A power distribution line load forecasting and optimal scheduling method and system thereof

Abstract

The application discloses a power distribution line load prediction and optimal scheduling method and system, relates to the technical field of intelligent scheduling of power systems, and comprises the following steps: generating time-aligned multi-source fusion input data sets; constructing a hybrid time series load prediction model; in the model training process, introducing a weighted quantile loss function; constructing a joint probability distribution model of renewable energy output and demand response participation rate, sampling the joint probability distribution; constructing a rolling time domain power distribution network optimal scheduling model; using a two-layer hybrid strategy to process uncertainty, decomposing the optimization problem into multiple sub-problems and using an alternating direction multiplier method with an adaptive penalty parameter to solve the problem in a distributed manner. The application establishes a multi-objective optimal scheduling model in a rolling time domain, realizes dynamic closed-loop optimization, and through the introduction of a two-layer hybrid solving strategy and an ADMM distributed algorithm with an adaptive penalty parameter, ensures global consistency, and significantly improves the calculation efficiency and scalability.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent scheduling of power systems, and particularly to a power distribution line load prediction and optimal scheduling method and system. BACKGROUND

[0002] With the rapid access of new energy, distributed power, electric vehicles, and demand response, the operation characteristics of distribution networks are gradually evolving from the traditional "one-way power flow and deterministic load" to the complex structure of "multi-source heterogeneity, two-way power flow, and strong randomness". The traditional distribution network scheduling mode mainly relies on experience rules or deterministic optimization methods, which cannot effectively cope with the random fluctuations of load and renewable energy output, resulting in problems such as voltage out-of-limit of distribution lines, increase of active power loss, and frequent triggering of equipment operation constraints. In the prior art, the load prediction methods for distribution networks mainly include time series analysis, neural network prediction, and grey prediction methods. These methods have certain accuracy in short-term prediction, but most of them only focus on a single time scale or a single prediction index, ignoring the non-stationarity of load characteristics and the correlation of multiple time levels; at the same time, the correlation between renewable energy output (such as photovoltaic and wind power) and user response behavior is not well described, resulting in that the prediction results cannot fully reflect the system uncertainty. At the optimization scheduling level, the existing distribution network scheduling methods mostly use a centralized solving framework, which models and solves the system power flow constraints, equipment operation constraints, economic objectives, etc. This method can be realized in small and medium-sized systems, but in intelligent distribution networks containing a large number of distributed energy and multi-layer control nodes, there are problems of high computational complexity, large data communication pressure, and difficulty in balancing privacy protection. In addition, the traditional deterministic scheduling model fails to fully consider the influence of prediction errors and scenario fluctuations, which easily causes the mismatch between scheduling plan and actual operation state.

[0003] In recent years, the rolling time domain optimization idea and distributed solving algorithm have been gradually introduced by academia and industry to realize dynamic scheduling and hierarchical collaborative control of distribution networks. However, the existing schemes generally have the following shortcomings:

[0004] The uncertainty distribution characteristics of load prediction are not fully combined in the rolling optimization process, and prediction is disconnected from scheduling; multi-objective optimization lacks a unified weighting and dynamic weight adjustment mechanism, making it difficult to simultaneously consider economy, safety, and user comfort; the distributed solving process fails to achieve adaptive coordination between sub-networks, resulting in slow convergence speed or inconsistent solutions.

[0005] Therefore, there is an urgent need for a power distribution line load prediction and optimal scheduling method that can jointly consider load prediction results, uncertainty modeling, and distributed solving mechanisms in the rolling time domain, to realize high-precision prediction, high-robustness scheduling, and real-time collaborative optimization of complex distribution network systems. SUMMARY

[0006] In view of the above problems, the present application is proposed.

[0007] Therefore, the problem to be solved by the present application is how to realize high-precision load prediction and rolling time domain optimization scheduling of distribution lines under the condition of high proportion of renewable energy access and high uncertainty of load behavior, so as to balance the economy, safety and reliability of system operation.

[0008] To solve the above technical problems, the present application provides the following technical solutions:

[0009] In the first aspect, the present application embodiment provides a distribution line load prediction and optimization scheduling method, including collecting and storing distribution network operation related data, and performing data quality processing to generate a time-aligned multi-source fusion input data set;

[0010] Taking the multi-source fusion input data set as input, a hybrid time series load prediction model is constructed, and in the model training process, a weighted quantile loss function is introduced to optimize the prediction accuracy of the prediction mean and multiple quantile points;

[0011] Based on the uncertainty characteristics of the distributed energy monitoring data and meteorological elements in the distribution network operation related data, a joint probability distribution model of renewable energy output and demand response participation rate is constructed, and a distribution-based scenario generation method is used to sample the joint probability distribution to generate a set of multiple renewable energy output and demand response participation rate scenarios in the future time domain;

[0012] A rolling time domain distribution network optimization scheduling model is constructed, and the weighted multi-objective function is used to jointly optimize the distribution network operation performance in the prediction time range of the rolling time domain;

[0013] The two-layer hybrid strategy is used to process uncertainty, and the optimization problem is decomposed into multiple sub-problems and distributedly solved using an alternating direction multiplier method with an adaptive penalty parameter.

