Power grid dispatching scheme generation method and system based on load optimization

Abstract

The invention discloses a power grid dispatching scheme generation method and system based on load optimization, and the method comprises the steps: predicting a reference load change curve and a confidence interval of each node in a future set time period according to the load data and meteorological data of each node in a power grid topological graph in a corresponding time period; continuously determining a risk area based on the risk assessment model and determining a predicted load offset and a standard deviation so as to correspondingly optimize and broaden a reference load change curve and a confidence interval of each node in the risk area; and constructing an uncertain scene set based on the reference load change curve and the confidence interval, then constructing an objective function, and solving the objective function based on the uncertain scene set to generate an elastic scheduling scheme with the minimum power generation cost and the minimum load vacancy in the worst load scene. According to the invention, through supplementing the prediction load offset and the standard deviation, the prediction precision of the short-term load change under the condition of sudden power consumption peak or extreme weather is obviously enhanced, and the power grid dispatching level is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of power grid dispatching, in particular to a power grid dispatching scheme generation method and system based on load optimization. BACKGROUND

[0002] With the integration of new energy into the power grid and the synchronization of power supply, as well as the increasing diversification and dynamic characteristics of industrial, residential and other power loads, the complexity of power grid operation has significantly increased. As a key link for ensuring the safe and stable operation of the power system and achieving efficient allocation of power resources, the prediction accuracy of user power load directly affects the power supply reliability, energy utilization efficiency and operation economy of the power grid.

[0003] In existing power grid load dispatching, the following problems exist: When dealing with new energy output fluctuations and multi-type load collaborative response, it is difficult to achieve real-time dynamic and accurate matching, resulting in low rationality of power resource allocation in some time periods; The prediction accuracy of traditional dispatching schemes for short-term load changes is low, and it is difficult to fully predict sudden power consumption peaks or load fluctuations under extreme weather conditions; In cross-regional and multi-level load coordination and dispatching, the information interaction efficiency and decision response speed are slow, and the overall dispatching capability of the power grid cannot be effectively realized; To this end, a power grid dispatching scheme generation method based on load optimization is needed. SUMMARY

[0004] To solve the above technical problems, a power grid dispatching scheme generation method and system based on load optimization are provided, which can accurately predict load changes during sudden power consumption peaks or extreme weather conditions, generate a dispatching method that can fully call a large number of new energy power sources based on the predicted load changes, and realize the coordinated dispatching of cross-regional and multi-level loads.

[0005] To achieve the above purpose, the technical solution adopted by the present application is: A power grid dispatching scheme generation method based on load optimization, comprising: Load prediction: Construct a power grid topology graph, and a load prediction model predicts the baseline load change curve and confidence interval of each node in the future time period according to the load data and meteorological data of each node in the power grid topology graph in the time period; Load optimization: the power grid coverage area is divided into several load areas, the risk level of each load area is continuously obtained based on the risk assessment model, and the load area exceeding the preset level is defined as a risk area, the bias prediction model predicts the predicted load deviation and standard deviation caused by the sudden characteristics based on the sudden characteristics data of the risk area, the reference load change curve of each node in the risk area is added to the predicted load deviation to generate a corrected reference load change curve, and the standard deviation is used to widen the confidence interval; Generating a scheduling scheme: based on the corrected reference load change curve and the widened confidence interval, an uncertain scenario set is constructed, a target function including generation cost and load shortage is constructed, and the scenario in the uncertain scenario set that maximizes the solution is defined as the worst load scenario by solving the target function, and the target function is continuously solved under the worst load scenario to generate an elastic scheduling scheme with minimum generation cost and load shortage.

[0006] Preferably, the risk assessment model comprises: continuously obtaining meteorological data, cloud data and Internet of Things data, constructing a mapping relationship between meteorological data and load areas, determining the moving track of the cloud layer based on the cloud data, determining the load weight of each load area based on the Internet of Things data, and predicting the risk level of each load area based on the moving track of the cloud layer, meteorological data in each load area and load weight.

