A power distribution network timely rolling optimization method and system

Abstract

The application discloses a power distribution network timely rolling optimization method and system, and belongs to the technical field of power system operation optimization. The method comprises the following steps: considering the super-short-term prediction of wind-solar-load prediction deviation correlation; constructing a multi-objective rolling optimization model based on a random model prediction control; solving the model by using an improved multi-objective milu optimization algorithm; and rolling executing control and closed-loop correction of prediction input. The application solves the technical problems that the three-dimensional correlation of wind-solar-load prediction deviation is ignored in the existing flexible interconnected power distribution network optimization operation, and the convergence precision of the traditional optimization algorithm is insufficient. The application can effectively improve the wind-solar-load prediction precision, fully develop the regulation and control potential of the flexible interconnected device, significantly reduce the system network loss, improve the voltage quality, balance the feeder load, and enhance the robustness and economy of the power distribution network operation.

Description

Technical Field

[0001] This invention relates to a real-time rolling optimization method and system for power distribution networks, belonging to the field of power system operation optimization. Background Technology

[0002] With the large-scale grid connection of new energy sources such as wind power and photovoltaics, the high proportion of new energy penetration has profoundly changed the operating characteristics of distribution networks. Traditional passive unidirectional radial distribution networks are gradually transforming into active bidirectional interactive flexible interconnected distribution networks. Flexible interconnected distribution networks rely on flexible interconnection devices such as VSC (Voltage Source Converter) and B2B-VSC (Back-to-Back Voltage Source Converter), which have bidirectional power regulation and reactive power rapid support capabilities. They are key technologies for solving the problem of new energy consumption and improving power quality, and have become the core direction of new power system construction.

[0003] However, the current rolling optimization of flexible interconnected distribution networks still has significant shortcomings: First, wind and solar power outputs are highly random and volatile, and the prediction deviations of the three have complex three-dimensional spatiotemporal correlations. Existing technologies mostly use single-point prediction or independent Monte Carlo deviation scenario prediction, ignoring the coupling relationship between deviations, resulting in insufficient prediction accuracy and difficulty in adapting to minute-level fluctuation characteristics. Second, the optimization of flexible interconnected distribution networks is a high-dimensional, strongly constrained, non-convex, nonlinear multi-objective problem. Traditional multi-objective optimization algorithms have defects such as low convergence accuracy, easy to get trapped in local optima, and uneven front distribution, making it difficult to efficiently solve complex optimization models. Third, existing rolling optimization strategies lack real-time correction mechanisms, which cannot correct the cumulative error of prediction deviations in a timely manner, easily leading to problems such as voltage exceeding limits, network loss increase, and feeder load imbalance, thus restricting the control efficiency of flexible interconnected devices. Summary of the Invention

[0004] The purpose of this invention is to provide a real-time rolling optimization method and system for distribution networks, which addresses the shortcomings of existing flexible interconnected distribution network optimization operations that neglect the three-dimensional correlation of wind, solar and load prediction deviations and the insufficient convergence accuracy of traditional optimization algorithms. By accurately characterizing the deviation correlation, improving the optimization algorithm, and introducing real-time correction, the prediction accuracy is improved, the control capability of flexible interconnected devices is fully utilized, network losses are reduced, voltage quality is improved, and the robustness and economy of distribution network operation are enhanced.

[0005] To achieve the above objectives, the present invention is implemented using the following technical solution.

[0006] In a first aspect, the present invention provides a method for real-time rolling optimization of a power distribution network, comprising:

[0007] Based on the actual values ​​of wind power, photovoltaic power and load at the current moment, the marginal distribution of prediction deviation is fitted. A three-dimensional joint distribution of wind power, photovoltaic power and load is constructed by R-Teng Copula. Monte Carlo sampling and reduction are performed on the three-dimensional joint distribution to obtain the correlation scenario of prediction deviation of wind power, photovoltaic power and load.

[0008] The aforementioned scenario of correlation between wind power, solar power and load forecast deviations is introduced into the AMRA ultra-short-term forecast model. By inputting the actual values ​​of wind power, solar power and load at the current moment, the predicted values ​​of wind power, solar power and load for the next n0 time periods are obtained.

[0009] Based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, a multi-objective rolling optimization model based on stochastic model predictive control is constructed, and an improved multi-objective stag optimization algorithm is used to solve the model to obtain control commands.

[0010] If the optimization cycle has ended, the final decision scheme is output according to the control command. If not, the optimization time domain is rolled forward by introducing timely residual feedback and a real-time correction mechanism for deviation scenario probabilities, and the above steps are repeated until the grid's full-day scheduling is completed and the final decision scheme is output.

[0011] The aforementioned real-time rolling optimization method for distribution networks, wherein fitting the marginal distribution of prediction deviations based on the actual values ​​of wind power, photovoltaics, and load at the current moment includes: fitting the prediction deviations of load and photovoltaics using a normal distribution, and fitting the prediction deviations of wind power using a Beta distribution.

