Power system multi-space-time supply and demand risk identification method based on scene simulation

Abstract

The invention relates to a scene simulation-based multi-space-time supply and demand risk identification method for a power system, and the method comprises the steps: generating a database with a space-time label through a data collection and preprocessing unit, and constructing a power grid model; the risk identification module, the scene simulation module and the dynamic mapping module are used for displaying the risk and adequacy data of each region and each time period in the form of a thermodynamic diagram and a distribution diagram to form a power grid fluctuation region; the method comprises the following steps: acquiring different variables in a power grid system to generate dynamic nodes, forming edges among the nodes by an interdependence relationship among the variables, constructing a time-varying relationship network of each time period, combining the time-varying relationship network with a machine learning model, and constructing a prediction model reflecting nonlinear dynamic change of the power grid; a power grid model is adjusted by constructing a time-varying relationship network to carry out a nonlinear dynamic change prediction model, and the evaluation precision of power grid space-time distribution is ensured; the method has the advantages that the time-varying relationship network is established, the prediction precision and the response speed are improved, and risk factors are accurately captured.

Description

Technical Field

[0001] The present invention belongs to the technical field of smart grids, and in particular relates to a method for identifying multi-temporal and spatial supply and demand risks of power systems based on scenario simulation. Background Art

[0002] The grid spatiotemporal distribution assessment system is a comprehensive platform designed to conduct a comprehensive assessment of the grid's operating status, supply and demand balance, risk status, and adequacy at different times and locations. Its goal is to provide a scientific basis for operational management, emergency warning, and long-term planning through data integration, risk analysis, scenario simulation, and intuitive display. For example, announcement number CN110429596B discloses a distribution network reliability assessment method that takes into account the spatiotemporal distribution of electric vehicles, which belongs to the field of smart grids. The method includes: S1: constructing a charging / discharging behavior model based on electric vehicle user travel; S2: constructing a distribution network reliability assessment model that takes into account electric vehicles to the grid, including: improving the distribution network Monte Carlo reliability assessment simulation time advancement method, calculating the amount of electric vehicles participating in V2G call power and distribution network reliability indicators during the fault period; based on the charging and discharging behavior model of electric vehicles, this case , the traditional Monte Carlo simulation time advancement method of distribution network is improved, which avoids the invalid calculation amount caused by directly using traditional methods and improves the efficiency of simulation; in the spatiotemporal distribution assessment of power grid, the issues affecting the accuracy of assessment cover all links from data acquisition, preprocessing to model construction, scenario simulation, and even the final result display. The errors in each link will accumulate in the final assessment results. Therefore, ensuring high data quality, building a model that conforms to reality, reasonably designing scenarios, and continuously correcting and updating the model are the key to improving the overall assessment accuracy; in the power grid system, there are complex nonlinear dynamics and time-varying relationships. If the model ignores these characteristics, the electromagnetic interactions between each node, power source and load in the power grid, the power flow and voltage stability are all affected by nonlinearity. Especially when facing large-scale new energy access, various uncertainties make the system response more complicated. New energy sources such as wind and solar energy are affected by weather, The output is affected by seasonal and even long-term climate changes. There are large differences between the time periods and they are in a constant changing process; therefore, it is very necessary to provide a scenario simulation-based method for identifying multi-temporal and spatiotemporal supply and demand risks in power systems, which can establish a time-varying relationship network, improve prediction accuracy and response speed, and accurately capture risk factors. Summary of the Invention

[0003] (1) Technical issues

[0004] In view of the above-mentioned existing technical status, this application mainly addresses the following technical problems:

[0005] 1. In the assessment of the spatiotemporal distribution of power grids, issues affecting assessment accuracy encompass every stage, from data collection and preprocessing to model construction, scenario simulation, and even the final result presentation. Errors in each stage will accumulate in the final assessment results. Therefore, ensuring high data quality, building realistic models, rationally designing scenarios, and continuously calibrating and updating models are key to improving overall assessment accuracy.

[0006] 2. The power grid system is subject to complex nonlinear dynamics and time-varying relationships. Electromagnetic interactions, power flows, and voltage stability among nodes, power sources, and loads in the grid are all affected by nonlinearities. This is especially true when facing the integration of large-scale renewable energy sources. Various uncertainties complicate system responses. Renewable energy sources such as wind and solar are affected by long-term changes in weather, seasons, and even climate. Their output varies significantly over time and is in a constant state of flux.

[0007] (2) Technical solution

[0008] The purpose of the present invention is to overcome the shortcomings of the existing technology and provide a scenario simulation-based method for identifying multi-temporal and spatial supply and demand risks in power systems, which can establish a time-varying relationship network, improve prediction accuracy and response speed, and accurately capture risk factors.

