently identified activation error scenarios. Its variables include the ellipse center and ellipse shape matrix of the first stage, the second-stage running adjustment variable corresponding to the current activation error scenario, as well as the Wasserstein dual variable and the empirical sample auxiliary upper bound variable. By solving the main problem, the current candidate elliptical running domain, the current Wasserstein dual variable, and the current sub-Bruker risk upper bound are obtained.
[0109] With the first-stage ellipse center and ellipse shape matrix obtained from the main problem fixed, a second-stage operation sub-problem is constructed and solved for each candidate error scenario in the photovoltaic prediction error scenario set. The second-stage operation sub-problem is used to determine the minimum second-stage operation cost required to satisfy the distribution network operation model described in step 5 under each candidate error scenario by adjusting the second-stage operation variables such as photovoltaic, energy storage system, distributed power source, flexible soft switch, branch power flow, node voltage and load shedding, without changing the current candidate ellipse operation domain.
[0110] 7.2 Calculate the maximum violation and perform iterative updates. Based on the minimum running cost of the second stage corresponding to each candidate error scenario, the current Wasserstein dual variable, the empirical sample auxiliary upper bound variable, and the scenario distance between error scenarios, calculate the violation of the Wasserstein dual constraint. Then, determine the maximum violation and its corresponding candidate error scenario from all combinations of empirical samples and candidate error scenarios. The maximum violation is determined by the following formula:
[0111]
[0112] In the formula, Indicates the iteration number of the column and constraint generation algorithm; Indicates the first The maximum violation in the next iteration; Indicates the first The center of the ellipse obtained in the next iteration; Indicates the first The elliptical shape matrix obtained from the next iteration; Indicates the first Wasserstein dual variables obtained in the next iteration; Indicates the first The result of the iteration is the first An auxiliary upper bound variable.
[0113] When the maximum violation is greater than the preset convergence threshold, the candidate error scenario that generates the maximum violation is added to the set of active error scenarios, and the Wasserstein dual cut constraint corresponding to the scenario is added to the main problem. The main problem is solved again and the second-stage running subproblem is checked. When the maximum violation is not greater than the preset convergence threshold, the iteration stops, and the ellipse center, ellipse shape matrix, ellipse boundary, ellipse area and Bruker bar running risk of each time period are output to obtain the root node PQ Bruker bar ellipse running domain result.
[0114] Beneficial effects: Compared with the prior art, the technical solution of the present invention has the following beneficial technical effects:
[0115] (1) This invention constructs a 24-hour photovoltaic prediction error scenario based on weather conditions, so that the error sample can reflect the output fluctuation differences under different meteorological conditions such as sunny, cloudy and overcast days, thereby improving the refinement of the operation domain assessment input.
[0116] (2) This invention describes the offset of the true error distribution relative to the empirical distribution under finite historical samples by using Wasserstein fuzzy sets, and incorporates the worst distribution operation risk into the elliptical operating domain optimization model, which can reduce the risk of the operating domain boundary being too optimistic or too conservative due to relying solely on the empirical distribution.
[0117] (3) This invention approximates the elliptical boundary by using finite direction points and applies power flow constraints and flexible resource operation constraints of the distribution network to the boundary points, which can reduce the complexity of solving the operating domain while ensuring the feasibility of decomposing points inside the ellipse.
[0118] (4) The present invention uses a column and constraint generation algorithm to iteratively add the scenario that most severely violates the Wasserstein dual constraint, avoiding the introduction of all candidate scenarios at once and causing the model size to be too large, thereby improving the solution efficiency of the elliptical running domain of the sub-Bruker bar.
[0119] (5) The present invention combines the active power balance, reactive power support and capacity constraints of the flexible soft switch port with the feasibility verification of the elliptical boundary point, and can quantify the impact of SOP access on the area, voltage level and branch load of the root node PQ operating domain under the condition of uncertain photovoltaic prediction error distribution. In the embodiment, after SOP access, the average area of the elliptical operating domain increased from 5.071 to 5.266 in 24 hours, the minimum voltage of the system increased from 0.950pu to 0.967pu, and the average load rate of typical branches decreased from 40.24% to 15.91%.
[0120] (6) This invention can characterize the impact of photovoltaic error distribution shift on the boundary of the operating domain and the operating risk under a limited sample, and can quantify the improvement effect on the area of the operating domain, voltage level and branch load before and after SOP access, providing a basis for flexible resource planning and safety assessment of the distribution network. Attached Figure Description
[0121] Figure 1 This is a flowchart of the present invention.
[0122] Figure 2 This is a comparison diagram of the elliptical operating domain before and after SOP access in this invention.
[0123] Figure 3 This is a comparison of the elliptical area before and after SOP connection in this invention over 24 hours.