[0012] The aforementioned real-time rolling optimization method for distribution networks, specifically the scenario involving the correlation between wind power, photovoltaic power, and load forecast deviations, is formulated as follows:

[0013] ;

[0014] in, For time t+g, the correlation scenario of wind power, photovoltaic and load forecast deviations is represented; N is the number of scenarios after scenario reduction. Let be the probability corresponding to scenario s; The prediction deviation value for scenario s at time t+g; g is the prediction time period;

[0015] The scenario of correlation between wind power, solar power, and load forecast deviation is introduced into the AMRA ultra-short-term forecasting model, as shown in the following formula:

[0016] ;

[0017] in, The probability values ​​of wind power, photovoltaic power, and load at time t+g are predicted for time t; m and n are the orders of autoregression and moving average, respectively; q is the lag order index in the ARMA model; t+gq is the lag time corresponding to the qth time period before the target prediction time t+g. These are the autoregressive coefficients; The moving average coefficient; The predicted value at time t+gq; This is a random variable representing the prediction bias.

[0018] The aforementioned real-time rolling optimization method for distribution networks, wherein the construction of the multi-objective rolling optimization model includes:

[0019] With the optimization objectives of minimizing system active power loss, minimizing node voltage deviation, and minimizing feeder load balance, a multi-objective optimization objective function is constructed, as shown in the following formula:

[0020] ;

[0021] ;

[0022] ;

[0023] in, , , These are the optimization objective functions for minimizing system active power loss, minimizing node voltage deviation, and optimizing feeder load balance, respectively. Let be the probability corresponding to scenario s; This represents the number of scenes after scene reduction; and These represent the number of branches in the AC and DC sections of the system, respectively. and These are the active power and reactive power at the end of the Lth branch in the s-th scenario, respectively; Let L be the impedance of the Lth branch; To optimize the rolling cycle; The number of VSCs; Number of B2B-VSCs; Let be the power loss of the VSC connected at node d at time t; Let be the power loss of the B2B-VSC accessed at node h at time t; and These are the rated voltage at the end of line L and the actual voltage under scenario s, respectively. The number of distribution network lines; Let be the active power probability of line L at time t+q; This represents the maximum transmittable power of line L;

[0024] The model is constrained, including power balance constraints, node voltage amplitude constraints, branch transmission capacity constraints, flexible interconnection device operation constraints, and energy storage system charging and discharging constraints.

[0025] The power balance constraint formula is as follows:

[0026] ;

[0027] ;

[0028] in, and These represent the total active power and reactive power injected into node j in scenario s at time t, respectively. and These represent the active power and reactive power flowing from node j to downstream node d in scenario s at time t, respectively. and These represent the active power and reactive power flowing from upstream node i to node j in scenario s at time t, respectively. , and Let i, j be the resistance, reactance, and current of the branch (i,j) in scenario s, respectively.

[0029] The formula for constraining the node voltage amplitude is as follows:

[0030] ;

[0031] in, and These are the upper and lower limits of the voltage at node i, respectively;

[0032] The formula for constraining the transmission capacity of the branch is as follows:

[0033] ;

[0034] in, This represents the upper limit of the current amplitude in branch ij;

[0035] The operational constraint formula for the flexible interconnection device is as follows:

[0036] VSC constraints:

[0037] ;

[0038] ;

[0039] ;

[0040] in, , and These represent the maximum active power, reactive power, and capacity limit of the k-th VSC output, respectively. Let be the active power output of the k-th VSC AC side in the s-th scenario at time t. Let be the reactive power output of the k-th VSC in the s-th scenario at time t. Let be the reactive power output of the k-th VSC AC side in the s-th scenario at time t;

[0041] B2B-VSC constraints:

[0042] ;

[0043] ;

[0044] ;

[0045] in, and These represent the active power injected by B2B-VSC at nodes i and j in scenario s at time t, respectively. and These represent the reactive power injected by B2B-VSC at nodes i and j in scenario s at time t, respectively. and These are the B2B-VSC loss power of nodes i and j in scenario s at time t, respectively. The loss factor for B2B-VSC; Let be the active power loss of the converter on the i-side of node B2B-VSC at time t. Let t be the active power loss of the converter on the j-side of the B2B-VSC node;

[0046] The energy storage system's charge and discharge constraint formula is as follows:

[0047] ;

[0048] ;

[0049] ;

[0050] in, Let z be the active power output of the z-th ESS in scenario s at time t; Let z be the reactive power output of the z-th ESS in scenario s at time t; , and These represent the capacity, maximum active power, and maximum reactive power of the z-th ESS, respectively.

[0051] The aforementioned real-time rolling optimization method for distribution networks, wherein the improved multi-objective elk optimization algorithm includes:

[0052] A chaotic sequence is generated using a symmetric Tent mapping and mapped to the value range of the decision variable to obtain the initial population. The ratio of stags is dynamically adjusted according to the number of iterations, and the population iteration is completed through mating period, breeding period, and selection period to output the Pareto optimal solution set.