[0009] The object of the present invention is achieved by providing a method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation, the method comprising the following steps:

[0010] Step 1: Collect power grid data through the data acquisition and preprocessing unit, generate a database with time and space labels, and build a power grid model;

[0011] Step 2: Use the risk identification module to identify risk factors affecting power supply and demand, build a risk matrix, and dynamically update the risk level;

[0012] Step 3: Use the scenario simulation module to simulate supply and demand changes under different scenarios based on the risk matrix and historical data, and generate multi-scenario plans;

[0013] Step 4: Use the dynamic mapping module to display the risk and adequacy data of each region and time period in the form of heat maps and distribution maps to form grid turbulence areas;

[0014] Step 5: Use the model correction unit to obtain different variables in the power grid system to generate dynamic nodes. The interdependence between variables constitutes the edges between nodes. The time dimension is introduced to divide the power grid data into different time periods. A time-varying relationship network is constructed for each time period. The time-varying relationship network is combined with the machine learning model to construct a prediction model that reflects the nonlinear dynamic changes of the power grid. The power grid model parameters are adjusted based on the prediction model.

[0015] Furthermore, the power grid data collected by the data acquisition preprocessing unit includes: voltage, current, load, meteorological conditions, new energy output and equipment status. The geographic information system is used to locate the power grid nodes, transmission lines, and substation equipment, and assign timestamps to form the spatial and temporal dual dimensions of the power grid data. The power grid data is cleaned, formatted, missing values ​​are filled, and noise is filtered.

[0016] Furthermore, the risk identification module extracts key influencing factors through correlation analysis to form characteristic indicators that describe power supply and demand behavior. Specifically, time series analysis, mean, variance, and deviation statistics are used to describe the fluctuations and abnormalities of variables; representative comprehensive indicators are extracted through principal component analysis to reduce data redundancy and highlight key risk signals; the correlation coefficient or mutual information between variables is calculated to determine risk characteristics that are highly correlated with historical risk data.

[0017] Furthermore, the risk identification module constructs a risk matrix, which specifically includes the following steps:

[0018] Step 1: Define the coordinate axes to construct a two-dimensional matrix, where the horizontal axis represents the frequency of risk occurrence and the vertical axis represents the degree of risk impact;

[0019] Step 2: Based on historical data distribution and statistical analysis, classify the probability and impact into low, medium, and high levels, and define the corresponding risk level for each combination;

[0020] Step 3: Calculate the risk score using the weighted sum or product method;

[0021] Step 4: Construct a multi-dimensional risk matrix to reflect the superposition effect and interaction of different risk factors in different regions and time periods of the power grid.

[0022] Furthermore, the scenario simulation module generates multiple scenario plans, specifically: simulating the response changes of power grid supply and demand and key processes under different risk combinations and external conditions to form several scenario plans, and constructing a scenario tree model based on the combination of risk factors and conditional probability. Each branch of the scenario tree represents a risk combination or a change in key control strategy. Through the scenario tree decomposition, the influence of each variable is expanded step by step, and the response effect of the power grid process at different decision nodes is evaluated to form a complete event path and plan set.

[0023] Furthermore, the construction of dynamic nodes and edges in the model correction unit includes the following steps:

[0024] Step 1: Determine the variables or grid data indicators used to reflect the operating status of the power grid. The variables or grid data indicators are regarded as dynamic nodes in the power grid. For each node, a feature vector describing its current state is constructed.

[0025] Step 2: Calculate the correlation coefficient between each pair of nodes, use the calculated correlation coefficient as the initial weight of the edge, and smooth the initial weight;

[0026] Step 3: Set a threshold and construct an edge after the correlation coefficient reaches the threshold.

[0027] Furthermore, the model correction unit introduces a time dimension to construct a time-varying relationship network, including the following steps:

[0028] Step 1: The time period is divided into fixed time periods and free adaptive time windows. According to the operation characteristics of the power grid, the fixed time window is pre-set to divide the continuous time series data into multiple independent time periods. The segmentation boundaries are dynamically determined by using the mutation points or periodic changes of the data;

[0029] Step 2: Use sliding window technology to construct overlapping time periods in the power grid data, and independently construct network snapshots within each window to capture the continuous changes of the power grid in time;

[0030] Step 3: All nodes and their edges in each time period constitute a network snapshot, depicting the dependency relationship between variables in that period.

[0031] Furthermore, the sequence construction of the time-varying relationship network includes the following steps:

[0032] Step 1: Sort the network snapshots in all time periods in chronological order to obtain a series of dynamic graphs;

[0033] Step 2: Extract global and local network features, i.e., node degree, centrality, and clustering coefficient, from the network snapshot and determine how the indicators change over time;

[0034] Step 3: Using the dynamic graph sequence as input, train the dynamic graph neural network to learn and predict the changes in the network structure over the entire time series;

[0035] Step 4: Use online learning algorithms to update the dynamic graph neural network parameters in real time.