[0014] As a preferred scheme of the distribution line load prediction and optimization scheduling method of the present application, wherein:

[0015] The hybrid time series load prediction model includes:

[0016] A spatial encoder based on a graph neural network is used to extract spatial features of the topological structure and electrical correlation between nodes of the distribution line;

[0017] A time series decoder based on an improved self-attention mechanism is used to model the time series dependence of load change characteristics at different time scales;

[0018] An uncertainty estimation module based on Monte Carlo Dropout and deep belief network is used to model the probability distribution of the prediction result, and output the load probability distribution P(L t ) of each prediction time t.

[0019] As a preferred scheme of the power distribution line load prediction and optimal scheduling method, the joint probability distribution model of the renewable energy output and the demand response participation rate is constructed, including:

[0020] A weighted quantile loss function is introduced to optimize the prediction accuracy of the prediction mean value and multiple quantile points, including:

[0021] The weighted quantile loss function jointly optimizes the prediction results of different quantile points q∈ ={0.1,0.5,0.9}, and its form is:

[0022] ;

[0023] Wherein, is the weighted quantile loss function, y is the true load value, is the prediction value corresponding to the quantile point q, is the weighted coefficient of each quantile point q, ( ) is the quantile loss function.

[0024] The definition of the quantile loss function is:

[0025] ;

[0026] Wherein, is the prediction error;

[0027] In the model training process, a dynamic weighting mechanism is adopted to dynamically adjust the quantile error on the validation set to jointly optimize the prediction mean value and the interval uncertainty.

[0028] As a preferred scheme of the power distribution line load prediction and optimal scheduling method, the joint probability distribution model of the renewable energy output and the demand response participation rate is constructed, including:

[0029] Edge probability models are established for photovoltaic output, wind power and demand response rate respectively; Gaussian mixture distribution is used to fit photovoltaic and wind power output to capture their asymmetry under extreme weather conditions; the demand response participation rate is modeled using truncated normal distribution or lognormal distribution according to the historical response probability characteristics;

[0030] To represent the dependence relationship between different variables, a joint distribution based on Copula function is constructed:

[0031] ;

[0032] where C( X, Y) is the correlation function, is the joint cumulative distribution function, is the marginal cumulative distribution function of the photovoltaic output, is the marginal cumulative distribution function of the wind power, is the marginal cumulative distribution function of the demand response participation rate, and are random variables.

[0033] As a preferred scheme of the power distribution line load prediction and optimal scheduling method, a distribution-based scenario generation method is used to sample the joint probability distribution, including:

[0034] A Copula function model is used to model the nonlinear correlation between meteorological variables and renewable output, or a Gaussian mixture model is used to fit and sample the output distribution characteristics under different operating states.

[0035] In each generated scenario, the occurrence probability corresponding to each generated scenario is determined, and the generated multiple sets of renewable output and demand response participation rate scenario set and the occurrence probability corresponding to each generated scenario are output together as the input of the subsequent optimal scheduling model, to represent the uncertainty of renewable energy output and demand response.

[0036] As a preferred scheme of the power distribution line load prediction and optimal scheduling method, a rolling time domain power distribution network optimal scheduling model is constructed, and within the prediction time range of the rolling time domain, the power distribution network operation performance is jointly optimized in the form of a weighted multi-objective function, including:

[0037] The scheduling time is divided into consecutive rolling periods, each rolling period corresponds to a prediction window, and within each period, optimization calculation is performed based on the latest observation data and prediction results to execute the first-step decision at the current time;

[0038] After each execution, the rolling window is moved forward, data is re-collected, prediction and scenarios are updated, and optimization is repeated to implement closed-loop control; the probabilistic load prediction and renewable energy output, demand response scenarios are combined to form multiple operating scenarios.

[0039] As a preferred scheme of the power distribution line load prediction and optimal scheduling method, a two-layer hybrid strategy is used to handle uncertainty, the optimization problem is decomposed into multiple sub-problems and distributedly solved using an alternating direction multiplier method with an adaptive penalty parameter, including:

[0040] ​​The risk budget constraint is introduced to the key constraint, and the original probability constraint is converted into a solvable deterministic constraint;

[0041] For the short-time fine-tuning stage in the rolling time domain, the model prediction control idea is adopted, each step is optimized starting from the current observation, and the mechanical and electrical resource output, energy storage charging and discharging and demand response are dynamically adjusted in combination with multiple scenarios;

[0042] The decomposed sub-problems are solved by ADMM iteration, each sub-network independently calculates a local optimal solution to ensure that the constraints are met, and the boundary variables are converged and consistency corrected in iteration, the Lagrange multiplier and the penalty term are used to guide the global consistency convergence of each sub-network solution, the ADMM introduces the constraint into the Lagrange form and adds a quadratic penalty term to form an enhanced Lagrange function:

[0043] ;

[0044] Wherein, the enhanced Lagrange function is, the local objective function of the sub-network m is, the Lagrange multiplier is, the quadratic penalty term coefficient is, the global coordinated boundary variable is, the local decision variable set of the sub-network m is, the boundary variable of the sub-network m is.