[0053] The expression for the symmetric Tent mapping is:

[0054] ;

[0055] in, This represents the state value during the (n+1)th iteration. This represents the state value at the nth iteration, which is the input variable for the current iteration step. For mapping control parameters;

[0056] The ratio of male deer is dynamically adjusted based on the number of iterations, so that the number of male deer decreases during the iteration process, in order to improve the global optimization accuracy. The formula is as follows:

[0057] ;

[0058] in, This represents the proportion of male deer in the vth iteration; and These represent the upper limit and the range of change of the dynamic selection probability, respectively. This represents the maximum number of iterations.

[0059] The aforementioned real-time rolling optimization method for distribution networks, in which the determination of whether the optimization cycle has ended includes: solving the optimization problem for the next N0 steps at each time t, executing only the control command for the current step, and then resolving the next window at time t=t+1 using the latest measured value to correct the prediction; when the time step advances to the end of the preset scheduling cycle, that is, when all time period control commands have been issued and executed, the rolling optimization cycle ends and the entire process terminates.

[0060] The aforementioned real-time rolling optimization method for distribution networks, wherein the real-time correction mechanism includes:

[0061] At the beginning of each optimization cycle, the measured values ​​of wind power output, photovoltaic power output, and load power at time t are collected and compared with the corresponding predicted values ​​obtained in the previous optimization cycle. The prediction residual vectors of wind power, photovoltaic power, and load are constructed, as shown in the following formula:

[0062] ;

[0063] in, To predict residual vectors in real time; , , These represent the measured values ​​of wind power, photovoltaic power, and load at time t, respectively. , , These represent the predicted values ​​for wind power, solar power, and load at time t, respectively. Let be the predicted residual of wind power output at time t. Let be the predicted residual of photovoltaic output at time t. Let be the predicted residual of the load power at time t;

[0064] The residual trend is calculated based on the predicted residual vectors of adjacent optimization periods, using the following formula:

[0065] ;

[0066] in, Calculate the residual trend for the predicted residual vector of adjacent optimization periods;

[0067] The real-time correlation changes among wind power, photovoltaic and load forecast deviations are calculated based on the historical residual sequence within the sliding time window.

[0068] Based on the residual vector, residual trend, and real-time correlation changes, a timely correction factor is constructed to correct the wind power, photovoltaic, and load forecast inputs for the next optimization cycle. The formula is as follows:

[0069] ;

[0070] in, This serves as the corrected forecast input for wind power, solar power, and load. Forecast inputs for wind power, solar power, and load before correction; This represents the time-domain attenuation coefficient that decreases with the prediction period. To correct the gain matrix; This is the residual trend compensation coefficient;

[0071] Based on the corrected real-time residual distribution, the probabilities of deviation scenarios are reorganized so that deviation scenarios that are closer to the real-time residual direction and magnitude receive higher probability weights.

[0072] The aforementioned real-time rolling optimization method for distribution networks, wherein the reorganization of the probability of deviation scenarios includes:

[0073] Based on timely prediction of residual vector The deviation vector corresponding to the s-th deviation scenario The distance between them is used to construct a scene matching index, and the formula is as follows:

[0074] ;

[0075] in, Let be the scene matching index corresponding to the s-th deviation scene at time t, which represents the Euclidean distance between the deviation scene vector and the corrected real-time residual vector. The smaller the value, the higher the matching degree between the deviation scene and the real-time running state.

[0076] The probability of updating the s-th deviation scenario is given by the following formula:

[0077] ;

[0078] in, This represents the probability of the s-th deviation scenario at time t after real-time correction. Let be the probability of the s-th deviation scenario at the previous time step; exp is the natural exponential function. The scaling parameter is determined based on the degree of fluctuation in the predicted residual within the sliding time window; This represents the total number of deviation scenarios;

[0079] By redistributing probabilities, scenarios that are closer to the actual deviation evolution trend in real-time operation are given higher weight in the next optimization cycle.

[0080] Secondly, the present invention provides a real-time rolling optimization system for a power distribution network, comprising:

[0081] The ultra-short-term forecasting module is used to fit the marginal distribution of forecast deviations based on the actual values ​​of wind power, solar power, and load at the current moment. It constructs a three-dimensional joint distribution of wind power, solar power, and load using R-copula, and performs Monte Carlo sampling and reduction on the three-dimensional joint distribution to obtain the correlation scenario of wind power, solar power, and load forecast deviations. The correlation scenario of wind power, solar power, and load forecast deviations is introduced into the AMRA ultra-short-term forecasting model. The actual values ​​of wind power, solar power, and load at the current moment are input to obtain the predicted values ​​of wind power, solar power, and load for the next n0 time periods.

[0082] The scheduling module is used to construct a multi-objective rolling optimization model based on stochastic model predictive control based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, and solve the model using an improved multi-objective stag optimization algorithm to obtain control commands.

[0083] The decision correction output module is used to determine whether the optimization cycle has ended. If so, it outputs the final decision scheme according to the control command. If not, it introduces a timely residual feedback and a real-time correction mechanism for deviation scenario probabilities to make the optimization time domain roll forward and repeatedly call the above modules until the whole-day scheduling of the power grid is completed and the final decision scheme is output.