[0036] Furthermore, the time-varying relationship network is combined with the machine learning model to construct a prediction model, including the following steps:

[0037] Step 1: Using the node features and edge weight matrix constructed in each time period as input, the graph convolution layer is used to extract spatial structural features;

[0038] Step 2: Input the graph representation sequence extracted from each time period into the recurrent neural network to learn the temporal dependency;

[0039] Step 3: Using the historical time-varying relationship network and the corresponding power grid operation indicators, the prediction model is trained through supervised learning methods;

[0040] Step 4: Select a loss function and optimize it according to the prediction target;

[0041] Step 5: Predict the operating status, risk changes, and early warning signals of the power grid in several time periods in the future.

[0042] Furthermore, the model correction unit adjusts the power grid model parameters based on the prediction model, specifically: the prediction model predicts the supply and demand situation and key indicators in the next period or several periods in the future, collects the actual power grid operation data provided by the real-time collection department, compares the predicted value with the actual measured value, calculates the error, determines the power grid operation adjustment parameters according to actual needs, establishes a mapping relationship based on the prediction output and the error input, and converts the prediction results into parameter adjustment instructions.

[0043] (3) Beneficial effects

[0044] 1. The present invention adjusts the power grid model by constructing a time-varying relationship network to perform a nonlinear dynamic change prediction model, thereby ensuring the accuracy of the temporal and spatial distribution assessment of the power grid;

[0045] 2. When the forecast model indicates increased near-term risk or a trend of imbalance in supply and demand, relevant parameters are adjusted in advance to reduce potential risks. Model parameters are updated in real time based on forecast results, enabling the dispatch center to support dispatch decisions with more accurate data and achieve refined management. Parameters are adjusted online based on feedback error information, forming an adaptive closed loop to continuously correct model deviations and improve forecast accuracy and response speed.

[0046] 3. It can accurately capture major risk factors such as power supply fluctuations, demand changes, and equipment abnormalities, convert large amounts of raw data into quantitative indicators with practical significance, and provide scientific and powerful data support for the construction of risk matrices, dynamic risk assessments, and subsequent scenario simulations.

[0047] Other features and advantages of the present invention will be described in the detailed description that follows, and some will become apparent from the description or be understood through implementation of the present invention; the purposes and other advantages of the present invention can be realized and obtained through the structures particularly pointed out in the written description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 It is a structural block diagram of the present invention. DETAILED DESCRIPTION

[0049] One of the purposes of the present invention is to provide a method for identifying multi-temporal and spatial supply and demand risks of power systems based on scenario simulation, by establishing a time-varying relationship network to determine the time-varying relationship, analyzing the relationship network to determine the persistence of the interactive relationship, and determining nonlinear dynamic changes based on the relationship network, so as to improve the model accuracy.

[0050] The present invention will be further described below with reference to the embodiments and / or drawings.

[0051] Example 1

[0052] like Figure 1 As shown, a method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation includes the following steps:

[0053] Step 1: Collect power grid data through the data acquisition and preprocessing unit, generate a database with time and space labels, and build a power grid model;

[0054] In the present invention, the collected power grid data includes: voltage, current, load, meteorological conditions, renewable energy output and equipment status. The geographic information system is used to locate the power grid nodes, transmission lines and substation equipment, and assign timestamps to form the spatial and temporal dual dimensions of the power grid data. The power grid data is then cleaned, formatted, missing values ​​are filled and noise filtered.

[0055] In this embodiment, by processing the power grid data to form a database, the construction of the overall power grid model is supported, accurate data input is provided for subsequent risk identification and scenario simulation, and the deep integration of data and models is achieved to improve prediction accuracy.

[0056] Step 2: Use the risk identification module to identify risk factors affecting power supply and demand, build a risk matrix, and dynamically update the risk level;

[0057] In the present invention, ① the risk identification module identifies risk factors by: extracting key influencing factors through correlation analysis to form characteristic indicators describing power supply and demand behavior; using time series analysis, mean, variance, and deviation statistics to describe the fluctuations and abnormalities of variables;

[0058] Extract representative comprehensive indicators through principal component analysis, reduce data redundancy, and highlight key risk signals;

[0059] Calculate the correlation coefficient or mutual information between variables and identify risk characteristics that are highly correlated with historical risk data.

[0060] In this embodiment, after analyzing the risk features and extracting a large number of candidate features, the following features are identified to determine if they truly reflect the risk:

[0061] Correlation and causal analysis uses statistical indicators such as Pearson correlation statistics to determine the correlation between each feature and known risk events, and applies causal relationship analysis to determine whether specific indicators can serve as leading signals of risk changes.

[0062] For the risk features that are initially screened, model validation methods such as cross-validation and holdout are used to evaluate their stability and generalization capabilities in different scenarios, time periods, and grid partitions, so as to determine the key factors that ultimately constitute the risk matrix.