[0045] In the second aspect, the embodiment of the present application provides a power distribution line load prediction and optimal scheduling system, comprising:

[0046] The acquisition and preprocessing module is used for acquiring and storing power distribution network operation related data, and performing data quality processing to generate a time-aligned multi-source fusion input data set;

[0047] The load prediction module is used for taking the multi-source fusion input data set as input, constructing a hybrid time series load prediction model, and introducing a weighted quantile loss function in the model training process to optimize the prediction accuracy of the prediction mean and multiple quantile points;

[0048] The demand generation module is used for constructing a joint probability distribution model of renewable energy output and demand response participation rate based on the uncertainty characteristics of distributed energy monitoring data and meteorological elements in the power distribution network operation related data, using a distribution-based scenario generation method to sample the joint probability distribution to generate a set of multiple renewable energy output and demand response participation rate scenarios in the future time domain;

[0049] The scheduling model establishment module is used for constructing a rolling time domain power distribution network optimal scheduling model, and jointly optimizing the power distribution network operation performance in the form of a weighted multi-objective function in the prediction time range of the rolling time domain.

[0050] A strategy generation module is configured to handle uncertainty by using a two-layer hybrid strategy, decompose the optimization problem into multiple sub-problems, and solve the sub-problems in a distributed manner using an alternating direction multiplier method with an adaptive penalty parameter.

[0051] In a third aspect, an embodiment of the present application provides a computer device, comprising a memory and a processor, and the memory stores a computer program, wherein the computer program instructions are executed by the processor to implement the steps of the power distribution line load forecasting and optimal scheduling method according to the first aspect of the present application.

[0052] In a fourth aspect, an embodiment of the present application provides a computer readable storage medium, which stores a computer program, wherein the computer program instructions are executed by the processor to implement the steps of the power distribution line load forecasting and optimal scheduling method according to the first aspect of the present application.

[0053] The present application has the following beneficial effects: the present application establishes a multi-objective optimal scheduling model in a rolling time domain, simultaneously considers active power loss, node voltage deviation, operation cost and risk penalty, and realizes dynamic closed-loop optimization of forecasting and scheduling; and by introducing a two-layer hybrid solving strategy and an ADMM distributed algorithm with an adaptive penalty parameter, the global consistency is ensured while the calculation efficiency and scalability are significantly improved, under the conditions of high proportion of renewable energy access and load fluctuation, high-precision forecasting, robust scheduling and intelligent collaborative control of the distribution network are realized, and the safety, economy and reliability of system operation are improved. BRIEF DESCRIPTION OF DRAWINGS

[0054] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed in the embodiment description. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0055] Figure 1 It is a flowchart of the power distribution line load forecasting and optimal scheduling method;

[0056] Figure 2 It is a structural schematic diagram of the power distribution line load forecasting and optimal scheduling system;

[0057] Figure 3 It is a medium structural schematic diagram of the power distribution line load forecasting and optimal scheduling method and system;

[0058] Figure 4 It is a computing device diagram of the power distribution line load forecasting and optimal scheduling method and system.

[0059] In the figure: 30, optical disc; 40, computing device; 401, processing unit; 402, system memory; 403, bus; 404, external device; 405, I / O interface; 406, network adapter; 4021, RAM; 4022, cache memory; 4023, ROM; 4024, program module; 4025, utility. DETAILED DESCRIPTION

[0060] In order to make the above objectives, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the accompanying drawings.

[0061] In the following description, a lot of specific details are set forth in order to give a thorough understanding of the present application, but the present application can also be implemented in other ways different from the description, and those skilled in the art can make similar generalizations without departing from the spirit of the present application, so the present application is not limited to the specific embodiments disclosed below.

[0062] Secondly, the "one embodiment" or "embodiment" referred to herein means that the specific features, structures or characteristics can be included in at least one implementation of the present application. "In one embodiment" appearing in different places in the specification does not mean the same embodiment, nor is it an independent or alternative embodiment that excludes other embodiments.

[0063] Embodiment 1

[0064] Reference Figures 1-4 For the first embodiment of the present application, the embodiment provides a power distribution line load prediction and optimal scheduling method, comprising:

[0065] S1: Collect and store power distribution network operation related data, and perform data quality processing to generate time-aligned multi-source fusion input data set.

[0066] Further, time series interpolation or autoregressive interpolation method based on historical pattern is used to restore data missing caused by sensor interruption; sliding window statistical method and Z-score detection algorithm are used to identify mutation points or out-of-limit values, and abnormal data affected by fault or communication noise are removed or corrected; according to the master station clock and the time stamp of each terminal device, the data of different sampling periods (1 min, 5 min, 15 min) are interpolated and resampled and time-aligned; all kinds of voltage, current, power, temperature, wind speed and other data are standardized and converted according to the unified engineering unit.

[0067] Further, after completing data cleaning, the data is fused and features are constructed based on time, space and statistical characteristics: historical mean, volatility, intraday periodicity, weekly periodicity and other indicators are extracted from the load curve; weather elements and calendar characteristics are encoded as external input variables; the operating state and power distribution of feeder, transformer and distributed energy nodes are mapped to a unified node attribute vector through a topological structure matrix; the minimum-maximum normalization or Z-score standardization method is used to adjust the characteristic values to a unified dimension interval to prevent model training deviation.

[0068] S2: A hybrid time series load forecasting model is constructed with a multi-source fusion input data set as input. In the model training process, a weighted quantile loss function is introduced to optimize the prediction accuracy of the prediction mean and multiple quantile points.