[0007] Preferably, the Internet of Things data comprises total power consumption data of each load area and power meter data of power meter power consumption in each load area; The load weight generation method is: obtaining people flow data and corresponding total power consumption data of different load areas at different time types, and establishing a people flow-load conversion coefficient table at different time types; Clustering the power meter data of each load area to determine the power consumption mode, obtaining the building density and per-house capacity of each load area to calculate the corresponding load density base value; Fusing the people flow-load conversion coefficient table of each load area, the load density base value and the power consumption mode, and outputting the load weight of each load area.

[0008] Preferably, the sudden characteristic data comprises temperature change gradient, wind speed mutation flag, cloud amount cumulative value, and Internet of Things device offline rate, and the bias prediction model comprises an encoder for converting the sudden characteristic data into a latent space vector, and a Bayesian linear regression calculation layer for calculating the predicted load deviation and standard deviation based on the latent space vector.

[0009] Preferably, solving the target function comprises: constructing a main problem and fixing an initial scenario to solve a corresponding elastic scheduling scheme, then constructing a sub-problem to find the worst scenario that maximizes the target function in the uncertain scenario set, adding the worst scenario to the main problem and repeatedly solving the elastic scheduling scheme, and stopping iteration when the convergence condition is met, to output the elastic scheduling scheme.

[0010] Preferably, the elastic scheduling scheme comprises a scheduling instruction framework and an elastic adjustment rule, the scheduling instruction framework comprises output plans of each energy unit to each node and flow limit values of transmission lines, and the elastic adjustment rule comprises a switching sequence of corresponding self-starting backup power and a calling priority of demand-side response resources under different deviation levels of predicted load and actual load of the node.

[0011] Preferably, the power grid scheduling scheme generation method comprises scheme evaluation: a digital twin model of the power grid simulates operation of a plurality of elastic scheduling schemes under different disturbance levels, and a scheme evaluation strategy determines an optimal elastic scheduling scheme based on simulation results of a plurality of indicators corresponding to each elastic scheduling scheme under different disturbance levels.

[0012] Preferably, the disturbance comprises a load prediction error disturbance, a unit fault disturbance, and a new energy output fluctuation disturbance.

[0013] Preferably, the scheme evaluation strategy comprises: obtaining simulation results of each indicator of the elastic scheduling scheme under a certain disturbance level, determining corresponding indicator levels based on the simulation results of each indicator, and performing weighted summation, and a summation result is an evaluation score of the elastic scheduling scheme under the disturbance level. The evaluation scores of the plurality of elastic scheduling schemes under each disturbance level are calculated respectively, the optimal elastic scheduling scheme is an elastic scheduling scheme whose evaluation scores under each disturbance level all meet a threshold value and whose total score is the highest.

[0014] A power grid scheduling scheme generation system based on load optimization comprises a dynamic risk profiling module for running a risk assessment model, a hybrid prediction engine module for performing load prediction and load optimization steps, a multi-objective robust optimization decision module for performing a scheduling scheme generation step, and a digital twin closed-loop correction module for correcting and screening elastic scheduling schemes.

[0015] Compared with the prior art, the power grid scheduling scheme generation system based on load optimization has the following beneficial effects: The power grid coverage area is divided into a plurality of load areas, the load weight of each load area is continuously updated through Internet of Things data, the risk level of each load area is continuously determined in combination with meteorological data and cloud image data, and whether a sudden disturbance (a sudden peak of electricity use or a sudden extreme weather) occurs in a corresponding load area is determined based on the monitored risk level, so that whether a sudden disturbance occurs in the power grid coverage area and an influence range of the disturbance can be quickly determined.

[0016] By setting sudden characteristic data of the risk area and a deviation prediction model, the deviation prediction model only comprises an encoder and a Bayesian linear regression calculation layer, the prediction load deviation and the standard deviation of the risk area can be quickly and accurately calculated using simple calculation, and the calculation efficiency is improved.

[0017] By predicting the load offset, optimizing the baseline load change curve and widening the confidence interval, the baseline load change curve in the interference area can be accurately optimized, and the confidence interval in the interference area can be widened, so as to realize accurate and comprehensive prediction of the power load change of each node in the power grid.

[0018] By constructing a target function containing power generation cost and load shortage, the elastic scheduling scheme with minimum power generation cost and minimum load shortage in the worst load scenario is solved based on the target function, so that the economic efficiency and the robustness of the scheduling scheme are considered, and the generated elastic scheduling strategy helps to improve the supply and demand balance ability of the power grid.