[0084] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0085] This invention provides a method and system for real-time rolling optimization of distribution networks. By accurately characterizing the correlation of deviations, improving the optimization algorithm, and introducing real-time correction, it enhances prediction accuracy, fully leverages the control capabilities of flexible interconnection devices, reduces network losses, improves voltage quality, and enhances the robustness and economy of distribution network operation.

[0086] This invention introduces R-copula to establish a three-dimensional joint distribution of wind power, photovoltaic and load forecasting deviations, which can characterize the correlation and coupling relationship between forecasting errors of different random variables, avoiding the problem of ignoring the correlation of deviations in traditional independent modeling methods; at the same time, it combines ARMA model to generate forecasting inputs containing correlated deviations, making the forecasting results closer to the actual operating state.

[0087] This invention employs an improved multi-objective elk optimization algorithm, Tent-AMEHO, which enhances the uniformity of the initial population through Tent chaotic mapping and utilizes an adaptive adjustment strategy for the ratio of male elk to balance global search and local exploitation capabilities. This improves the algorithm's convergence accuracy and the quality of the Pareto solution set distribution, thereby reducing the risk of traditional intelligent optimization algorithms getting trapped in local optima.

[0088] This invention employs a stochastic model predictive control rolling optimization framework and introduces a real-time residual feedback and a probabilistic real-time correction mechanism for deviation scenarios. This allows the prediction input for the next optimization time domain to be dynamically corrected based on real-time measurement data, reducing the cumulative impact of prediction errors and fully leveraging the control capabilities of flexible interconnection devices and energy storage systems. Simultaneously, based on real-time residual feedback, the deviation scenarios that better fit the changes in the next moment are reweighted, thereby reducing network losses, improving voltage quality, and balancing feeder loads. Attached Figure Description

[0089] Figure 1 This is a schematic diagram of the process steps of a real-time rolling optimization method for a power distribution network in this invention;

[0090] Figure 2 This is a schematic diagram of the prediction considering the correlation between wind and solar load prediction deviation in this invention;

[0091] Figure 3 This is a schematic diagram illustrating the process of using a stochastic model predictive control optimization strategy in this invention;

[0092] Figure 4 This is a schematic diagram of the rolling optimization process of the present invention;

[0093] Figure 5 A flowchart illustrating the process of improving the multi-objective elk optimization algorithm. Detailed Implementation

[0094] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0095] The term "and / or" simply describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. Additionally, the character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0096] Example 1:

[0097] like Figure 1 As shown in the figure, this embodiment introduces a real-time rolling optimization method for distribution networks, including:

[0098] Based on the actual values ​​of wind power, photovoltaic power, and load at the current moment, the marginal distribution of prediction deviation is fitted. A three-dimensional joint distribution of wind power, photovoltaic power, and load is constructed using R-Teng Copula. Monte Carlo sampling and reduction are then performed on the three-dimensional joint distribution to obtain the correlation scenario of prediction deviation of wind power, photovoltaic power, and load.

[0099] The aforementioned scenario of wind power, solar power, and load forecast deviation correlation is introduced into the AMRA ultra-short-term forecast model. By inputting the actual values ​​of wind power, solar power, and load at the current moment, the predicted values ​​of wind power, solar power, and load for the next n0 time periods are obtained.

[0100] Based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, a multi-objective rolling optimization model based on stochastic model predictive control is constructed. An improved multi-objective stag optimization algorithm is used to solve the model to obtain control commands.

[0101] If the optimization cycle has ended, the final decision scheme is output according to the control command. If not, the optimization time domain is rolled forward by introducing timely residual feedback and a real-time correction mechanism for deviation scenario probabilities, and the above steps are repeated until the grid's full-day scheduling is completed and the final decision scheme is output.

[0102] Among them, the prediction that takes into account the correlation between wind and solar load prediction bias, such as Figure 2 As shown, the marginal distributions of the prediction deviations for each fitted variable are as follows: the prediction deviations for load and photovoltaic power are fitted using a normal distribution, while the prediction deviations for wind power are fitted using a Beta distribution. The probability density function of the prediction deviations for wind and solar loads is as follows:

[0103] The deviations in load and photovoltaic output predictions follow a normal distribution, i.e. Its probability density function is:

[0104]

[0105] In the formula, and These represent the expected value and standard deviation of the prediction bias, respectively.

[0106] The prediction error of active power output from wind turbines exhibits a significant double-peak, thick-tailed characteristic. Modeling this error using only a normal distribution will lead to substantial errors, while the Beta distribution can more accurately characterize the distribution of this deviation. Its probability density function can be expressed as:

[0107]

[0108] In the formula, and These are the shape parameter and the scale parameter, respectively, which can be obtained from historical data.

[0109] Specifically, based on the marginal distribution of prediction biases for each variable of wind and solar load, a three-dimensional joint distribution of wind and solar load is constructed using R-vine Copula. Monte Carlo sampling and reduction are then applied to this three-dimensional joint distribution to obtain the correlation scenario of wind and solar load prediction biases. The specific steps are as follows:

[0110] Step 1: Based on the three sets of random variables representing the prediction deviations of photovoltaic, wind turbine, and load... , , The probability density function is obtained. , , and marginal distribution , , .