[0069] Further, the hybrid time series load forecasting model includes: a spatial encoder based on a graph neural network, which is used for spatial feature extraction of the topological structure and electrical correlation between nodes of the distribution line; a time series decoder based on an improved self-attention mechanism, which is used for time series dependence modeling of load change characteristics at different time scales; an uncertainty estimation module based on Monte Carlo Dropout and deep belief network, which is used for probability distribution modeling of the prediction result, and outputs the load probability distribution P(L t ) of each prediction time t.

[0070] Further, the weighted quantile loss function jointly optimizes the prediction results for different quantile points q∈ ={0.1,0.5,0.9}, and its form is:

[0071] ;

[0072] Wherein, is the weighted quantile loss function, y is the true load value, is the predicted value corresponding to the quantile point q, is the weighted coefficient of each quantile point q, ( ) is the quantile loss function.

[0073] The definition of the quantile loss function is:

[0074] ;

[0075] Wherein, is the prediction error;

[0076] In the model training process, a dynamic weighting mechanism is used to dynamically adjust the prediction mean and interval uncertainty jointly optimized according to the quantile error on the validation set.

[0077] Further, the spatial encoder performs neighborhood aggregation operation on each node through multi-layer graph convolution operation to extract spatial correlation features:

[0078] ;

[0079] wherein, is the feature vector of node i at the l+1 layer, is the adjacent weight, σ( ) is a nonlinear activation function, and N(i) is the neighbor set of node i, is the feature vector of node at the l layer, is a bias term;

[0080] Through multi-layer iteration, the node-level spatial embedding representation H spatial is obtained, which is used to reflect the structural characteristics and power distribution patterns of the power distribution network.

[0081] The time decoder takes the node-level spatial embedding representation H spatial and the load sequence in the historical time window as input, and uses an improved self-attention mechanism to model the time dependence; in the traditional attention mechanism, long-term dependence is easily overwhelmed by local features. Therefore, the attention weight is subjected to multi-scale normalization processing, and a time decay factor γ is introduced, so that the model is more sensitive to changes in recent time and pays less attention to trends in distant time:

[0082] ;

[0083] wherein, is the attention weight of time k to time t, , K are the query matrix and the key matrix respectively, is the query vector of time t, is the key vector of time k, is the key vector of the nth sequence index, and γ is the time decay coefficient, is the dimension of the key vector k, and n is all possible historical time indexes;

[0084] After multi-head attention and feedforward network layer calculation, the time sequence feature representation H temporal is obtained, which can accurately capture the load change law at different time scales, including periodic trends and sudden disturbances.

[0085] At the output end, the sample distribution is further density estimated to obtain the load probability distribution P(L t ) of each prediction time t, and the confidence interval is calculated.

[0086] Further, to strengthen the adaptability of the model to different load scenarios, a dynamic weighting mechanism is introduced:

[0087] Initial stage: let w 0.1 = 0.3, w 0.5 = 0.4, w 0.9 = 0.3; at the end of each training round, calculate the average absolute percentage error of each quantile on the validation set; if the error of the high quantile is large, automatically increase w 0.9 .

[0088] If the error of the low quantile is large, increase w 0.1 to enhance the robustness of the low load interval.

[0089] S3: Based on the uncertainty characteristics of the distributed energy monitoring data and meteorological elements in the power distribution network operation related data, a joint probability distribution model of renewable energy output and demand response participation rate is constructed, and a distribution-based scenario generation method is used to sample the joint probability distribution to generate a set of renewable energy output and demand response participation rate scenarios in the future time domain.

[0090] Further, edge probability models are established for photovoltaic output, wind power and demand response rate respectively; Gaussian mixture distribution is used to fit photovoltaic and wind power output to capture their asymmetry under extreme weather conditions; demand response participation rate is modeled using truncated normal distribution or lognormal distribution according to historical response probability characteristics;

[0091] To represent the dependence relationship between different variables, a joint distribution based on Copula function is constructed:

[0092] ;

[0093] Where C( ) is the correlation function, is the joint cumulative distribution function, is the edge cumulative distribution function of photovoltaic output, is the edge cumulative distribution function of wind power, is the edge cumulative distribution function of demand response participation rate, , and are random variables.

[0094] Further, the Copula function model is used to model the nonlinear correlation between meteorological variables and renewable output, or the Gaussian mixture model is used to fit the output distribution characteristics under different operating conditions;

[0095] In each generated scenario, the occurrence probability of each generated scenario is determined, and the generated multiple sets of renewable power output and demand response participation rate scenario set and the occurrence probability of each scenario thereof are output together as inputs of a subsequent power distribution network optimization scheduling model, so as to represent the uncertainty of renewable energy output and demand response.

[0096] It should be noted that each power generation output is normalized to [0, 1] according to the rated capacity, and the demand response rate R t is maintained as a proportion (0-1); if there are a large number of 0s or 1s, mark as a special sample in advance. Detrend and deseasonalize the time series: if the intra-day and seasonality are very significant, first model by time period (hour) or use regression to remove the seasonal term, and then build the marginal distribution; truncate or separately model the sparse extreme values. Divide the training, validation and test sets, and remove or correct the outliers.