[0019] The elastic scheduling scheme includes a scheduling instruction framework and an elastic adjustment rule. The scheduling instruction framework is used to determine the scheduling process, and is matched with the elastic adjustment rule according to the actual response of the power supply end and the load end. The specific scheduling process is determined based on the matching result. The actual working condition of the power grid is dynamically adjusted to improve the rationality of the scheduling scheme and the utilization rate of power resources.

[0020] The digital twin model of the power grid simulates the feasibility of the elastic scheduling scheme in advance to adjust and optimize the elastic scheduling scheme in advance, and also avoids the problem of slow information interaction efficiency and decision response speed between cross-regions, improves the effectiveness of the elastic scheduling scheme, and realizes the overall scheduling of the power grid.

[0021] The scheme evaluation strategy is set, and the index for judging the effectiveness of the scheme is set. The optimal elastic scheduling scheme is determined based on the simulation results of the index, so as to ensure the reliability and effectiveness of the scheduling scheme in actual application. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A method flowchart of a power grid scheduling scheme generation method based on load optimization; Figure 2 A system framework diagram of a power grid scheduling scheme generation system based on load optimization. DETAILED DESCRIPTION

[0023] The following description is used to disclose the present application so that those skilled in the art can implement the present application. The preferred embodiments in the following description are only examples, and other obvious modifications can be made by those skilled in the art.

[0024] A power grid scheduling scheme generation method based on load optimization, as Figure 1As shown, the method comprises load prediction: constructing a power grid topology map, and a load prediction model predicting a reference load change curve and a confidence interval of each node in a future time period based on load data and meteorological data of each node in the power grid topology map in a time period.

[0025] The construction of the power grid topology map comprises: based on the actual distribution structure of the power grid, setting specific substations and distribution network load points (such as distribution transformers for residential areas, industrial areas or commercial areas) as nodes, transmission lines between adjacent nodes as edges, and setting weights of corresponding edges based on impedances of each transmission line to generate a power grid topology map of the power grid. For example, two nodes include two edges, and the smaller the impedance, the greater the edge weight.

[0026] The preset time period of the load data is a time period within the current several hours, and the preset time period of the meteorological data includes a time period within the current several hours and a time period within the future several hours. An embodiment of the present application is that the current several hours are 6 hours before the current time, and the future several hours are 6 hours after the current time.

[0027] The reference load change curve of a certain node represents the predicted change of the electric energy flowing through the node in the future time period, and the corresponding confidence interval is the prediction error range, which represents the actual electric energy flowing through the node fluctuating within the range formed by the predicted value plus the confidence interval.

[0028] Training the load prediction model comprises: representing historical parameter data contained by each node in the power grid topology map in the form of an adjacency matrix, and the historical parameter data contained by the node includes load data, meteorological data and corresponding actual load change curves. Several historical parameter data contained by all nodes are normalized and standardized to generate a data set, and the data set is proportionally divided into a training set, a validation set and a test set. The load prediction model selects a spatio-temporal graph neural network architecture, trains the spatio-temporal graph neural network architecture based on the training set, the validation set and the test set, simultaneously adopts an optimizer and an early stopping mechanism to prevent model overfitting, and completes the training of the load prediction model.

[0029] Load optimization: the power grid coverage area is divided into several load areas, the risk level of each load area is continuously obtained based on a risk assessment model, and the load area exceeding a preset level is defined as a risk area, the bias prediction model predicts the prediction load offset and standard deviation caused by the sudden characteristics based on the sudden characteristics data of the risk area, the reference load change curve of each node in the risk area is added to the prediction load offset to generate a corrected reference load change curve, and the standard deviation is used to widen the confidence interval.

[0030] Load area division: the user's power consumption type includes residential power consumption, office power consumption, factory power consumption, etc., and different power consumption types correspond to different power consumption, power consumption peak period and power loss. In order to accurately predict the interference degree of the emergency to different areas, the power grid coverage area is divided into several load areas based on the power consumption type. When an emergency occurs, the risk level of the load area is evaluated based on the influence degree of different load areas. An embodiment of the present application is that: when a sudden rainstorm weather occurs, the risk level of the load area corresponding to the office power consumption and the residential power consumption is lower than that of the open-air aquaculture power consumption area. For example: when a sudden rainstorm occurs, the water pump of the open-air fish farm needs to work continuously and cannot be powered off.