[0063] ②The method for constructing the risk matrix in the risk identification module is as follows:

[0064] Define the coordinate axes to construct a two-dimensional matrix, where the horizontal axis represents the frequency of risk occurrence and the vertical axis represents the degree of risk impact;

[0065] Based on historical data distribution and statistical analysis, both probability and impact are divided into low, medium and high levels, and the corresponding risk level of each combination is defined;

[0066] Risk scores were calculated using a weighted sum or product approach;

[0067] A multidimensional risk matrix is ​​constructed. For each risk factor, the risk score is calculated based on the normalized indicators and weights. The weighted average is used to obtain the comprehensive risk value of each risk factor in a specific time period or region. The overall data distribution is used to set the threshold of the risk score, and the risks are divided into low, medium and high levels. This forms a risk matrix with distinguishing capabilities in spatial and temporal dimensions, reflecting the superposition effects and interactions of different risk factors in various regions and time periods of the power grid.

[0068] Step 3: Use the scenario simulation module to simulate supply and demand changes under different scenarios based on the risk matrix and historical data, and generate multi-scenario plans;

[0069] The present invention specifically simulates the response changes of power grid supply and demand and key processes under different risk combinations and external conditions to form several scenario plans. According to the combination of risk factors and conditional probability, a scenario tree model is constructed. Each branch of the scenario tree represents a risk combination or a change in key control strategy. Through the scenario tree decomposition, the influence of each variable is expanded step by step, and the response effect of the power grid process at different decision nodes is evaluated to form a complete event path and plan set.

[0070] Step 4: Use the dynamic mapping module to display the risk and adequacy data of each region and time period in the form of heat maps and distribution maps to form grid turbulence areas;

[0071] Step 5: Use the model correction unit to obtain different variables in the power grid system to generate dynamic nodes. The interdependence between variables constitutes the edges between nodes. The time dimension is introduced to divide the power grid data into different time periods. A time-varying relationship network is constructed for each time period. The time-varying relationship network is combined with the machine learning model to construct a prediction model that reflects the nonlinear dynamic changes of the power grid. The power grid model parameters are adjusted based on the prediction model.

[0072] In the present invention, ① a method for constructing dynamic nodes and edges:

[0073] Determine variables or grid data indicators used to reflect the operating status of the power grid, the variables or grid data indicators are used as dynamic nodes in the power grid, and construct a feature vector describing the current state of each node;

[0074] Calculate the correlation coefficient between each pair of nodes, use the calculated correlation coefficient as the initial weight of the edge, and smooth the initial weight;

[0075] Set a threshold, and build an edge after the correlation coefficient reaches the threshold.

[0076] ② The steps to introduce the time dimension and construct a time-varying relationship network are as follows:

[0077] The time period is divided into fixed time periods and free adaptive time windows. According to the operation characteristics of the power grid, the fixed time window is pre-set to divide the continuous time series data into multiple independent time periods. The segmentation boundaries are dynamically determined by using the mutation points or periodic changes of the data;

[0078] Sliding window technology is used to construct overlapping time periods in the power grid data, and a network snapshot is independently constructed within each window to capture the continuous changes of the power grid in time;

[0079] All nodes and their edges in each time period constitute a network snapshot, depicting the dependency relationships between variables during that period.

[0080] ③ Sequence construction of time-varying relationship network:

[0081] Sort the network snapshots in all time periods into time series to obtain a series of dynamic graphs;

[0082] Extract global and local network features, i.e., node degree, centrality, and clustering coefficient, from the network snapshot to determine how the indicators change over time;

[0083] Using dynamic graph sequences as input, we train dynamic graph neural networks to learn and predict changes in network structure over the entire time series.

[0084] The online learning algorithm is used to update the parameters of the dynamic graph neural network in real time.

[0085] ④ Combining the time-varying relationship network with the machine learning model to build a prediction model: Taking the node features and edge weight matrix constructed in each time period in the time-varying relationship network as input, the graph convolution layer is used to extract spatial structure features;

[0086] The graph representation sequence extracted from each time period is input into the recurrent neural network to learn the temporal dependency;

[0087] Using the historical time-varying relationship network and the corresponding power grid operation indicators, the prediction model is trained through supervised learning methods;

[0088] Select a loss function and optimize it based on the prediction target;

[0089] Predict the operating status, risk changes and early warning signals of the power grid in several time periods in the future.

[0090] ⑤ The prediction model predicts the supply and demand situation and key indicators in the next period or several periods in the future, collects the actual operation data of the power grid provided by the real-time collection department, compares the predicted value with the actual measurement value, calculates the error, determines the power grid operation adjustment parameters according to actual needs, establishes a mapping relationship between the prediction output and the error input, and converts the prediction results into parameter adjustment instructions.

[0091] The present invention is a method for identifying multi-temporal and spatial supply and demand risks of power systems based on scenario simulation. During use, the present invention adjusts the power grid model by constructing a time-varying relationship network to carry out a nonlinear dynamic change prediction model, thereby ensuring the accuracy of the temporal and spatial distribution assessment of the power grid; when the prediction model indicates that the recent risk increases or the supply and demand balance shows an imbalance trend, the relevant parameters are adjusted in advance to reduce potential risks, and the model parameters are updated in real time based on the prediction results, so that the dispatching center can support dispatching decisions with more accurate data and realize refined management. The parameters are adjusted online based on feedback error information to form an adaptive closed loop, continuously correct model deviations, and improve prediction accuracy and response speed; the present invention has the advantages of establishing a time-varying relationship network, improving prediction accuracy and response speed, and accurately capturing risk factors.