[0097] If the demand response rate R t is mostly continuous distribution in (0, 1), and there are no large number of 0s or 1s, use Beta distribution, or lognormal and truncated normal; if there are a large number of 0s (no response) or 1s (full response), use a zero inflation model: consider the demand response rate R t as a mixture of point mass combined with continuous distribution, and the rest follow the truncated normal. Steps: estimate whether there is zero and one inflation; if it is a simple continuous distribution, use MLE to fit the parameters of truncated normal or lognormal, and save the CDF and inverse CDF. If it is a mixture model, use EM or directly optimize the mixed log-likelihood to estimate the parameters.

[0098] Further, use the data of the same period or similar weather conditions as the sample. Use BIC and AIC or cross-validation to select the number of components K, use EM algorithm to fit GMM, and get the component weight, mean and variance. If multiple grid-connected points are jointly modeled, GMM can be fitted for each point separately, and then the dependence between different points is coupled by Copula; or GMM is directly fitted for the aggregated output. First, randomly select a component according to the component weight, and then sample the output value from the Gaussian component; if there are physical boundaries (such as non-negative, upper limit), truncate or resample the sampling results.

[0099] Set to generate N sample trajectories (for example, N = 5000-20000), and each trajectory is a sequence of PV, WIND and DR values for the entire prediction window. If the independent sampling method is used, the time correlation can be restored by residual autoregression or bootstrap. For each trajectory, a representative feature vector is extracted, such as peak value, peak time position, mean, standard deviation, energy, maximum drop rate, correlation index, etc.; or the entire trajectory vector is directly used. The cluster center or cluster barycenter of each cluster c is selected as the multiple sets of renewable power output and demand response participation rate scenario set, and the occurrence probability of each generated scenario is calculated according to the proportion of the number of samples in the cluster.

[0100] S4: Construct the optimization scheduling model of distribution network in rolling time domain, and jointly optimize the operation performance of distribution network in the prediction time range of rolling time domain in the form of weighted multi-objective function.

[0101] Further, the scheduling time is divided into continuous rolling periods, each rolling period corresponds to a prediction window, and in each period, optimization calculation is performed based on the latest observation data and prediction results to execute the first-step decision at the current time;

[0102] After each execution, the rolling window is moved forward, the data is reacquired, the prediction and scenario are updated, and the optimization is repeated to realize closed-loop control; the probabilistic load prediction is combined with renewable energy output and demand response scenarios to form multiple operation scenarios.

[0103] It should be noted that the scheduling period of a day is divided into several continuous rolling periods, for example, one rolling period every 15 minutes. For each rolling period k, the prediction time range of the corresponding rolling time domain is defined, which can be 1-4 hours; the rolling step is determined to ensure that the output of each optimization only executes the first-step decision. At the beginning of the rolling period k, the latest observation data is collected, including real-time load of the distribution network, renewable energy output, weather information, demand response participation rate and available capacity, switch state and topology and other device parameters; the data is preprocessed: missing value interpolation, outlier removal, normalization, and a time-aligned multi-source input data set is generated; the time-aligned multi-source input data set is input into the hybrid time series prediction model, and the load probability distribution P(L t ), the GNN spatial encoder extracts the line topology and electrical correlation between nodes; the improved self-attention decoder models the time series dependence; Monte Carlo Dropout and deep belief network estimate uncertainty; the weighted quantile loss function is used to optimize the prediction mean and quantile point, and the high peak and low valley load prediction accuracy is strengthened; the output load prediction is a probability distribution at each time, which can be used for risk assessment in subsequent scheduling.

[0104] S5: A two-layer hybrid strategy is used to handle uncertainty, and the optimization problem is decomposed into multiple sub-problems and solved distributedly using an alternating direction multiplier method with an adaptive penalty parameter.

[0105] Further, risk budget constraints are introduced for key constraints, converting the original probability constraints into solvable deterministic constraints; for the short-time fine-tuning stage in the rolling time domain, model predictive control is adopted, and each optimization starts from the current observation, combined with the weighted period of multiple scenarios to dynamically adjust the output of mechanical and electrical resources, energy storage charging and discharging, and demand response;

[0106] The decomposed subproblems are solved iteratively using ADMM, with each subnet independently calculating its local optimum to ensure constraint satisfaction. Convergence and consistency correction are performed during iterations using boundary variables, and Lagrange multipliers and penalty terms are used to guide the solutions of each subnet towards global consistency. ADMM incorporates constraints into Lagrangian form and adds a quadratic penalty term to form an enhanced Lagrangian function.

[0107] ;

[0108] in, To enhance the Lagrange function, Let m be the local objective function of subnet m. For Lagrange multipliers, This is the coefficient for secondary penalties. For globally coordinated boundary variables, Let m be the set of local decision variables for subnet m. Let m be the boundary variable of subnet m.