[0031] The risk assessment model includes: continuously acquiring meteorological data, cloud data and Internet of Things data, constructing the mapping relationship between meteorological data and load area, determining the moving track of the cloud layer based on the cloud data, determining the load weight of each load area based on the Internet of Things data, and predicting the risk level of each load area based on the meteorological data, the moving track of the cloud layer and the load weight in each load area.

[0032] The meteorological data includes temperature, humidity, wind speed, cloud amount and other parameters. When constructing the mapping relationship, the above parameters are normalized, and then the meteorological data is interpolated into a specific geographic grid for spatio-temporal alignment. The load area and the geographic grid construct the mapping relationship, realizing one-to-one correspondence between the load area and the meteorological data.

[0033] The cloud data includes satellite radar images. The satellite lightning image is identified to determine the cloud coverage rate. The cloud layer area of continuous multiple satellite lightning images is determined by a filtering algorithm to obtain the moving track of the cloud layer. An embodiment of the present application is: a convolutional neural network model is used to identify the satellite lightning image, historical satellite radar images and ground observation cloud coverage rate data are used as a data set, and the convolutional neural network model is trained using the data set.

[0034] The Internet of Things data includes the total power consumption data of each load area. The people flow change data in each load area is acquired, and the time type is set. The people flow change data and the power consumption are divided and counted based on the time type, so as to acquire the people flow and the power consumption of different load areas under different time types, and establish a people flow-load conversion coefficient table. An embodiment of the present application is that: the time type includes weekdays, holidays, daytime, night, late night, etc., and the people flow-load conversion coefficient table records the average people flow and the average power consumption of each load area under different time types.

[0035] An embodiment of the present application is that: the people flow change data is derived from the signaling data of mobile operators. The real-time number of people in each load area is calculated by analyzing the connection relationship between the user's mobile phone and the base station, and a people flow heat map is generated. Based on the people flow heat map, the people flow change data of each load area can be determined.

[0036] The Internet of Things data includes the electric meter data of the power consumption of the electric meter in each load area, the power consumption mode of each load area is determined through clustering analysis of the electric meter data, the smart meter data is collected according to a specific period and the abnormal values are removed, and then the clustering algorithm is used to divide the load area into different power consumption modes.

[0037] One embodiment of the present application is that a certain load area is a residential area, the power consumption data of the electric meter of the residents is obtained, and clustering is performed to determine the power consumption type of the residential area. The power consumption type includes high-load power consumption (such as villa area), medium-load power consumption (such as multi-family member type community), low-load power consumption (such as few family member type community), etc.

[0038] The load density base value corresponding to the power consumption type is calculated based on the building density and the average capacity per household. The building density is obtained through the GIS data provided by the urban planning department, and the average capacity per household is queried through the power marketing system. The load density base value is used to represent the initial reference value of the average power consumption per unit area in the load area. The load density base value is also high in areas with high building density (high floors and dense floor distribution) and high average capacity per household (five rooms per household).

[0039] The gradient boosting tree algorithm is used to fuse the people-flow-load conversion coefficient table of each load area, the load density base value and the power consumption mode, and the load weight of each load area is output. The greater the load weight, the greater the power consumption and the greater the power loss. The weight matrix is updated according to the preset period, and the update trigger conditions include that the people flow density changes by a set amount or the meteorological warning level is improved. When the trigger condition is met, the system immediately starts the weight update process without waiting for the preset period.

[0040] The burst feature data of the risk area includes temperature change gradient, wind speed mutation flag, cloud amount cumulative value, and Internet of Things device offline rate. The bias prediction model includes an encoder for converting the burst feature data into a spatial vector and a Bayesian linear regression calculation based on the spatial vector to calculate the load offset and standard deviation. The encoder encodes the burst feature data of each risk area to generate a feature vector, and then maps the feature vector to a latent space vector; the Bayesian linear regression calculation receives the latent space vector of each load area, and uses the Markov Chain Monte Carlo sampling method to collect a number of candidate vectors close to the latent space vector in the latent space, and calculates the load offset and standard deviation of each risk area based on the prior error offset of the candidate vector.