[0092] Example 2

[0093] like Figure 1 As shown, a method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation is provided. The method improves the evaluation accuracy of the model through a time-varying relationship network, thereby increasing the accuracy of supply and demand risk identification and scenario simulation. The method includes the following steps:

[0094] Step 1: Collect power grid data through the data acquisition and preprocessing unit, generate a database with time and space labels, and build a power grid model;

[0095] Step 2: Use the risk identification module to identify risk factors affecting power supply and demand, build a risk matrix, and dynamically update the risk level;

[0096] In the present invention, the method for constructing the risk matrix by the risk identification module is as follows:

[0097] Define the coordinate axes to construct a two-dimensional matrix, where the horizontal axis represents the frequency of risk occurrence and the vertical axis represents the degree of risk impact;

[0098] Based on historical data distribution and statistical analysis, both probability and impact are divided into low, medium and high levels, and the corresponding risk level of each combination is defined;

[0099] Risk scores were calculated using a weighted sum or product approach;

[0100] A multidimensional risk matrix is ​​constructed. For each risk factor, the risk score is calculated based on the normalized indicators and weights. The weighted average is used to obtain the comprehensive risk value of each risk factor in a specific time period or region. The overall data distribution is used to set the threshold of the risk score, and the risks are divided into low, medium and high levels. This forms a risk matrix with distinguishing capabilities in spatial and temporal dimensions, reflecting the superposition effects and interactions of different risk factors in various regions and time periods of the power grid.

[0101] In this embodiment, in the risk matrix, the data acquisition system is used to input the latest data into the risk matrix. Based on real-time data and historical trends, machine learning is used to adjust the risk indicators and weights to ensure that the matrix always reflects the current implementation situation. The dynamic matrix is ​​used as the initial condition and combined with the simulation method to improve the risk plan.

[0102] For example, the risk score = w1×P+w2×I, where P represents the probability level; I represents the impact level; w1 and w2 are preset weights, and the weight factors are dynamically adjusted to adapt to risk changes in different scenarios.

[0103] In the coordinate axis, the horizontal axis is the frequency of risk occurrence, the higher the frequency, the more likely it is to occur, and the vertical axis is the degree of risk impact, such as the degree of power shortage and the magnitude of load abnormality.

[0104] Step 3: Use the scenario simulation module to simulate supply and demand changes under different scenarios based on the risk matrix and historical data, and generate multi-scenario plans;

[0105] The present invention specifically simulates the response changes of power grid supply and demand and key processes under different risk combinations and external conditions to form several scenario plans. According to the combination of risk factors and conditional probability, a scenario tree model is constructed. Each branch of the scenario tree represents a risk combination or a change in key control strategy. Through the scenario tree decomposition, the influence of each variable is expanded step by step, and the response effect of the power grid process at different decision nodes is evaluated to form a complete event path and plan set.

[0106] In this embodiment, the risk factors in each scenario are comprehensively evaluated, and the comprehensive risk score is calculated by weighted summation or product method. According to the plan risk score and its corresponding process changes, the plan is divided into low risk, medium risk, and high risk levels, and specific operational recommendations are specified for each level.

[0107] Output method of multi-scenario plan:

[0108] Quantitatively and in detail describe risk indicators, process changes, and emergency measures under different scenarios;

[0109] The changes in each area and each stage of the power grid in each scenario are intuitively presented through heat maps, flow charts and time sequence diagrams;

[0110] Real-time warning, combined with dynamically updated real-time data, compares simulation results with actual working conditions, and promptly issues warning information and control suggestions for scenarios that exceed risk warning thresholds.

[0111] Step 4: Use the dynamic mapping module to display the risk and adequacy data of each region and time period in the form of heat maps and distribution maps to form grid turbulence areas;

[0112] Step 5: Use the model correction unit to obtain different variables in the power grid system to generate dynamic nodes. The interdependence between variables constitutes the edges between nodes. The time dimension is introduced to divide the power grid data into different time periods. A time-varying relationship network is constructed for each time period. The time-varying relationship network is combined with the machine learning model to construct a prediction model that reflects the nonlinear dynamic changes of the power grid. The power grid model parameters are adjusted based on the prediction model.

[0113] ① Dynamic node and edge construction method:

[0114] Determine variables or grid data indicators used to reflect the operating status of the power grid, the variables or grid data indicators are used as dynamic nodes in the power grid, and construct a feature vector describing the current state of each node;

[0115] Calculate the correlation coefficient between each pair of nodes, use the calculated correlation coefficient as the initial weight of the edge, and smooth the initial weight;

[0116] Set a threshold, and build an edge after the correlation coefficient reaches the threshold.