[0109] It should be noted that the boundary variables of all subnets The boundary variable z is aggregated to form a globally coordinated boundary variable, and consistency correction is performed in each iteration. A weighted aggregation strategy is adopted, which dynamically adjusts the influence weight of each subnetwork on the global boundary variable based on the importance of the subnetwork or the sensitivity of key nodes in the power grid, improving the accuracy of global scheduling. Each subnetwork independently solves the enhanced Lagrangian function, achieving parallel computation and significantly reducing the overall solution time. Combining dynamic aggregation of boundary variables with Lagrange multiplier adjustment enables distributed iterative solution while ensuring global constraint satisfaction, suitable for rolling scheduling of large-scale distribution networks. Using probabilistic load forecasting, renewable energy output, and demand response scenarios as inputs, and combining the above-mentioned distributed ADMM solution, each round of rolling time-domain optimization finds local optima in multiple scenarios, while ensuring global consistency through boundary variables and the Lagrange multiplier mechanism. Integrating distributed ADMM with probabilistic rolling optimization improves the robustness and economy of the distribution network under load fluctuations, DER output uncertainty, and demand response participation.

[0110] Example 2

[0111] This embodiment is the second embodiment of the present invention, which provides a method for predicting and optimizing the scheduling of power distribution line loads. In order to verify the beneficial effects of the present invention, a simulation experiment is conducted for scientific demonstration.

[0112] Set up a medium-voltage distribution network, including 3 substations, 12 feeders, and a total of 120 nodes, equipped with smart meters, distributed energy monitoring devices, meteorological monitoring equipment, and historical archives.

[0113] Training strategy: Weighted quantile loss function, quantile set ={0.1, 0.5, 0.9}, high quantile points are weighted to improve the peak load prediction accuracy; low quantile points are weighted to enhance the low valley load prediction robustness, and a load probability distribution of 24 hours in the future is output.

[0114] Sampling generates a set of 50 groups of renewable power and demand response participation rate scenarios for 24 hours in the future, and the occurrence probability corresponding to each generated scenario;

[0115] Table 1: Renewable energy and demand response scenario data

[0116]

[0117] Each hour is a rolling period, and the first step optimization decision is executed every minute, the scenario tree + risk budget is used to convert the probability constraint into a deterministic constraint; based on the model predictive control idea, the current observation is taken as the starting point, and the multi-scene weighting or worst scenario is combined for solving; the simulation condition is 24 hours in time range and 1 minute in time step,

[0118] The load prediction and optimization scheduling is updated in real time, and the sub-network distributed calculation is parallelly solved.

[0119] Table 2: Simulation result comparison

[0120]

[0121] Table 2 shows that the active loss of the distribution line is reduced by 12%, the probability load prediction and multi-scene power flow optimization of the method of the application make the node power distribution more balanced, reduce the feeder line current at the peak load, and reduce the line loss; the number of node voltage overruns is reduced by 80%, in the traditional method, the random fluctuation of photovoltaic output easily causes voltage fluctuation, the confidence interval scheduling and risk budget constraint are introduced in the application, the voltage overrun is effectively inhibited, and the voltage stability and operation safety of the distribution network are significantly improved; the DER scheduling cost is reduced by 13.7%, the application jointly optimizes the DER output and the demand response participation rate, so that the energy storage device is charged at low price and discharged at high price, and the overall scheduling is more economical; the demand response penalty cost is reduced by 40%, the traditional deterministic strategy leads to user response overrun or deficiency, the application adopts scene generation and response probability distribution fitting, so that the scheduling signal is more in line with the actual response ability; the solving time is reduced by 64%, the application adopts ADMM distributed solving, divides the global problem into 6 sub-networks for parallel calculation, adjusts the convergence speed through an adaptive penalty term, and reduces the average solving time of each step from 2.5 seconds to 0.9 seconds.

[0122] After introducing the method of the exemplary embodiment of the application, next, with reference to Figure 2 An exemplary embodiment of the application is described below with reference to a power distribution line load prediction and optimization scheduling system.

[0123] The collection preprocessing module is configured to collect and store power distribution network operation related data, and perform data quality processing to generate a multi-source fusion input data set aligned in time;

[0124] The load prediction module is configured to use the multi-source fusion input data set as input to construct a hybrid time series load prediction model, and introduce a weighted quantile loss function in the model training process to optimize the prediction accuracy of the prediction mean and multiple quantile points.

[0125] The demand generation module is configured to construct a joint probability distribution model of renewable energy output and demand response participation rate based on the uncertainty characteristics of distributed energy monitoring data and meteorological elements in the power distribution network operation related data, and use a distribution-based scenario generation method to sample the joint probability distribution to generate a set of multiple renewable energy output and demand response participation rate scenarios in the future time domain.

[0126] The dispatch model establishment module is configured to construct a rolling time domain power distribution network optimization dispatch model to jointly optimize the power distribution network operation performance in the form of a weighted multi-objective function within the prediction time range of the rolling time domain.

[0127] The strategy generation module is configured to use a two-layer hybrid strategy to handle uncertainty, decompose the optimization problem into multiple sub-problems, and use an alternating direction multiplier method with an adaptive penalty parameter to solve the sub-problems in a distributed manner.

[0128] After introducing the method and system of the exemplary embodiments of the present application, next, with reference to Figure 3 The computer readable storage medium of the exemplary embodiments of the present application is described with reference to Figure 3 The computer readable storage medium shown is an optical disc 30, which stores a computer program (i.e. a program product) thereon, the computer program when executed by a processor will implement each step described in the above method embodiments, and the specific implementation of each step will not be repeated here.

[0129] It should be noted that examples of the computer readable storage medium can also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical, magnetic storage medium, which will not be repeated here.