[0041] One embodiment of the present application is: assuming that a load area (risk area) for residential living has a sudden temperature drop, a number of candidate vectors close to the latent space vector of the load area are extracted, all of which refer to the sudden temperature drop of the residential area, the predicted load offset of the risk area in the future period is calculated based on the candidate predicted load offset corresponding to the number of candidate vectors, and the standard deviation is calculated based on the number of candidate predicted load offsets.

[0042] The reference load change curve of each node in the risk area is added to the predicted load offset to generate an optimized reference load change curve, and the confidence interval width corresponding to the reference load change curve is dynamically adjusted according to the standard deviation, so as to ensure that the confidence interval covers more comprehensive possible load values in a high-risk scenario. Realize accurate prediction of load change of each node in the power grid when a local load area appears a sudden situation, improve the rational allocation of subsequent power resource scheduling, and facilitate efficient use of electric energy to realize reliable power distribution.

[0043] Generate a scheduling scheme: based on the corrected reference load change curve and the widened confidence interval, construct an uncertain scenario set, construct a target function containing generation cost and load shortage, define the scenario that maximizes the target function in the uncertain scenario set as the worst load scenario, and continue to solve the target function under the worst load scenario to generate an elastic scheduling scheme with minimum generation cost and load shortage.

[0044] Set the constraint conditions of power grid operation, and the reference load change curve and the confidence interval constitute the load prediction interval, and generate a set representing the possible operation scenarios of the power grid based on the load prediction interval and the constraint conditions to generate the uncertain scenario set. One embodiment of the present application is: the constraint conditions include unit ramp rate constraint, reserve capacity constraint, and network safety transmission constraint.

[0045] The target function is the sum of the generation cost and the load shortage in the worst scenario, which can make the generated elastic scheduling scheme have the characteristics of economy and scheduling robustness.

[0046] The column constraint generation method is used to solve the target function, and the specific steps are: first, construct the main problem and fix the initial scenario to solve the scheduling scheme, then construct the sub-problem to find the worst scenario that maximizes the target function in the uncertainty set, add the worst scenario to the main problem and repeat the solution, until the convergence condition is met, then stop iteration, realize automatic identification of the worst load scenario in the confidence interval, and generate the elastic scheduling scheme with the minimum cost.

[0047] The elastic scheduling scheme includes a scheduling instruction framework and a local elastic adjustment rule of the framework. The scheduling instruction framework includes an output plan of each energy unit for energy supply and a power flow limit value of a power transmission line for energy transmission; and the local elastic adjustment rule of the framework includes a self-enabled backup power switching sequence when an actual load of a node deviates from a predicted value by a preset deviation threshold and a calling priority of a demand side response resource. An embodiment of the present application is that the backup power switching sequence is sorted according to a response time, and the calling priority of the demand side response resource is sorted according to a unit adjustment cost.

[0048] The generation logic of the local elastic adjustment rule of the framework includes establishing a backup resource response characteristic database, recording parameters including a gas turbine startup time, a storage system charging and discharging rate and an interruptible load reduction delay, the database data is derived from technical manuals and historical operation records provided by equipment manufacturers, is updated periodically, and is sorted according to a unit adjustment cost of each type of backup resource. When the actual load of the node deviates from the predicted value to different levels, different types of backup resources are preferentially called according to the above sorting results.

[0049] Scheme evaluation: the digital twin model of the power grid simulates the operation of a plurality of elastic scheduling schemes under different disturbance levels, and the scheme evaluation strategy determines the optimal elastic scheduling scheme based on the simulation results of a plurality of indicators corresponding to each elastic scheduling scheme under different disturbance levels.

[0050] The digital twin model of the power grid includes a unit dynamic characteristic model, a power transmission line thermal stability limit model and a load frequency response model. An embodiment of the present application is that when the digital twin model of the power grid is used to simulate the operation of the elastic scheduling scheme, the feasibility of the local elastic adjustment rule of the framework can be verified, scenarios in which the actual load deviates from the predicted value to different levels are deduced, and whether the calling of the backup resource meets the response time requirement and whether it causes system frequency or power flow overrun are checked. If there is a problem, the scheduling scheme is returned to the step of generating the scheduling scheme to solve the objective function again.