[0117] In this embodiment, the directionality of the edge is based on the needs of the power grid scenario. If there is an obvious causal relationship, a directed edge is used to represent the sequence relationship. For example, the change in renewable energy output precedes the load response in time. Conversely, when there is no obvious factor relationship, an undirected edge is used to represent the bidirectional dependency relationship between variables.

[0118] For example, three key variables are determined in the power grid, namely, node A (wind farm output), B (regional load), and C (substation voltage), and two consecutive time periods, T1 and T2, are selected.

[0119] Time period T1 (08:00-08:30)

[0120] Node A (wind power output): average value: 200MW, standard deviation: 20MW;

[0121] Node B (regional load): average: 180MW, standard deviation: 15MW;

[0122] Node C (substation voltage): average value: 110 kV, standard deviation: 2 kV.

[0123] After preprocessing, each node forms a feature vector, and the mean and standard deviation are used to form the vector [mean, standard deviation].

[0124] Node A feature vector: [200, 20], node B feature vector: [180, 15], node C feature vector: [110, 2], the features describe the status of each node in time period T1.

[0125] Using sampling data and time series statistical methods, the correlation between nodes is calculated (for example, using the Pearson correlation coefficient): A and B: The calculated correlation coefficient is 0.8 → indicating that the relationship between wind power output and load is strong during this period.

[0126] B and C: The calculated correlation coefficient is -0.6 → The strong negative correlation may indicate that the voltage may decrease when the load increases (showing negative feedback characteristics).

[0127] A and C: The calculated correlation coefficient is 0.2 → The correlation is weak, and it can be considered that there is no obvious relationship between the two during the T1 period.

[0128] Set a correlation threshold (for example, 0.5), and only relationships with an absolute value of the correlation coefficient greater than 0.5 are considered valid edges:

[0129] AB: |0.8|>0.5, retain edge (weight 0.8);

[0130] BC: |–0.6|>0.5, retain the edge (weight -0.6);

[0131] AC: |0.2|<0.5, discarded, does not constitute an edge.

[0132] Therefore, in period T1, the constructed network diagram is:

[0133] Node set: {A, B, C}; Edge set: {(A, B, 0.8), (B, C, -0.6)}; Weight indicates the strength of correlation.

[0134] Edge set: edge AB, weight 0.8, edge B–C, weight -0.6.

[0135] It reflects the interdependence between variables during the T1 period.

[0136] ② The steps to introduce the time dimension and construct a time-varying relationship network are as follows:

[0137] The time period is divided into fixed time periods and free adaptive time windows. According to the operation characteristics of the power grid, the fixed time window is pre-set to divide the continuous time series data into multiple independent time periods. The segmentation boundaries are dynamically determined by using the mutation points or periodic changes of the data;

[0138] Sliding window technology is used to construct overlapping time periods in the power grid data, and a network snapshot is independently constructed within each window to capture the continuous changes of the power grid in time;

[0139] All nodes and their edges in each time period constitute a network snapshot, depicting the dependency relationships between variables during that period.

[0140] ③ Sequence construction of time-varying relationship network:

[0141] Sort the network snapshots in all time periods into time series to obtain a series of dynamic graphs;

[0142] Extract global and local network features, i.e., node degree, centrality, and clustering coefficient, from the network snapshot to determine how the indicators change over time;

[0143] Using dynamic graph sequences as input, we train dynamic graph neural networks to learn and predict changes in network structure over the entire time series.

[0144] The online learning algorithm is used to update the parameters of the dynamic graph neural network in real time.

[0145] ④ Combining the time-varying relationship network with the machine learning model to build a prediction model: Taking the node features and edge weight matrix constructed in each time period in the time-varying relationship network as input, the graph convolution layer is used to extract spatial structure features;

[0146] The graph representation sequence extracted from each time period is input into the recurrent neural network to learn the temporal dependency;

[0147] Using the historical time-varying relationship network and the corresponding power grid operation indicators, the prediction model is trained through supervised learning methods;

[0148] Select a loss function and optimize it based on the prediction target;

[0149] Predict the operating status, risk changes and early warning signals of the power grid in several time periods in the future.

[0150] In this embodiment, for example, the grid data is used as a time window every half hour, and three consecutive time periods T1, T2, and T3 are selected as training data. The following operations are performed in each time period:

[0151] In each time period, wind power output, regional load, and substation voltage are collected, and statistics (such as mean and standard deviation) are calculated to form node feature vectors.

[0152] Feature example: Using the mean and standard deviation to form a vector, the feature of node A in T1 is [300,15] (300MW, standard deviation 15).

[0153] Taking T1 as an example, assume that the following statistical data is obtained after data processing (the sample data are fictitious values ​​and are only used to illustrate the process):

[0154] Period T1:

[0155] Node A (wind power output): mean = 300MW, standard deviation = 15;

[0156] Node B (regional load): mean = 280MW, standard deviation = 20;

[0157] Node C (substation voltage): mean = 110 kV, standard deviation = 2.