[0124] Figure 4 This refers to the change in branch load before and after SOP access in this invention. Detailed Implementation
[0125] The present invention will be further illustrated below with reference to the accompanying drawings and specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. After reading this invention, any modifications of the invention in various equivalent forms by those skilled in the art will fall within the scope defined by the appended claims.
[0126] like Figure 1As shown in the figure, this embodiment provides a method for characterizing the elliptical operating domain of a flexible power distribution system that takes into account soft switching. The method includes the following steps.
[0127] Step 1: Obtain the target distribution network's network topology parameters, electrical parameters, root node information, photovoltaic (PV) unit parameters, energy storage system parameters, distributed generation (DG) parameters, flexible switching parameters, historical PV output data, and historical meteorological data. Network topology parameters include node sets, branch sets, and branch connection relationships; electrical parameters include branch resistance, branch reactance, node voltage upper and lower limits, branch capacity upper limits, and load data; flexible resource parameters include PV unit capacity and inverter capacity, energy storage system capacity and charging / discharging power limits, active and reactive power output ranges of DG, flexible switching capacity, and port connection nodes. The evaluation and scheduling cycle is set to 24 consecutive 1-hour time periods.
[0128] Step 2: Perform maximum and minimum normalization on historical photovoltaic power output data and historical meteorological data according to formula (1). Meteorological variables include total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure. Determine daytime periods based on total solar irradiance and global horizontal irradiance. Use the irradiance quantile of daytime samples of the same month and hour as the reference irradiance to calculate the daily relative irradiance mean and daily relative irradiance standard deviation. Based on this, classify the weather conditions as sunny, cloudy, and overcast. Further construct the sine and cosine periodic characteristics corresponding to the hour, month, and day within the year according to formula (2).
[0129] Step 3: Use the historical sequence of the past 72 hours as the input window of the LSTM photovoltaic power prediction model. The historical sequence includes at least the normalized photovoltaic output of the previous hour, meteorological variables, and time period characteristics. Among them, meteorological variables include total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure. When meteorological data for future periods are unavailable, weather state predictions can be used instead or set as empty vectors. Obtain the normalized photovoltaic prediction value according to formula (3), obtain the photovoltaic prediction output in actual dimensions by inverse normalization according to formula (4), calculate the photovoltaic prediction error according to formula (5), and combine the prediction errors of 24 consecutive time periods to form a 24-dimensional error scenario vector according to formula (6). Then, obtain the available photovoltaic output after considering the prediction error according to formula (7). The LSTM model constructs a 72-hour input window and a single-period feature vector according to formulas (8) and (9). It completes the gating state update of the two-layer LSTM according to formulas (10) to (16). It obtains the normalized photovoltaic prediction value through Dropout, ReLU fully connected layer and linear output layer according to formulas (17) to (19). The model training calculates Huber loss and LSTM training loss according to formulas (20) and (21), and uses Adam optimizer to update model parameters. When making continuous 24-hour rolling predictions, the normalized prediction value of the previous period is backfilled into the photovoltaic output feature of the previous hour of the next period according to formula (22). Finally, based on the weather conditions obtained in step 2, a set of 24-hour photovoltaic prediction error scenarios corresponding to sunny, cloudy and overcast days is formed respectively.
[0130] Step 4: Taking the root node PQ power vector as the object, the candidate ellipse running domain of the root node PQ for each time period is characterized by the ellipse center and ellipse shape matrix according to formula (23), and the matrix form of the two-dimensional symmetric shape matrix is given according to formula (24). In order to perform finite point verification on the candidate ellipse, the vertices of the finite external approximation polygon containing the unit circle are constructed according to formula (25), and then each reference vertex is mapped to the root node PQ plane according to formula (26) to obtain the root node PQ verification vertices corresponding to each time period. The number of external approximation polygon vertices can be set according to the calculation accuracy and solution scale. In this embodiment, it can be set to 8.
[0131] Step 5: Establish a distribution network scenario operation model for each scheduling period, each root node PQ verification vertex, and each photovoltaic prediction error scenario. Establish node active power balance, node reactive power balance, node net injected power composition, and linearized branch voltage drop equations according to formulas (27) to (31); apply node voltage upper and lower limits and branch apparent power capacity constraints according to formulas (32) and (33); establish photovoltaic active power availability, prediction error corrected available output, reactive power adjustment range, and inverter capacity constraints according to formulas (34) to (37); establish energy storage charging and discharging power, charging and discharging mode, energy state update, and remaining energy margin constraints according to formulas (38) to (41); establish distributed power active and reactive power output upper and lower limit constraints according to formulas (42) and (43); establish flexible soft switch two-port active power balance and port capacity constraints according to formulas (44) to (46); establish active and reactive power load shedding constraints according to formulas (47) and (48). Therefore, it can be determined whether there are internal operating variables that satisfy all operating equations, safety constraints, and flexible resource operating constraints under a given error scenario and root node PQ check vertex.