[0130] After introducing the method and medium of the exemplary embodiments of the present application, next, with reference to Figure 4 The computing device for power distribution line load prediction and optimization dispatch of the exemplary embodiments of the present application.

[0131] Figure 4A block diagram is shown of an exemplary computing device 40 suitable for implementing embodiments of the present invention. The computing device 40 may be a computer system or a server. Figure 4 The computing device 40 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0132] like Figure 4 As shown, the components of computing device 40 may include, but are not limited to: one or more processors or processing units 401, system memory 402, and bus 403 connecting different system components (including system memory 402 and processing unit 401).

[0133] The computing device 40 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computing device 40, including volatile and non-volatile media, and removable and non-removable media.

[0134] System memory 402 may include computer system readable media in the form of volatile memory, such as at least one of random access memory (RAM) 4021 or cache memory 4022. Computing device 40 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, ROM 4023 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 4 (Not shown in the image, usually referred to as "hard drive"). Although not shown in... Figure 4 The diagram illustrates that disk drives for reading and writing to removable non-volatile disks (e.g., "floppy disks") and optical disc drives for reading and writing to removable non-volatile optical discs (e.g., CD-ROMs, DVD-ROMs, or other optical media) can be provided. In these cases, each drive can be connected to bus 403 via one or more data media interfaces. System memory 402 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0135] A program / utility 4025 having a set (at least one) of program modules 4024 may be stored, for example, in system memory 402, and such program modules 4024 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data, each or some combination of these examples may include an implementation of a network environment. Program modules 4024 typically perform the functions and / or methods described in the embodiments of the present invention.

[0136] Computing device 40 can also communicate with one or more external devices 404 such as a keyboard, a pointing device, a display, etc. through I / O interface 405. Further, computing device 40 can communicate with one or more networks such as a local area network (LAN), a wide area network (WAN) and / or a public network such as the Internet, through network adapter 406. As Figure 4 illustrated, network adapter 406 communicates with the other modules of computing device 40 such as processing unit 401, etc. through bus 403. It should be appreciated that although not shown, other hardware and / or software modules could be used in conjunction with computing device 40. Examples, include, but are not limited to, microcode, device drivers, redundant processing units, and external disk drive arrays, to name a few. Figure 4 Figure 4

[0137] Processing unit 401 executes various functions applications and data processing by running programs stored in system memory 402.

[0138] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0139] In several embodiments provided by the present application, it should be understood that the disclosed system, device and method can be implemented by other ways. The device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and there can be another division manner in actual implementation, and for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some communication interface, device or unit, and can be electrical, mechanical or other forms.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application rather than limit the present application. Although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and all modifications or replacements should be covered in the scope of the claims of the present application.​

Claims

1. A power distribution line load forecasting and optimal dispatching method, characterized in that, The method comprises the following steps: Collect and store power distribution network operation related data, and perform data quality processing to generate a time-aligned multi-source fusion input data set; A hybrid time series load prediction model is constructed by taking the multi-source fusion input data set as input. In the model training process, a weighted quantile loss function is introduced to optimize the prediction accuracy of the prediction mean and multiple quantile points. The hybrid time series load prediction model comprises: The hybrid time series load prediction model comprises a spatial encoder based on a graph neural network, which is used for spatial feature extraction of the topological structure of the power distribution line and the electrical correlation between nodes; a time series decoder based on an improved self-attention mechanism, which is used for time series dependence modeling of load change characteristics at different time scales; and an uncertainty estimation module based on Monte Carlo Dropout and deep belief network, which is used for probability distribution modeling of the prediction results and outputs the load probability distribution P(Lt) at each prediction time t. Based on the uncertainty characteristics of the distributed energy monitoring data and meteorological elements in the power distribution network operation related data, a joint probability distribution model of renewable energy output and demand response participation rate is constructed, and a distribution-based scenario generation method is used to sample the joint probability distribution to generate a set of renewable energy output and demand response participation rate scenarios in the future time domain. The construction of the joint probability distribution model of renewable energy output and demand response participation rate comprises: Edge probability models are respectively established for photovoltaic output, wind power and demand response rate. Gaussian mixture distribution is used to fit photovoltaic and wind power outputs to capture their asymmetry under extreme weather conditions. The demand response participation rate is modeled using truncated normal distribution or lognormal distribution according to the historical response probability characteristics. To represent the dependence between different variables, a joint distribution based on Copula function is constructed. A rolling time domain power distribution network optimization scheduling model is constructed to jointly optimize the performance of the power distribution network in the form of a weighted multi-objective function within the prediction time range of the rolling time domain. The construction of the rolling time domain power distribution network optimization scheduling model to jointly optimize the performance of the power distribution network in the form of a weighted multi-objective function within the prediction time range of the rolling time domain comprises: The scheduling time is divided into consecutive rolling periods, each corresponding to a prediction window. In each rolling period, optimization calculations are performed based on the latest observation data and prediction results to execute the first-step decision at the current time. After each execution, the rolling window is moved forward, the data is re-collected, the prediction and scenarios are updated, and the optimization execution is repeated to realize closed-loop control. The probabilistic load prediction is combined with renewable energy output and demand response scenarios to form multiple operation scenarios. A two-layer hybrid strategy is used to handle uncertainty, which divides the optimization problem into multiple sub-problems and solves them using an alternating direction multiplier method with an adaptive penalty parameter. The two-layer hybrid strategy for handling uncertainty, which divides the optimization problem into multiple sub-problems and solves them using an alternating direction multiplier method with an adaptive penalty parameter, comprises: A risk budget constraint is introduced for key constraints to convert the original probability constraint into a solvable deterministic constraint. For the short-time fine-tuning stage in the rolling time domain, the model predictive control idea is adopted, each step is optimized from the current observation, and the output of the electro-mechanical resource, the charging and discharging of the energy storage and the demand response are dynamically adjusted in combination with multiple scenarios.