[0158] Based on the original sampling data, the correlation between nodes in T1 is obtained by calculating the Pearson correlation coefficient:

[0159] A and B: correlation coefficient = 0.85;

[0160] B and C: correlation coefficient = –0.70;

[0161] A and C: correlation coefficient = 0.30;

[0162] After setting the correlation absolute value threshold to 0.5, only the following are retained:

[0163] Edge AB (weight 0.85); edge BC (weight –0.70).

[0164] Therefore, the network snapshot at period T1 contains 3 nodes and 2 valid edges.

[0165] Similarly, in time periods T2 and T3, we can calculate:

[0166] T2: Node A: feature [320, 10], Node B: feature [300, 18], Node C: feature [108, 3];

[0167] Correlation: AB = 0.90, BC = –0.65, AC = 0.40 → construct edges AB (0.90), BC (–0.65);

[0168] T3: Node A: feature [310, 20], Node B: feature [310, 25], Node C: feature [107, 2];

[0169] Correlation: AB = 0.75, BC = –0.55, AC = 0.35 → construct edges AB(0.75), BC(–0.55).

[0170] The network snapshots obtained over three time periods are a "dynamic graph sequence" that reflects the time-varying dependencies between key variables.

[0171] Assuming we want to predict the changes in regional load within T4 and the possible risk score, the training process is as follows:

[0172] The training data is constructed by using multiple time periods in the past (e.g., T1 to T3) to form the model input X∈R T×N×F , where T represents the time window length, N is the number of nodes (3), and F is the node feature dimension.

[0173] Supervision tags Supervision signals may include the actual load value during the T4 period or the risk level scored by experts.

[0174] The training objective is to minimize the error between the predicted and actual values ​​(e.g., mean squared error (MSE)).

[0175] After training, the model can predict the state of key variables of the power grid in the future period (T4).

[0176] The load in the T4 area is predicted to rise to 330MW;

[0177] The corresponding risk score is predicted to be 0.8 (assuming 1 indicates high risk);

[0178] Based on the prediction results, the dispatching center can adjust the control strategy in advance, such as increasing spare capacity or optimizing new energy dispatch, to reduce the risk of supply and demand imbalance.

[0179] ⑤ The prediction model predicts the supply and demand situation and key indicators in the next period or several periods in the future, collects the actual operation data of the power grid provided by the real-time collection department, compares the predicted value with the actual measurement value, calculates the error, determines the power grid operation adjustment parameters according to actual needs, establishes a mapping relationship between the prediction output and the error input, and converts the prediction results into parameter adjustment instructions.

[0180] In this embodiment, when the predicted regional load is higher than the warning threshold, the spare capacity is adjusted or more generator sets are dispatched online; when the predicted risk score increases, the sensitivity of the risk warning is increased or emergency response measures are triggered.

[0181] When the forecasting model indicates that the risk in the near future is increasing or the supply and demand balance is showing an imbalance trend, the relevant parameters are adjusted in advance to reduce potential risks, and the model parameters are updated in real time based on the forecast results, so that the dispatching center can support dispatching decisions with more accurate data and realize refined management. The parameters are adjusted online based on the feedback error information to form an adaptive closed loop, continuously correct the model deviation, and improve the forecast accuracy and response speed.

[0182] In summary, in the present invention, the power grid spatiotemporal distribution assessment system utilizes big data collection, in-depth analysis, and refined simulation to build a closed loop from data to decision-making, providing a solid data foundation and decision-making support for ensuring the safe, stable, and efficient operation of the power grid.

[0183] Among them, through real-time monitoring and risk identification, potential supply and demand imbalance risks can be quickly identified, early warnings can be issued, the supply and demand conditions and abundance in different regions and time periods can be revealed, and backup capacity and energy storage systems can be deployed in advance in weak areas to achieve optimized energy distribution across regions and time periods.

[0184] Through dynamic mapping and model correction, the power grid model is continuously optimized to ensure its high accuracy in a complex and changing power grid environment, thereby effectively improving the reliability and economy of power grid operation.

[0185] The present invention is a method for identifying multi-temporal and spatial supply and demand risks of power systems based on scenario simulation. During use, when the prediction model indicates that the risk in the near future is increasing or the supply and demand balance is out of balance, the relevant parameters are adjusted in advance to reduce potential risks. The model parameters are updated in real time based on the prediction results, so that the dispatching center can support dispatching decisions with more accurate data and realize refined management. The parameters are adjusted online based on feedback error information to form an adaptive closed loop, and the model deviation is continuously corrected to improve the prediction accuracy and response speed. The method can accurately capture the main risk factors such as power supply side fluctuations, demand side changes and equipment abnormalities, convert a large amount of raw data into quantitative indicators with practical significance, and provide scientific and powerful data support for the construction of risk matrix, dynamic risk assessment and subsequent scenario simulation. The present invention has the advantages of establishing a time-varying relationship network, improving prediction accuracy and response speed, and accurately capturing risk factors.

[0186] Although the present invention is disclosed in conjunction with the above embodiments, it is not intended to limit the present invention. Anyone familiar with this technology can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the definition of the attached claims.