[0132] Step 6: Based on the 24-hour photovoltaic prediction error scenario set, construct the empirical distribution according to formula (49), calculate the first norm distance between error scenarios according to formula (50), construct the first-order Wasserstein fuzzy set according to formula (51), and determine the Wasserstein radius according to formula (52) based on the error sample dispersion, scenario number, and confidence level parameters. On this basis, couple the root node PQ candidate ellipse established in step 4 with the distribution network scenario operation model established in step 5 to construct a two-stage Wasserstein sub-Bruker ellipse operation domain optimization model; the first stage determines the ellipse center and ellipse shape matrix for each time period, and the second stage adjusts the distribution network operation for the given photovoltaic prediction error scenario. Establish the objective function that simultaneously considers the operation domain extension term and the sub-Bruker operation risk term according to formula (53), determine the minimum operation cost of the second stage under each error scenario according to formula (54), and establish Wasserstein dual constraints according to formula (55).
[0133] Step 7: The column and constraint generation algorithm is used to solve the two-stage Wasserstein sub-Bruker elliptical running domain optimization model. First, the photovoltaic prediction error scenario with the smallest distance to the first norm of the zero error vector is selected as the initial activation scenario according to formula (56). The main problem is constructed and solved based on the current set of activated error scenarios to obtain the current candidate elliptical running domain, Wasserstein dual variables, and sub-Bruker risk upper bound. Then, the first-stage elliptical center and elliptical shape matrix obtained from the main problem are fixed, and the second-stage running subproblems are solved for all candidate error scenarios. The maximum violation of the Wasserstein dual constraint is calculated according to formula (57). When the maximum violation is greater than the preset convergence threshold, the corresponding unfavorable error scenario and its Wasserstein dual cut constraint are added to the main problem and the iteration continues. When the maximum violation is not greater than the preset convergence threshold, the iteration stops, and the elliptical center, elliptical shape matrix, elliptical boundary, elliptical area, and sub-Bruker running risk for each time period are output to obtain the root node PQ sub-Bruker elliptical running domain result.
[0134] The root node PQ sub-bar elliptical operating domain results include the ellipse center, ellipse shape matrix, ellipse boundary points, ellipse area, worst-case expected operating cost, and worst-case distribution main scenarios for each time period. These results can be used for distribution network security assessment, flexible scheduling, and evaluation of the effectiveness of flexible soft-switching.
[0135] In one implementation, a computational system is constructed based on the IEEE 33-node power distribution system, with the root node set at node 1. The scheduling cycle is 24 hours, the time interval is 1 hour, and the elliptical operating domain is approximated using 8 direction points. Multiple photovoltaic power sources, energy storage devices, distributed power sources, and flexible soft switches can be connected to the system to compare the changes in the root node's PQ elliptical operating domain before and after the flexible soft switch is connected.
[0136] During verification, the root node PQ elliptical operating domain can be calculated separately for both cases with and without flexible soft switches, and the 24-hour elliptical area, node voltage, branch load rate, and load shedding status can be compared. For example... Figure 2 As shown, during typical periods, the elliptical boundary expands outward after SOP connection, indicating that the active power transfer between feeders and the reactive power support at the ports can release voltage constraints and branch capacity constraints; as Figure 3As shown, in this embodiment, the average area of the elliptical operating domain increased from 5.071 to 5.266 over 24 hours. The most significant area increase occurred during periods t21, t20, and t22, with corresponding increases of 15.579%, 14.469%, and 12.308%, respectively. Simultaneously, the system minimum voltage increased from 0.950 pu to 0.967 pu, the average voltage increased from 0.988 pu to 1.017 pu, and the average load factor of typical branches decreased from 40.24% to 15.91%, a decrease of 24.33 percentage points on average. This indicates that the SOP can alleviate branch load pressure through power flow transfer and power redistribution. The above values represent the effects of this embodiment and do not constitute a limitation on the scope of the claims.
[0137] Monte Carlo sampling can also be used to verify the feasibility of decomposing points inside the elliptical operating domain: randomly sample the root node PQ point inside the ellipse at each time period to form a 24-hour root node power trajectory; fix the sampling trajectory as the root node injected power, re-establish the operational feasibility verification model, and use non-negative relaxation variables to determine whether the voltage, branch capacity, photovoltaic inverter capacity, flexible soft switching capacity and energy storage constraints are met.
[0138] Under different weather conditions, Wasserstein fuzzy sets can be constructed using error samples from sunny, cloudy, and overcast days, and the sub-Bruker elliptical running domain model can be solved under different confidence levels. Table 1 shows the running domain area, Wasserstein radius, worst expected running cost, and worst distribution main scenarios for different weather types under different confidence levels.