2. The power distribution line load forecasting and optimization scheduling method of claim 1, wherein: The introduction of the weighted quantile loss function optimizes the prediction accuracy of the prediction mean and multiple quantile points, including: The weighted quantile loss function is applied to different quantiles. The prediction results are jointly optimized, and the form is as follows: ； wherein, is a weighted quantile loss function, is a true load value, is a quantile point corresponding predicted value, is a weighting coefficient for each quantile point is a quantile loss function;​ The definition of "C1-C6alkyl" is: ； wherein is the prediction error; In the process of model training, a dynamic weighting mechanism is adopted to dynamically adjust the quantile error on the validation set The prediction mean and interval uncertainty are jointly optimized.

3. The power distribution line load forecasting and optimization scheduling method of claim 1, wherein: The expression of the joint distribution is as follows: ； wherein, is the correlation function, is the joint cumulative distribution function, is the marginal cumulative distribution function of the photovoltaic output, is the marginal cumulative distribution function of the wind power, is the marginal cumulative distribution function of the demand response participation rate, and are random variables.

4. The power distribution line load forecasting and optimization scheduling method of claim 1, wherein, The distribution-based scenario generation method is adopted to sample the joint probability distribution, including: The Copula function model is used to model the nonlinear correlation between meteorological variables and renewable output, or the Gaussian mixture model is used to fit and sample the output distribution characteristics under different operating states; In each generated scenario, the occurrence probability corresponding to each generated scenario is determined, and the generated multiple sets of renewable output and demand response participation rate scenario set and the occurrence probability corresponding to each generated scenario are output together as the input of the subsequent optimization scheduling model to represent the uncertainty of renewable energy output and demand response.

5. The power distribution line load forecasting and optimization scheduling method of claim 1, wherein: The decomposed sub-problems are solved by ADMM iteration, each sub-network independently calculates the local optimal solution to ensure that the constraints are met; the boundary variables are converged and consistency corrected in the iteration, the Lagrange multiplier and the penalty term are used to guide the convergence of each sub-network solution to global consistency, ADMM introduces constraints into the Lagrange form and adds a quadratic penalty term to form an enhanced Lagrange function: ； wherein, is a local objective function of the subnetwork is a local objective function of the subnetwork is a local objective function of the subnetwork is a Lagrange multiplier, is a quadratic penalty coefficient, is a globally coordinated boundary variable, is a local decision variable set of the subnetwork is a local decision variable set of the subnetwork is a boundary variable of the subnetwork is a boundary variable of the subnetwork 6. The system of power distribution line load forecasting and optimal scheduling method based on claim 1, characterized in that, Including: The acquisition and preprocessing module is used to acquire and store power distribution network operation related data, and perform data quality processing to generate a time-aligned multi-source fusion input data set; The load prediction module is used to construct a hybrid time series load prediction model with the multi-source fusion input data set as input, and in the model training process, a weighted quantile loss function is introduced to optimize the prediction accuracy of the prediction mean and multiple quantile points. The hybrid time series load prediction model includes: the hybrid time series load prediction model includes a spatial encoder based on a graph neural network, which is used for spatial feature extraction of the topological structure and electrical correlation between nodes of the power distribution line; a time series decoder based on an improved self-attention mechanism, which is used for time series dependent modeling of load change characteristics under different time scales; an uncertainty estimation module based on Monte Carlo Dropout and deep belief network, which is used for probability distribution modeling of the prediction result, and outputs the load probability distribution P(Lt) of each prediction time t. The demand generation module is configured to construct a joint probability distribution model of renewable energy output and demand response participation rate based on the uncertainty characteristics of distributed energy monitoring data and meteorological elements in power distribution network operation related data, generate a plurality of sets of renewable energy output and demand response participation rate scene sets in a future time domain by sampling the joint probability distribution using a distribution-based scene generation method, and construct an edge probability model for photovoltaic output, wind power and demand response rate respectively. The scheduling model establishment module is configured to construct a rolling time domain power distribution network optimization scheduling model, and jointly optimize power distribution network operation performance in a weighted multi-objective function form within a predicted time range of the rolling time domain. The strategy generation module is configured to use a two-layer hybrid strategy to handle uncertainty, decompose the optimization problem into a plurality of sub-problems, and solve the sub-problems using an alternating direction multiplier method with an adaptive penalty parameter. 7.A computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the computer device is characterized in that: The processor executes the computer program to implement the steps of the power distribution line load prediction and optimization scheduling method of any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that: The computer program is executed by the processor to implement the steps of the power distribution line load prediction and optimization scheduling method of any one of claims 1-5.

Images