[0111] Step 2: Use Kendall's rank correlation coefficient As an edge in the R-vine structure The weighting index is optimized by improving the topology of the vine structure to achieve the optimal characterization of the correlation between prediction errors of high-dimensional wind-solar-load.

[0112] Step 3: Select the Pair Copula function for each edge in the vine structure, and estimate the parameters based on the BIC criterion to construct a high-dimensional joint distribution function. .

[0113] Step four: Generate a joint bias scenario by performing Monte Carlo sampling on the high-dimensional bias joint distribution. .

[0114] Step 5: Combined bias scenario reduction uses K-means clustering.

[0115] Ultimately, the prediction bias random variable The possible values ​​for are as follows.

[0116]

[0117] In the formula, Number of scenes For the scene The corresponding probability, for Moment Scene The predicted deviation value.

[0118] Furthermore, the ultra-short-term prediction that takes into account the correlation between wind and solar load prediction bias is introduced into the autoregressive moving average (ARMA) prediction model:

[0119]

[0120] In the formula, and These are the orders of autoregression and moving average, respectively; and The correlation coefficient; express Time Prediction Real-time load, wind turbine and photovoltaic power values; for The predicted value at any given time; To predict the bias random variable, in Actual output at time With predicted output The prediction deviation between them is .

[0121] Furthermore, a stochastic model predictive control optimization strategy is adopted during the rolling optimization process, such as... Figure 3 The system described above addresses random fluctuations in distributed power sources and loads by periodically updating the system state. Its operating mechanism is as follows: at each sampling time, the optimal control sequence within a finite time domain is solved based on the latest predicted data. Only the control command for the first time period is executed. When the next sampling time arrives, the optimization window rolls forward and the optimization process is repeated, as shown below. Figure 4 As shown. In the specific rolling optimization process, the flexible interconnected distribution network aims to minimize system losses. Minimum system voltage deviation and load imbalance The objective function is:

[0122]

[0123]

[0124]

[0125] In the formula: For the rolling phase The probability of a scenario; This represents the number of scenes after scene reduction; and This refers to the number of branches in the AC and DC sections of the system. Let be the current at the end of the Lth branch in the sth scenario at time t+q. The cycle interval for rolling control is set to 15 minutes. The optimization cycle is a rolling cycle. Since the control cycle is 15 minutes and the optimization cycle is 1 hour, therefore... =4; This refers to the number of feeders in the distribution network. for The active power of feeder L at any given time; Let L be the maximum transmittable power of feeder L. Constraints satisfy fundamental constraints such as system power flow equations, node voltage limits, and line current capacity, as well as constraints on flexible interconnection devices and energy storage systems during the operation of the flexible interconnected distribution network.

[0126] Furthermore, to solve the aforementioned multi-objective function, an improved multi-objective elk optimization algorithm is used to solve the model. Specifically, the initial elk population is generated using the Tent chaotic mapping: a chaotic sequence is generated using a symmetric Tent mapping, and then mapped to the value range of the decision variables to obtain the initial population. The expression for the symmetric Tent mapping is:

[0127]

[0128] In the formula, For mapping control parameters (typically taking values ​​in the range of 1000-10000) ).when When the mapping degenerates into a symmetric Tent mapping, the generated sequence has optimal traversal uniformity, while also possessing deep chaotic characteristics and stronger spatial coverage.

[0129] Furthermore, in the improved multi-objective elk optimization algorithm solution model of the rolling optimization, step three involves dynamically adjusting the proportion of male elk based on the number of iterations. During the iterative update process, the number of male elk exhibits a decreasing trend, thereby improving the global optimization accuracy. The formula is as follows:

[0130]

[0131] In the formula, This indicates the proportion of male deer in the v-th iteration; and These represent the upper limit and the range of change of the dynamic selection probability, respectively. The value of is in the range of (0,1).

[0132] Furthermore, the improved multi-objective elk optimization algorithm's workflow framework, such as... Figure 5 As shown. The specific solution steps in flexible interconnected distribution networks are as follows:

[0133] Step 1: Initialize the Tent chaotic population. The initial population is generated using the Tent chaotic mapping, which utilizes its uniform traversal property to solve the problems of insufficient diversity and easy getting trapped in local optima in traditional random initialization.

[0134] Step 2: Map the individuals in the population to the decision variables for day-ahead optimization of the flexible interconnected distribution network, substitute them into a multi-objective function with the objectives of minimizing system network loss and node voltage deviation, and calculate the fitness value of each individual.

[0135] Step 3: Adopt an adaptive partitioning strategy to dynamically adjust the ratio of male deer based on the current iteration number: maintain a high ratio of male deer in the early stage of iteration to expand the global search scope, and gradually reduce the ratio of male deer in the later stage of iteration to focus on local fine development, so as to achieve a dynamic balance between global exploration and local development capabilities.