[0139] Table 1. Scoring results at different confidence levels
[0140] (a) Risk outcome at a confidence level of 0.05
[0141] Weather type Wasserstein radius Worst expected operating cost Worst scenario Worst-case scenario probability sunny 0.127 0.474 Scene 9 0.333 partly cloudy 0.118 162.615 Scene 10 0.184 cloudy day 0.117 0.464 Scene 4 0.286
[0142] (b) Risk outcome at a confidence level of 0.25
[0143] Weather type Wasserstein radius Worst expected operating cost Worst scenario Worst-case scenario probability sunny 0.300 0.699 Scene 2 0.793 partly cloudy 0.280 248.866 Scene 10 0.332 cloudy day 0.276 0.517 Scene 2 0.574
[0144] (c) Risk outcome at a confidence level of 0.45
[0145] Weather type Wasserstein radius Worst expected operating cost Worst scenario Worst-case scenario probability sunny 0.433 0.736 Scene 2 0.942 partly cloudy 0.403 312.566 Scene 10 0.470 cloudy day 0.398 0.518 Scene 2 0.825
[0146] (d) Risk outcome at a confidence level of 0.65
[0147] Weather type Wasserstein radius Worst expected operating cost Worst scenario Worst-case scenario probability sunny 0.574 0.734 Scene 2 1.000 partly cloudy 0.534 377.131 Scene 10 0.581 cloudy day 0.527 0.518 Scene 2 0.993
Claims
1. A method for characterizing robust elliptic operation region of a flexible power distribution system considering soft-switching, characterized in that, The method includes the following steps: Step 1: Obtain the network topology parameters, electrical parameters, root node information, photovoltaic unit parameters, energy storage system parameters, distributed power generation parameters, flexible soft switching parameters, historical photovoltaic output data, and historical meteorological data of the target distribution network. The evaluation and scheduling cycle consists of 24 consecutive 1-hour time periods. Step 2: Normalize the historical photovoltaic power output data and historical meteorological data, classify the weather conditions as sunny, cloudy and overcast based on the relative irradiance level and its intraday fluctuation characteristics, and construct the corresponding periodic features for hours, days within the year and months. Step 3: Using the historical sequence of the past 72 hours as the input window for the LSTM photovoltaic power prediction model in each prediction period, perform hourly rolling predictions for 24 consecutive 1-hour periods within the scheduling cycle to be evaluated; compare the predicted output of each period with the corresponding actual output, combine the 24 consecutive periods to form a 24-dimensional photovoltaic prediction error scenario vector, and construct photovoltaic prediction error scenario sets corresponding to sunny, cloudy, and overcast days according to the weather conditions obtained in Step 2; Step 4: For the projection of the high-dimensional feasible set of the distribution network onto the active-reactive power plane of the root node, the PQ operating domain of the root node is parameterized by an ellipse formed by translation, rotation and scaling of the unit circle. The center of the ellipse and the positive semi-definite matrix are used as the operating domain parameters to be optimized. Verification vertices are constructed by constructing a finite outer approximation polygon containing the unit circle, and the finite vertices on the PQ plane of the root node are obtained by mapping through the center of the ellipse and the shape matrix, providing input for the establishment of the elliptical approximation operating domain model of the distribution network in Step 5. Step 5: Based on the vertices of the approximate polygon outside the ellipse formed in Step 4, establish a distribution network scenario operation model. The model consists of distribution network operation equations, network security constraints, and flexible resource operation constraints. The distribution network operation equations include node active power balance, node reactive power balance, node net injected power composition, and linearized branch voltage drop equations. Network security constraints include node voltage upper and lower limits and branch capacity constraints. Flexible resource operation constraints include photovoltaic, energy storage systems, distributed power sources, flexible soft switching, and load shedding constraints. Step 6: Based on the 24-dimensional photovoltaic prediction error scenario set obtained in Step 3, construct the empirical distribution, scenario distance, Wasserstein radius, and Wasserstein fuzzy set. Combined with the candidate elliptical operating domain established in Step 4 and the distribution network elliptical approximate operating domain model established in Step 5, construct a two-stage Wasserstein sparse bar elliptical operating domain optimization model. In the first stage, determine the ellipse center and ellipse shape matrix for each time period. In the second stage, adjust the distribution network operation for different photovoltaic prediction error scenarios to expand the elliptical operating domain of the root node PQ and reduce the operating risk corresponding to the worst probability distribution within the Wasserstein fuzzy set. Step 7: Use the column and constraint generation algorithm to solve the two-stage Wasserstein sub-Brubar elliptical running domain optimization model established in Step 6; determine the candidate elliptical running domain and sub-Brubar risk upper bound under the current set of active error scenarios through the main problem; then check all candidate error scenarios through the second-stage running sub-problems and identify the unfavorable scenario that generates the maximum risk constraint violation; iteratively add the scenario to the main problem until the maximum violation does not exceed the preset convergence threshold; and output the root node PQ sub-Brubar elliptical running domain result.