[0051] In order to screen out an elastic scheduling scheme with strong anti-disturbance ability, the digital twin model is simulated with increased load prediction error disturbance, unit fault disturbance and new energy output fluctuation disturbance. The disturbance range corresponding to each disturbance level is artificially set to inject disturbance interference into the digital twin model. Specifically, the load prediction error disturbance is that an error value is randomly generated based on the normal distribution of historical prediction errors and is superimposed on the predicted load; the unit fault disturbance is that a plurality of units are randomly selected to trigger a fault to reduce the output to 0, the fault lasts for a certain time, and the fault unit type is selected according to a certain proportion; and the new energy output fluctuation disturbance is that a fluctuation sequence is generated based on historical wind power / photovoltaic output data using a corresponding model and is superimposed on the predicted output of the new energy.

[0052] The scheme evaluation strategy includes: obtaining simulation results of each index of the elastic scheduling scheme under a certain disturbance level, determining the corresponding index level based on the simulation results of each index, and performing weighted summation on each index level, and the summation result is the evaluation score of the elastic scheduling scheme under a certain disturbance level. The evaluation scores of each elastic scheduling scheme under each disturbance level are calculated respectively, and the elastic scheduling scheme with the highest total score and meeting the threshold value of each disturbance level evaluation score is the optimal elastic scheduling scheme.

[0053] One embodiment of the present application is: the indexes include the system frequency deviation of the power grid, the key section overrun probability, and the standby capacity utilization rate, the evaluation standards and weights of each index level of each index are set, the evaluation score of the elastic scheduling scheme under a certain disturbance level is calculated by the weighted summation calculation method, and the closer the evaluation score is to the upper limit, the more robust the scheme is. The same method is used to obtain the evaluation scores of each elastic scheduling scheme under each disturbance level, and the scheme with the highest total score and meeting the set standard under each disturbance level is selected as the optimal elastic scheduling scheme.

[0054] Reverse optimization strategy: based on the digital twin model, difference data between actual operation and simulation operation is obtained, and the load prediction model and the objective function are optimized in reverse based on the difference data.

[0055] The difference data includes actual load deviation, unit output error and standby resource utilization ratio. The actual load deviation is fed back to the load prediction model and optimized, and the unit output error and the standby resource utilization ratio are fed back to the objective function and optimized.

[0056] One embodiment of the present application is: the correction rule of the prediction model is that when the continuous multiple deduction deviation reaches a set amplitude, an online incremental learning mechanism is triggered to update the graph convolution weight of the spatio-temporal graph neural network with a sliding time window.

[0057] One embodiment of the present application is: the load deficiency in the objective function is calculated by using a conditional value at risk model, and based on the unit output error and the standby resource utilization ratio, the difference in loss of load probability in power supply can be counted, when the difference between the loss of load probability predicted by the digital twin and the actual value reaches a set standard, the confidence level parameter of the conditional value at risk is adjusted according to the size relationship between the predicted value and the actual value, so that the objective function can adapt to the actual operation situation.

[0058] A power grid scheduling scheme generation system based on load optimization, as shown in Figure 2 The system includes a dynamic risk profiling module for running a risk assessment model, a hybrid prediction engine module for performing load prediction and load optimization steps, a multi-objective robust optimization decision module for performing a scheduling scheme generation step, and a digital twin closed-loop correction module for performing a scheme evaluation step.

[0059] The dynamic risk portrait module is used to access weather forecast data, satellite radar images and Internet of Things device data in real time. The weather forecast data is sourced from the meteorological data sharing platform of the provincial meteorological bureau, the satellite radar images are obtained by the national satellite meteorological center through a dedicated satellite receiving station, and the Internet of Things device data is gathered through the power Internet of Things platform. The module generates a dynamic load risk portrait by fusing multiple source heterogeneous data, quantifying the potential impact strength and range of extreme weather on the load; the weather forecast data includes high spatiotemporal resolution meteorological element data, and the Internet of Things devices include smart power meters, power transmission tower state sensors, distributed power supply monitoring terminals and user-side smart power consumption equipment.