[0136] Step 4: Select the individuals with the best fitness as male deer, and assign female deer to families using a roulette wheel method; based on the offspring index affiliation, adopt two differentiated strategies: paternal lineage update with random perturbation and maternal lineage update with fusion of parental genes, to generate an offspring population that combines diversity and superior genes.

[0137] Step 5: Update and merge all individuals from the parent and child generations, recalculate fitness and sort them, retain the top N best individuals as the next generation population, and complete one iteration cycle.

[0138] Step 6: Repeat steps 2 through 5 until the maximum number of iterations is reached.

[0139] Furthermore, the rolling optimization method, wherein the real-time correction mechanism includes the following steps: at the beginning of each control cycle t, the measured values ​​of wind power output, photovoltaic power output, and load power at time t are collected and compared with the corresponding predicted values ​​obtained in the previous control cycle to construct a wind-solar-load prediction residual vector:

[0140]

[0141] in, , , These represent the measured values ​​of wind power, photovoltaic power, and load at time t, respectively. , These represent the predicted values ​​of wind power, solar power, and load at time t, respectively. To predict the residual vector in real time; calculate the residual trend based on the predicted residual vectors of adjacent control cycles:

[0142]

[0143] The real-time correlation change among wind power, photovoltaic, and load forecast deviations is calculated based on the historical residual sequence within the sliding time window. An adaptive correction is constructed based on the forecast residuals, residual trends, and real-time correlation changes to correct the wind, solar, and load forecast input for the g-th forecast period in the next optimization domain.

[0144]

[0145] in, This is the input for the wind-solar load prediction before correction. This is the input for the corrected wind-solar load prediction. To correct the gain matrix, The time-domain decay coefficient decreases with the prediction period. This is the residual trend compensation coefficient;

[0146] Based on the corrected prediction residual distribution, the probability of deviation scenarios is redistributed, so that deviation scenarios that are closer to the real-time residual direction and magnitude receive higher probability weights.

[0147] Furthermore, the deviation scenario probability reallocation specifically involves redistributing the probability based on the real-time predicted residual vector. The deviation vector corresponding to the s-th deviation scenario The distance between them is used to construct a scene matching index:

[0148]

[0149] And update the probability of the s-th deviation scenario according to the following formula:

[0150]

[0151] in, Let be the probability of the s-th deviation scenario in the previous control cycle. This represents the probability of the s-th deviation scenario after real-time correction. The scaling parameter is determined based on the degree of fluctuation in the predicted residuals within the sliding time window. This represents the total number of deviation scenarios.

[0152] By redistributing the probabilities as described above, scenarios that are closer to the actual deviation evolution trend in real-time operation are given higher weight in the next rolling optimization time domain.

[0153] Finally, based on the ultra-short-term forecast values, the improved multi-objective elk optimization algorithm is used to solve the multi-objective rolling optimization model of the flexible interconnected distribution network to obtain the Pareto optimal solution set. The optimal scheduling scheme is selected by the minimum deviation method, and the day-ahead scheduling instructions of the flexible interconnected device and energy storage system are output.

[0154] Example 2:

[0155] Based on the same inventive concept as Embodiment 1, this embodiment introduces a real-time rolling optimization system for a distribution network, comprising:

[0156] The ultra-short-term forecasting module is used to fit the marginal distribution of forecast deviations based on the actual values ​​of wind power, solar power, and load at the current moment. It constructs a three-dimensional joint distribution of wind power, solar power, and load using R-copula, and performs Monte Carlo sampling and reduction on the three-dimensional joint distribution to obtain the correlation scenario of wind power, solar power, and load forecast deviations. The correlation scenario of wind power, solar power, and load forecast deviations is introduced into the AMRA ultra-short-term forecasting model. The actual values ​​of wind power, solar power, and load at the current moment are input to obtain the predicted values ​​of wind power, solar power, and load for the next n0 time periods.

[0157] The scheduling module is used to construct a multi-objective rolling optimization model based on stochastic model predictive control based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, and solve the model using an improved multi-objective stag optimization algorithm to obtain control commands.

[0158] The decision correction output module is used to determine whether the optimization cycle has ended. If so, it outputs the final decision scheme according to the control command. If not, it introduces a timely residual feedback and a real-time correction mechanism for deviation scenario probabilities to make the optimization time domain roll forward and repeat the above steps until the whole-day scheduling of the power grid is completed and the final decision scheme is output.