2. The method of claim 1, wherein, Step 2 includes: performing maximum and minimum normalization on photovoltaic power, total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure; determining daytime periods based on total solar irradiance and global horizontal irradiance; using the irradiance quantiles of daytime samples from the same month and hour as reference irradiance to calculate the daily relative irradiance mean and daily relative irradiance standard deviation, and classifying weather conditions as sunny, cloudy, and overcast based on this; and constructing sine and cosine periodic characteristics corresponding to hours, months, and days within the year. The maximum and minimum normalization processing and the construction of the time period feature are performed according to the following formula: (1) (2) In the formula, Represents the normalized variable value; This represents the original variable value before normalization; This represents the minimum value of the variable in the training samples; This represents the maximum value of the variable in the training samples; Indicates time period Time-period feature vector; Indicates time period The corresponding hour number; Indicates time period The corresponding day number within the year; Indicates time period The corresponding month number.
3. The method according to claim 1, characterized in that, In step 3, the LSTM photovoltaic power prediction model uses its previous 72-hour historical sequence as the input window for each prediction period, where 72 hours represents the historical review length of a single prediction. The evaluation scheduling period consists of 24 consecutive 1-hour periods, where 24 hours represents the time dimension of the error scenario vector. Predictions are performed sequentially for the 1st to 24th periods within the evaluation period, with the input window shifted forward by 72 hours hourly with each prediction time, thus forming a photovoltaic prediction sequence of 24 consecutive periods. The historical sequence includes at least the normalized photovoltaic output, irradiance, ambient temperature, air pressure, and time period characteristics of the previous hour. The LSTM photovoltaic power prediction model includes a first LSTM layer, a first Dropout layer, a second LSTM layer, a second Dropout layer, a ReLU fully connected layer, and a linear output layer connected in sequence. Normalized photovoltaic prediction values are obtained based on LSTM gating state updates. The 24-hour photovoltaic prediction error scenario set consists of multiple error scenario vectors, each of which includes photovoltaic prediction errors for 24 consecutive periods. The LSTM photovoltaic power prediction, inverse normalization, prediction error, error scenario vector, and available photovoltaic output are determined by the following formula: (3) (4) (5) (6) (7) In the formula, This indicates a scenario of photovoltaic prediction error. This represents the set of candidate photovoltaic prediction error scenarios; This represents the normalized photovoltaic prediction value output by the LSTM. This represents a nonlinear mapping constructed from an LSTM network. This represents the trainable parameters of the LSTM model; This represents the historical sequence features input to the LSTM; Indicates time period The projected output of photovoltaic power; This represents the maximum photovoltaic output in the training samples; This represents the minimum photovoltaic output in the training samples; Indicates time period The photovoltaic prediction error; Indicates time period The actual output of photovoltaic power; Indicates the first A 24-hour photovoltaic forecast error scenario vector; superscript Indicates vector transpose; Indicates the first The scene in time period The photovoltaic prediction error; Indicates the first One error scenario in the time period The photovoltaic power available for use; Indicates the apparent power capacity of photovoltaics; The historical sequence input, input gate, forget gate and output gate calculation of the two-layer LSTM photovoltaic power prediction model, candidate cell state and cell state update, Dropout random inactivation, ReLU fully connected layer, linear output layer, training loss, and continuous multi-time recursive input relationship are expanded as follows: (8) (9) (10) (11) (12) (13) (14) (15) (16) (17) (18) (19) (20) (21) (22) In the formula, Indicates the length of the history window; Indicates time period Single-time feature vector, superscript This indicates that the corresponding variable has undergone normalization. This represents the normalized photovoltaic output for the previous hour. Indicates time period Normalized actual photovoltaic output This represents a normalized meteorological characteristic vector, which includes at least total solar irradiance, direct normal irradiance, global horizontal irradiance, ambient temperature, and air pressure; when meteorological data for future periods are unavailable. It can be set to an empty vector or replaced by weather state forecasts. Indicates the LSTM layer number. Corresponding to the first LSTM layer, Corresponding to the second LSTM layer; Indicates the first Layer in time period Input, This indicates a hidden state. Indicates cell state; , , These represent the input gate, forget gate, and output gate, respectively. Indicates the state of candidate cells; and This represents the input weight matrix and the cyclic weight matrix used in the corresponding calculations. Denotes the corresponding calculated bias vector, where, Pick , , and The time parameters correspond to the calculation parameters of the input gate, forget gate, output gate, and candidate cell state, respectively; where the cell state is the internal memory variable used by LSTM to store and transmit historical information in the time series. This represents the Sigmoid function. Represents the hyperbolic tangent function. Represents the Hadamard product; and This represents the Dropout operator. and Indicates the discard rate. This indicates that the first layer of LSTM is in the time period The output hidden state is discarded at a rate of The feature vector obtained after processing by the first Dropout layer is used as the input of the second LSTM layer; This indicates that the second layer of LSTM is in the time period The output hidden state is discarded at a rate of The feature vector obtained after processing by the second Dropout layer is used as the input to the subsequent ReLU fully connected layer; This represents a linear rectified function; random deactivation is performed during the training phase and disabled during the prediction phase. This represents the hidden state of the second LSTM layer output after Dropout; This indicates the output of the ReLU fully connected layer. , This represents the weights and biases of the ReLU fully connected layer. , This represents the weights and biases of the linear output layer; Indicates time periods in the training samples Normalized actual photovoltaic output Indicates time period The corresponding Huber loss, This represents the Huber threshold. This represents the training loss of LSTM. This represents the set of target time period indices corresponding to the training samples of the LSTM model; This indicates the number of training samples in this set; The prediction step number is indicated by the continuous rolling prediction. The Adam optimizer is used to train the parameters in the LSTM layer, the fully connected layer and the output layer. Formula (22) represents the recursive input relationship during continuous prediction, that is, the normalized prediction value of the previous time period is used as the photovoltaic power output feature of the previous hour in the next time period.