[0060] The hybrid prediction engine module is used to efficiently and accurately predict the load change data of each node in the future period within the power grid coverage area. When no extreme disturbance is identified, only the load prediction step is run, and when an extreme disturbance is identified, the load prediction step and the load optimization step are run concentrically. The baseline load change curve is corrected and the corresponding confidence interval is dynamically widened.

[0061] The multi-objective robust optimization decision module takes the load point prediction value and the confidence interval as input, and constructs a target function that optimizes both economic efficiency and scheduling scheme robustness. The target function is defined as the sum of the minimum generation cost and the minimum load shortage in the worst case scenario; the constraint conditions of the target function include unit ramp rate constraints, reserve capacity constraints, and network safety transmission constraints.

[0062] The digital twin closed-loop correction module performs super-real-time deduction and stress testing on the several flexible scheduling schemes generated by the multi-objective robust optimization decision module in the power grid digital twin platform, selects the most robust scheme based on the simulation results, and feeds back the difference data between the actual scheduling results and the simulation results to the hybrid prediction engine and the multi-objective robust optimization decision module for online optimization fine-tuning of the model.

[0063] The construction method of the digital twin platform is as follows: a corresponding protocol is used to realize synchronous updating with the physical power grid, ensuring that the state deviation between the digital twin and the physical power grid is at a low level; the constructed digital twin includes unit dynamic characteristic models, power transmission line thermal stability limit models and load frequency response models.

[0064] The system cooperatively operates according to the following process: The timing control logic controls the dynamic risk portrait module to output updates at a preset period to trigger the hybrid prediction engine to respond at a minute level, controls the multi-objective robust optimization decision module to run at a set interval, and ensures that the digital twin preview is completed before the dispatching instruction is issued. The timing control is realized by combining a timer and event triggering, and the running states of each module are monitored in real time through heartbeat signals.

[0065] If any module runs overtime for multiple times in succession, an abnormal processing mechanism is triggered, and the timeout threshold of each module is set; the system automatically switches to a degraded mode of generating a scheduling scheme by using a historical similar scene matching strategy, the historical similar scene matching is retrieved from a historical database, the similarity calculation uses a corresponding algorithm, the input features are the current risk level, weather conditions and load level, and the scene that reaches the set similarity is determined as a similar scene, and the scheduling scheme with the highest robustness score in the similar scene is selected for execution.

[0066] In addition, through the man-machine cooperative interface, the risk portrait atlas, the prediction interval strip chart (a strip chart composed of a reference load change curve and a confidence interval), the optimization scheme comparison radar chart and the digital twin deduction video playback are visually pushed to the dispatcher to assist his final decision; the man-machine cooperative interface is constructed based on a corresponding platform, the risk portrait atlas is displayed in a grid heat map, and the colors correspond to the risk levels; the prediction interval strip chart is drawn using a corresponding tool, the horizontal axis is time, the vertical axis is load, the strip area is the confidence interval, and the middle solid line is the predicted value. The optimization scheme comparison radar chart includes economy, robustness, environmental protection and backup adequacy, and each dimension has a value in a specific interval; the digital twin deduction video playback is rendered using a corresponding technology, and the power grid topology, unit output and line flow change are displayed in real time, supporting different playback speeds and key node positioning viewing.

[0067] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Without departing from the spirit and scope of the present application, various changes and improvements can be made to the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection claimed by the present application is defined by the appended claims and their equivalents.

Claims

1. A method for generating power grid dispatch schemes based on load optimization, characterized in that, include: Load forecasting: Construct a power grid topology map. Based on the load data and meteorological data of each node in the power grid topology map for the corresponding time period, the load forecasting model predicts the baseline load change curve and confidence interval of each node for the future time period. Load optimization: The power grid coverage area is divided into several load areas. Based on the risk assessment model, the risk level of each load area is continuously obtained and load areas exceeding the preset level are defined as risk areas. The deviation prediction model predicts the predicted load offset and standard deviation caused by the sudden characteristics based on the sudden characteristic data of the risk area. The baseline load change curve of each node in the risk area is added to the predicted load offset to generate the corrected baseline load change curve. The standard deviation is used to widen the confidence interval. Generate a scheduling scheme: Based on the modified baseline load change curve and the widened confidence interval, construct a set of uncertain scenarios, construct an objective function that includes generation cost and load deficit, solve the objective function, and define the scenario that maximizes the solution in the set of uncertain scenarios as the worst load scenario. Under the worst load scenario, continue to solve the objective function to generate an elastic scheduling scheme that minimizes both generation cost and load deficit.