[0159] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A real-time rolling optimization method for a power distribution network, characterized in that, include: Based on the actual values ​​of wind power, photovoltaic power and load at the current moment, the marginal distribution of prediction deviation is fitted. A three-dimensional joint distribution of wind power, photovoltaic power and load is constructed by R-Teng Copula. Monte Carlo sampling and reduction are performed on the three-dimensional joint distribution to obtain the correlation scenario of prediction deviation of wind power, photovoltaic power and load. The aforementioned scenario of correlation between wind power, solar power and load forecast deviations is introduced into the AMRA ultra-short-term forecast model. By inputting the actual values ​​of wind power, solar power and load at the current moment, the predicted values ​​of wind power, solar power and load for the next n0 time periods are obtained. Based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, a multi-objective rolling optimization model based on stochastic model predictive control is constructed, and an improved multi-objective stag optimization algorithm is used to solve the model to obtain control commands. If the optimization cycle has ended, the final decision scheme is output according to the control command. If not, the optimization time domain is rolled forward by introducing timely residual feedback and a real-time correction mechanism for deviation scenario probabilities, and the above steps are repeated until the grid's full-day scheduling is completed and the final decision scheme is output.

2. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The method of fitting and predicting the marginal distribution of deviations based on the actual values ​​of wind power, photovoltaic power, and load at the current moment includes: The load and photovoltaic forecast deviations were fitted using a normal distribution, while the wind power forecast deviations were fitted using a Beta distribution.

3. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The formula for the correlation scenario of wind power, photovoltaic and load forecast deviation is as follows: ； in, For time t+g, the correlation scenario of wind power, photovoltaic and load forecast deviations is represented; N is the number of scenarios after scenario reduction. Let be the probability corresponding to scenario s; The prediction deviation value for scenario s at time t+g; g is the prediction time period; The scenario of correlation between wind power, solar power, and load forecast deviation is introduced into the AMRA ultra-short-term forecasting model, as shown in the following formula: ； in, The probability values ​​of wind power, photovoltaic power, and load at time t+g are predicted for time t; m and n are the orders of autoregression and moving average, respectively; q is the lag order index in the ARMA model; t+gq is the lag time corresponding to the qth time period before the target prediction time t+g. These are the autoregressive coefficients; The moving average coefficient; The predicted value at time t+gq; This is a random variable representing the prediction bias.

4. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The construction of the multi-objective rolling optimization model includes: With the optimization objectives of minimizing system active power loss, minimizing node voltage deviation, and minimizing feeder load balance, a multi-objective optimization objective function is constructed, as shown in the following formula: ； ； ； in, , , These are the optimization objective functions for minimizing system active power loss, minimizing node voltage deviation, and optimizing feeder load balance, respectively. Let be the probability corresponding to scenario s; This represents the number of scenes after scene reduction; and These represent the number of branches in the AC and DC sections of the system, respectively. and These are the active power and reactive power at the end of the Lth branch in the s-th scenario, respectively; Let L be the impedance of the Lth branch; To optimize the rolling cycle; The number of VSCs; Number of B2B-VSCs; Let be the power loss of the VSC connected at node d at time t; Let be the power loss of the B2B-VSC accessed at node h at time t; and These are the rated voltage at the end of line L and the actual voltage under scenario s, respectively. The number of distribution network lines; Let be the active power probability of line L at time t+q; This represents the maximum transmittable power of line L; The model is constrained, including power balance constraints, node voltage amplitude constraints, branch transmission capacity constraints, flexible interconnection device operation constraints, and energy storage system charging and discharging constraints. The power balance constraint formula is as follows: ； ； in, and These represent the total active power and reactive power injected into node j in scenario s at time t, respectively. and These represent the active power and reactive power flowing from node j to downstream node d in scenario s at time t, respectively. and These represent the active power and reactive power flowing from upstream node i to node j in scenario s at time t, respectively. , and Let i, j be the resistance, reactance, and current of the branch (i,j) in scenario s, respectively. The formula for constraining the node voltage amplitude is as follows: ； in, and These are the upper and lower limits of the voltage at node i, respectively; The formula for constraining the transmission capacity of the branch is as follows: ； in, This represents the upper limit of the current amplitude in branch ij; The operational constraint formula for the flexible interconnection device is as follows: VSC constraints: ； ； ； in, , and These represent the maximum active power, reactive power, and capacity limit of the k-th VSC output, respectively. Let be the active power output of the k-th VSC AC side in the s-th scenario at time t. Let be the reactive power output of the k-th VSC in the s-th scenario at time t. Let be the reactive power output of the k-th VSC AC side in the s-th scenario at time t; B2B-VSC constraints: ； ； ； in, and These represent the active power injected by B2B-VSC at nodes i and j in scenario s at time t, respectively. and These represent the reactive power injected by B2B-VSC at nodes i and j in scenario s at time t, respectively. and These are the B2B-VSC loss power of nodes i and j in scenario s at time t, respectively. The loss factor for B2B-VSC; Let be the active power loss of the converter on the i-side of node B2B-VSC at time t. Let t be the active power loss of the converter on the j-side of the B2B-VSC node; The energy storage system's charge and discharge constraint formula is as follows: ； ； ； in, Let z be the active power output of the z-th ESS in scenario s at time t; Let z be the reactive power output of the z-th ESS in scenario s at time t; , and These represent the capacity, maximum active power, and maximum reactive power of the z-th ESS, respectively.

5. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The improved multi-objective elk optimization algorithm includes: A chaotic sequence is generated using a symmetric Tent mapping and mapped to the value range of the decision variable to obtain the initial population. The ratio of stags is dynamically adjusted according to the number of iterations, and the population iteration is completed through mating period, breeding period, and selection period to output the Pareto optimal solution set. The expression for the symmetric Tent mapping is: ； in, This represents the state value during the (n+1)th iteration. This represents the state value at the nth iteration, which is the input variable for the current iteration step. For mapping control parameters; The ratio of male deer is dynamically adjusted based on the number of iterations, so that the number of male deer decreases during the iteration process, in order to improve the global optimization accuracy. The formula is as follows: ； in, This represents the proportion of male deer in the vth iteration; and These represent the upper limit and the range of change of the dynamic selection probability, respectively. This represents the maximum number of iterations.

6. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The determination of whether the optimization cycle has ended includes: solving the optimization problem for the next N0 steps at each time t, executing only the control instruction for the current step, and then correcting the prediction with the latest measured value at time t=t+1 and solving the next window again; when the time step advances to the end of the preset scheduling cycle, that is, when all time period control instructions have been issued and executed, the rolling optimization cycle ends and the entire process terminates.

7. The real-time rolling optimization method for distribution networks according to claim 1, characterized in that, The real-time correction mechanism includes: At the beginning of each optimization cycle, the measured values ​​of wind power output, photovoltaic power output, and load power at time t are collected and compared with the corresponding predicted values ​​obtained in the previous optimization cycle. The prediction residual vectors of wind power, photovoltaic power, and load are constructed, as shown in the following formula: ； in, To predict residual vectors in real time; , , These represent the measured values ​​of wind power, photovoltaic power, and load at time t, respectively. , , These represent the predicted values ​​for wind power, solar power, and load at time t, respectively. Let be the predicted residual of wind power output at time t. Let be the predicted residual of photovoltaic output at time t. Let be the predicted residual of the load power at time t; The residual trend is calculated based on the predicted residual vectors of adjacent optimization periods, using the following formula: ； in, Calculate the residual trend for the predicted residual vector of adjacent optimization periods; The real-time correlation changes among wind power, photovoltaic and load forecast deviations are calculated based on the historical residual sequence within the sliding time window. Based on the residual vector, residual trend, and real-time correlation changes, a timely correction factor is constructed to correct the wind power, photovoltaic, and load forecast inputs for the next optimization cycle. The formula is as follows: ； in, This serves as the corrected forecast input for wind power, solar power, and load. Forecast inputs for wind power, solar power, and load before correction; This represents the time-domain attenuation coefficient that decreases with the prediction period. To correct the gain matrix; This is the residual trend compensation coefficient; Based on the corrected real-time residual distribution, the probabilities of deviation scenarios are reorganized so that deviation scenarios that are closer to the real-time residual direction and magnitude receive higher probability weights.

8. The real-time rolling optimization method for distribution networks according to claim 7, characterized in that, The reorganization of the probabilities of the deviation scenarios includes: Based on timely prediction of residual vector The deviation vector corresponding to the s-th deviation scenario The distance between them is used to construct a scene matching index, and the formula is as follows: ； in, Let be the scene matching index corresponding to the s-th deviation scene at time t, which represents the Euclidean distance between the deviation scene vector and the corrected real-time residual vector. The smaller the value, the higher the matching degree between the deviation scene and the real-time running state. The probability of updating the s-th deviation scenario is given by the following formula: ； in, This represents the probability of the s-th deviation scenario at time t after real-time correction. Let be the probability of the s-th deviation scenario at the previous time step; exp is the natural exponential function. The scaling parameter is determined based on the degree of fluctuation in the predicted residual within the sliding time window; This represents the total number of deviation scenarios. By redistributing probabilities, scenarios that are closer to the actual deviation evolution trend in real-time operation are given higher weight in the next optimization cycle.

9. A real-time rolling optimization system for a power distribution network, characterized in that, include: The ultra-short-term forecasting module is used to fit the marginal distribution of forecast deviations based on the actual values ​​of wind power, solar power, and load at the current moment. It constructs a three-dimensional joint distribution of wind power, solar power, and load using R-copula, and performs Monte Carlo sampling and reduction on the three-dimensional joint distribution to obtain the correlation scenario of wind power, solar power, and load forecast deviations. The correlation scenario of wind power, solar power, and load forecast deviations is introduced into the AMRA ultra-short-term forecasting model. The actual values ​​of wind power, solar power, and load at the current moment are input to obtain the predicted values ​​of wind power, solar power, and load for the next n0 time periods. The scheduling module is used to construct a multi-objective rolling optimization model based on stochastic model predictive control based on the predicted values ​​of wind power, photovoltaic power and load for the next n0 time periods, and solve the model using an improved multi-objective stag optimization algorithm to obtain control commands. The decision correction output module is used to determine whether the optimization cycle has ended. If so, it outputs the final decision scheme according to the control command. If not, it introduces a timely residual feedback and a real-time correction mechanism for deviation scenario probabilities to make the optimization time domain roll forward and repeatedly call the above modules until the whole-day scheduling of the power grid is completed and the final decision scheme is output.

Images