4. The method according to claim 1, characterized in that, Step 4 includes: (4.1) For each scheduling period within the 24-hour evaluation scheduling cycle Using the active and reactive power of the root node to form a two-dimensional power vector, the candidate elliptical operating domains of the root node PQ for each time period are constructed according to equation (23): (23) Among them, the two-dimensional symmetric shape matrix Represented as: (24) In the formula, The root node PQ represents the elliptical running domain of a continuous time period; Represents the Cartesian product of the elliptical running domain of the root node PQ at each time period; , They represent time periods respectively. Active and reactive power at the center of the ellipse; A positive semi-definite two-dimensional matrix representing the shape of an ellipse; Indicates the set of evaluation periods; Indicates the scheduling period; , They represent time periods respectively. The active and reactive power of the root node; , , and Representing shape matrices respectively The corresponding matrix elements; Represents a two-dimensional reference vector within the unit circle; The second norm of a vector; (4.2) Construct the finite number of vertices of the outer approximation polygon of the unit circle, and set the total number of vertices of the outer approximation polygon to be . , No. The vertices of the approximate polygon are denoted as _____. And determine its two coordinate components according to equation (25): (25) In the formula, Indicates the first Two-dimensional coordinate vectors of the vertices of an approximate polygon; and They represent the first The coordinate components of the vertices of an external approximate polygon in the directions of the first and second coordinate axes; Indicates the vertex index of the external approximate polygon; (4.3) Map the vertices of the outer approximation polygon to the root node PQ plane, for each scheduling period. and the vertices of each external approximate polygon Using the ellipse center and shape matrix determined in step 4.1, the reference vertex obtained in step 4.2 is... Mapping to the root node PQ plane, the corresponding root node PQ verification vertex is obtained according to equation (26): (26) In the formula, , They represent time periods respectively. Next The active and reactive power of the root node obtained by mapping the vertices of the external approximation polygon will be used to form the active and reactive power of each scheduling period. Each root node PQ verification vertex serves as the root node power input for establishing the distribution network scenario operation model in step 5. Distribution network operation equations, network security constraints, and photovoltaic, energy storage system, distributed power source, flexible soft switch and load shedding operation constraints are applied to each verification vertex.