2. The method for generating a power grid dispatching scheme based on load optimization according to claim 1, characterized in that, The risk assessment model includes: continuously acquiring meteorological data, cloud map data, and IoT data; constructing a mapping relationship between meteorological data and load areas; determining the movement trajectory of clouds based on cloud map data; determining the load weight of each load area based on IoT data; and predicting the risk level of each load area based on the movement trajectory of clouds, meteorological data within each load area, and load weight.

3. The method for generating a power grid dispatching scheme based on load optimization according to claim 2, characterized in that, The IoT data includes total electricity consumption data for each load area and electricity meter data for electricity consumption in each load area. The method for generating load weights is as follows: obtain pedestrian flow data and corresponding total electricity consumption data for different load areas under different time types, and establish a pedestrian flow-load conversion coefficient table under different time types; Cluster the meter data of each load area to determine the electricity consumption pattern, and obtain the building density and average household capacity of each load area to calculate the corresponding load density base value; By integrating the population-load conversion coefficient table, load density base value, and electricity consumption pattern of each load area, the load weight of each load area is output.

4. The method for generating a power grid dispatching scheme based on load optimization according to claim 1, characterized in that, The sudden characteristic data includes temperature change gradient, wind speed change flag, cloud cover accumulation, and IoT device offline rate. The deviation prediction model includes an encoder that converts the sudden characteristic data into a latent space vector and a Bayesian linear regression calculation layer that calculates the predicted load offset and standard deviation based on the latent space vector.

5. The method for generating a power grid dispatching scheme based on load optimization according to claim 1, characterized in that, Solving the objective function involves: constructing a main problem and fixing the initial scenario to solve for the corresponding elastic scheduling scheme; then constructing a subproblem to find the worst scenario that maximizes the objective function in the set of uncertain scenarios; adding the worst scenario to the main problem and repeatedly solving for the elastic scheduling scheme; stopping the iteration when the convergence condition is met, and outputting the elastic scheduling scheme.

6. The method for generating a power grid dispatching scheme based on load optimization according to claim 1, characterized in that, The flexible scheduling scheme includes a scheduling instruction framework and flexible adjustment rules. The scheduling instruction framework includes the output plan of each energy unit for each node and the power flow limit value of the transmission line. The flexible adjustment rules include the corresponding self-activated backup power switching sequence and the priority of demand-side response resource call under different deviation levels between the predicted load and the actual load of the node.

7. The method for generating a power grid dispatching scheme based on load optimization according to claim 1, characterized in that, The method for generating power grid dispatch schemes includes scheme evaluation: a digital twin model of the power grid simulates the operation of several flexible dispatch schemes under different disturbance levels, and the scheme evaluation strategy determines the optimal flexible dispatch scheme based on the simulation results of several indicators corresponding to each flexible dispatch scheme under different disturbance levels.

8. The method for generating a power grid dispatching scheme based on load optimization according to claim 7, characterized in that, The disturbances include load forecasting error disturbances, unit fault disturbances, and new energy output fluctuation disturbances.

9. The method for generating a power grid dispatching scheme based on load optimization according to claim 7, characterized in that, The scheme evaluation strategy includes: obtaining the simulation results of each indicator of the elastic scheduling scheme under a certain disturbance level, determining the corresponding indicator level based on the simulation results of each indicator and performing weighted summation, and the summation result is the evaluation score of the elastic scheduling scheme under that disturbance level. Calculate the evaluation scores of several elastic scheduling schemes under each disturbance level. The elastic scheduling scheme with the highest total score and all evaluation scores at each disturbance level that meet the threshold is the optimal elastic scheduling scheme.

10. A power grid dispatching scheme generation system based on load optimization, comprising a power grid dispatching scheme generation method based on load optimization according to any one of claims 1-9, characterized in that, It includes a dynamic risk profiling module for determining the risk level of each load area, a hybrid forecasting engine module for executing load forecasting and load optimization steps, a multi-objective robust optimization decision module for executing the generation of scheduling schemes, and a digital twin closed-loop correction module for correcting and screening flexible scheduling schemes.

Images