5. The method according to claim 1, characterized in that, In step 5, a power distribution network scenario operation model is established for the vertices of the approximate polygon outside the ellipse. The power distribution network scenario operation model consists of three parts: "power distribution network operation equations", "network security constraints", and "flexible resource operation constraints". The power distribution network operation equations are used to describe the physical relationship between nodes and branches; the network security constraints are used to limit node voltage and branch capacity; and the flexible resource operation constraints are used to limit the adjustable range of photovoltaic, energy storage systems, distributed power sources, flexible soft switching, and load shedding. (5.1) The power distribution network operation equations consist of node power balance, node net injection, and linearized branch voltage drop: (27) (28) (29) (30) (31) (5.2) The network security constraints consist of node voltage upper and lower limits and branch apparent power capacity constraints: (32) (33) In the formula, Represents the set of nodes in the distribution network. Indicates the node index; Represents the set of branches in the distribution network. Indicates the connection node and nodes The side road; , Representing nodes respectively Net active power and net reactive power injection; , Representing branch roads The active and reactive power flowing through; , Representing nodes respectively During the period Ellipse boundary points and error scenarios The photovoltaic system outputs both active and reactive power. , These represent the energy storage discharge and charging power, respectively. , These represent the active and reactive power outputs of the distributed power source, respectively. , These represent the flexible soft-switching injection nodes. Active and reactive power; , These represent the active and reactive load shedding quantities at the node, respectively. , These represent the active and reactive loads of the node, respectively. Represents a node Voltage squared, , Representing branch roads Resistance and reactance, , These represent the lower and upper limits of the square of the node voltage, respectively. Indicates the maximum capacity of the branch; (5.3) Photovoltaic operation constraints consist of available active power, available output after prediction error correction, inverter capacity, and reactive power regulation range: (34) (35) (36) (37) (5.4) The operating constraints of the energy storage system consist of upper and lower limits of charge and discharge power, charge and discharge mode restrictions, energy state updates, and remaining energy margin restrictions: (38) (39) (40) (41) (5.5) The operating constraints of distributed power sources consist of upper and lower limits for active and reactive power output: (42) (43) (5.6) The operational constraints of flexible soft switching consist of two-port capacity constraints and active power balance: (44) (45) (46) (5.7) The load shedding constraint is that neither the active nor reactive load shedding amount shall exceed the corresponding load demand: (47) (48) In the formula, This indicates the usable active power output of photovoltaic systems after accounting for prediction errors. Indicates the first Apparent power capacity of Taiwan's solar panels This indicates the maximum reactive power regulation capability of photovoltaic systems. Indicates the projected output of photovoltaic power; and These represent the maximum charging power and the maximum discharging power, respectively. and These represent the charging state and the discharging state, respectively. Formula (39) is used to avoid simultaneous charging and discharging at the same time. Indicates the energy storage status. Indicates the energy storage self-sufficiency coefficient. and These represent charging efficiency and discharging efficiency, respectively. Indicates the scheduling time interval. Indicates energy storage capacity. and These represent the lower and upper limits of the energy storage state of charge ratio, respectively. , These represent the lower and upper limits of the active power output of distributed generation, respectively. , These represent the lower and upper limits of reactive power output of distributed generation, respectively. This represents the set of flexible soft-switching connection ports. and These represent the flexible soft switches at the ports. ,port The rated capacity at the point; Formula (44) represents the active power balance between the two ports of the flexible soft switch when active power loss is ignored; Formulas (47) and (48) are used to constrain the load shedding amount to not exceed the corresponding load demand; Through Formulas (27)-(48), for any given photovoltaic prediction error scenario and any node PQ operating point, it can be determined whether there are internal operating variables that satisfy all distribution network operating equations, network security constraints and flexible resource operating constraints.
6. The method according to claim 1, characterized in that, The method for step 6 is as follows: (6.1) Construct a Wasserstein fuzzy set. Based on the 24-dimensional photovoltaic prediction error scenario set formed in step 3, construct an empirical distribution and assign corresponding empirical probabilities to each historical error scenario. Use the norm distance between any two 24-dimensional photovoltaic prediction error scenario vectors as the scenario distance. Determine the Wasserstein radius based on the error scenario sample mean, sample bias statistic, number of error scenarios, and confidence level parameter. Construct a first-order Wasserstein fuzzy set with the empirical distribution as the center and the Wasserstein radius as the upper limit of the distance. This set is used to describe the possible offset range of the true photovoltaic prediction error probability distribution relative to the empirical distribution under limited historical sample conditions. The empirical distribution, scene distance, Wasserstein fuzzy set, and Wasserstein radius are determined by the following formula: (49) (50) (51) (52) In the formula, Indicates by The empirical distribution formed by candidate photovoltaic prediction error scenarios; and Indicates the error scenario index; This indicates that the focus is on the error scene vector. Dirac measure at the location; Represents the error scenario vector and The distance between them; Denotes the norm 1; Represents the Wasserstein fuzzy set; This represents the candidate true distribution within the Wasserstein fuzzy set; Represents the first-order Wasserstein distance; Represents the Wasserstein radius; This represents the radius statistic determined by the dispersion of the error samples; Indicates the confidence level parameter; (6.2) Construct a two-stage Wasserstein fuzzy set elliptical operating domain optimization model. Couple the Wasserstein fuzzy set constructed in step 6.1 with the root node PQ candidate ellipse established in step 4 and the distribution network scenario operation model established in step 5 to establish a two-stage Wasserstein fuzzy set elliptical operating domain optimization model. The first-stage decision variables include the ellipse center and ellipse shape matrix for each time period, which are used to determine the root node PQ candidate elliptical operating domain before the realization of photovoltaic prediction error. The second-stage decision variables include the photovoltaic, energy storage system, distributed power source, flexible soft switch, branch power flow, node voltage and load shedding operation variables that are adjusted to meet the distribution network operation equation, network security constraints and flexible resource operation constraints after the first-stage elliptical operating domain and photovoltaic prediction error scenario. The objective function of the two-stage Wasserstein sub-Bruker elliptical running domain optimization model includes a running domain extension term determined by the determinant of the elliptical shape matrix and a sub-Bruker running risk term under the Wasserstein fuzzy set; wherein, the sub-Bruker running risk term is the maximum value of the expected value of the second-stage running cost among all candidate error probability distributions allowed by the Wasserstein fuzzy set. The objective function is determined by the following formula: (53) In the formula, Represents the determinant of a matrix; This represents the risk weighting coefficient for the operation of the sub-bar. Represent Wasserstein's dual variable; Indicates the first Auxiliary upper bound variables corresponding to each empirical sample; (6.3) Determine the minimum operating cost of the second stage and establish Wasserstein dual constraints. For a given first-stage ellipse parameter and any photovoltaic prediction error scenario, the second-stage operation adjustment ensures that each ellipse checkpoint meets the requirements for safe operation of the distribution network. The weighted sum of the active power load shedding cost, reactive power load shedding cost, and curtailment cost is defined as the second-stage operating cost corresponding to the error scenario, and its minimum value is taken as the second-stage minimum operating cost of the scenario. Among all error probability distributions allowed by the Wasserstein fuzzy set, the maximum expected value of the second-stage minimum operating cost is taken as the sub-Bruker operating risk. The sub-Bruker operating risk is transformed into a solvable dual constraint through Wasserstein dual variables and empirical sample auxiliary upper bound variables. The minimum operating cost of the second stage and the Wasserstein duality constraint are determined by the following formula: (54) (55) In the formula, This represents the minimum running cost of the second stage under given ellipse parameters and error scenarios; This represents the set of variables composed of the centers of the ellipse at each time interval; Representing a scene The second stage of operation involves regulating variables; Represents the elliptical shape matrix of each time period A set of variables; This represents the set of feasible operations under given ellipse parameters and error scenarios; This represents the active power reduction cost coefficient; This represents the reactive power reduction cost coefficient; This represents the cost coefficient for wasted light. Representing a scene The amount of active power reduction below; Representing a scene The amount of reactive power reduction; Representing a scene The amount of light discarded.
7. The method according to claim 6, characterized in that, The method in step 7 includes the following steps: (7.1) Construct the main problem and solve the second-stage running subproblems. Select an initial activation error scenario from the photovoltaic prediction error scenario set formed in step 3, and establish an activation scenario set based on the initial activation error scenario. Select the photovoltaic prediction error scenario with the smallest 1-norm distance to the zero error vector as the initial activation error scenario. The initial activation scenario is determined by the following formula: (56) In the formula, The initial activation scenario number is represented. The main problem is constructed and solved based on the current set of activation error scenarios. The main problem is to solve the two-stage Wasserstein elliptic running domain optimization model described in step 6 within the range of the currently identified activation error scenarios. Its variables include the ellipse center and ellipse shape matrix of the first stage, the second stage running adjustment variables corresponding to the current activation error scenario, as well as Wasserstein dual variables and empirical sample auxiliary upper bound variables. By solving the main problem, we obtain the current candidate elliptic running domain, the current Wasserstein dual variable, and the current upper bound of the risk of the split-Bruker bar. By fixing the first-stage ellipse center and ellipse shape matrix obtained from the main problem, the second-stage running sub-problems are constructed and solved for each candidate error scenario in the photovoltaic prediction error scenario set. The second-stage operation subproblem is used to determine the minimum second-stage operation cost required to satisfy the distribution network operation model described in step 5 under each candidate error scenario without changing the current candidate elliptical operating domain, by adjusting the second-stage operation variables such as photovoltaic, energy storage system, distributed power source, flexible soft switch, branch power flow, node voltage and load shedding. (7.2) Calculate the maximum violation and perform iterative updates. Based on the minimum running cost of the second stage corresponding to each candidate error scenario, the current Wasserstein dual variable, the empirical sample auxiliary upper bound variable, and the scenario distance between error scenarios, calculate the violation of the Wasserstein dual constraint, and determine the maximum violation and its corresponding candidate error scenario from all combinations of empirical samples and candidate error scenarios. The maximum violation is determined by the following formula: (57) In the formula, Indicates the iteration number of the column and constraint generation algorithm; Indicates the first The maximum violation in the next iteration; Indicates the first The center of the ellipse obtained in the next iteration; Indicates the first The elliptical shape matrix obtained from the next iteration; Indicates the first Wasserstein dual variables obtained in the next iteration; Indicates the first The iteration yields the... One auxiliary upper bound variable; When the maximum violation amount is greater than the preset convergence threshold, the candidate error scenario that generates the maximum violation amount is added to the set of active error scenarios, and the Wasserstein dual cut constraint corresponding to the scenario is added to the main problem. The main problem is solved again and the second-stage running sub-problem verification is performed. When the maximum violation is not greater than the preset convergence threshold, the iteration stops, and the ellipse center, ellipse shape matrix, ellipse boundary, ellipse area and Brubar running risk for each time period are output, thus obtaining the root node PQ Brubar ellipse running domain result.