Claims

1. A method for identifying multi-temporal and spatial supply and demand risks in power systems based on scenario simulation, characterized by: The method comprises the following steps: Step 1: Collect power grid data through the data acquisition and preprocessing unit, generate a database with time and space labels, and build a power grid model; Step 2: Use the risk identification module to identify risk factors affecting power supply and demand, build a risk matrix, and dynamically update the risk level; Step 3: Use the scenario simulation module to simulate supply and demand changes under different scenarios based on the risk matrix and historical data, and generate multi-scenario plans; Step 4: Use the dynamic mapping module to display the risk and adequacy data of each region and time period in the form of heat maps and distribution maps to form grid turbulence areas; Step 5: Use the model correction unit to obtain different variables in the power grid system to generate dynamic nodes. The interdependence between variables constitutes the edges between nodes. The time dimension is introduced to divide the power grid data into different time periods. A time-varying relationship network is constructed for each time period. The time-varying relationship network is combined with the machine learning model to construct a prediction model that reflects the nonlinear dynamic changes of the power grid. The power grid model parameters are adjusted based on the prediction model.

2. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 1, characterized in that: The power grid data collected by the data acquisition and preprocessing unit include: voltage, current, load, weather conditions, new energy output and equipment status. The geographic information system is used to locate the power grid nodes, transmission lines, and substation equipment, and assign timestamps to form the spatial and temporal dual dimensions of the power grid data. The power grid data is cleaned, formatted, missing values ​​are filled, and noise is filtered.

3. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 2, characterized in that: The risk identification module extracts key influencing factors through correlation analysis to form characteristic indicators that describe power supply and demand behavior. Specifically, it uses time series analysis, mean, variance, and deviation statistics to describe the fluctuations and anomalies of variables; and extracts representative comprehensive indicators through principal component analysis to reduce data redundancy and highlight key risk signals. Calculate the correlation coefficient or mutual information between variables and identify risk characteristics that are highly correlated with historical risk data.

4. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 3, characterized in that: The risk identification module constructs a risk matrix, specifically including the following steps: Step 1: Define the coordinate axes to construct a two-dimensional matrix, where the horizontal axis represents the frequency of risk occurrence and the vertical axis represents the degree of risk impact; Step 2: Based on historical data distribution and statistical analysis, classify the probability and impact into low, medium, and high levels, and define the corresponding risk level for each combination; Step 3: Calculate the risk score using the weighted sum or product method; Step 4: Construct a multi-dimensional risk matrix to reflect the superposition effect and interaction of different risk factors in different regions and time periods of the power grid.

5. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 1, characterized in that: The scenario simulation module generates multiple scenario plans, specifically by simulating the response changes of power grid supply and demand and key processes under different risk combinations and external conditions to form several scenario plans. Based on the combination of risk factors and conditional probabilities, a scenario tree model is constructed. Each branch of the scenario tree represents a risk combination or a change in key control strategy. Through the scenario tree decomposition, the influence of each variable is expanded step by step, and the response effect of the power grid process at different decision nodes is evaluated to form a complete event path and plan set.

6. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 1, characterized in that: The construction of dynamic nodes and edges in the model correction unit includes the following steps: Step 1: Determine the variables or grid data indicators used to reflect the operating status of the power grid. The variables or grid data indicators are regarded as dynamic nodes in the power grid. For each node, a feature vector describing its current state is constructed. Step 2: Calculate the correlation coefficient between each pair of nodes, use the calculated correlation coefficient as the initial weight of the edge, and smooth the initial weight; Step 3: Set a threshold and construct an edge after the correlation coefficient reaches the threshold.

7. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 6, characterized in that: The model correction unit introduces a time dimension to construct a time-varying relationship network, including the following steps: Step 1: The time period is divided into fixed time periods and free adaptive time windows. According to the operation characteristics of the power grid, the fixed time window is pre-set to divide the continuous time series data into multiple independent time periods. The segmentation boundaries are dynamically determined by using the mutation points or periodic changes of the data; Step 2: Use sliding window technology to construct overlapping time periods in the power grid data, and independently construct network snapshots within each window to capture the continuous changes of the power grid in time; Step 3: All nodes and their edges in each time period constitute a network snapshot, depicting the dependency relationship between variables in that period.

8. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 7, characterized in that: The sequence construction of the time-varying relationship network includes the following steps: Step 1: Sort the network snapshots in all time periods in chronological order to obtain a series of dynamic graphs; Step 2: Extract global and local network features, i.e., node degree, centrality, and clustering coefficient, from the network snapshot and determine how the indicators change over time; Step 3: Using the dynamic graph sequence as input, train the dynamic graph neural network to learn and predict the changes in the network structure over the entire time series; Step 4: Use online learning algorithms to update the dynamic graph neural network parameters in real time.

9. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 8, characterized in that: The time-varying relationship network is combined with the machine learning model to construct a prediction model, including the following steps: Step 1: Using the node features and edge weight matrix constructed in each time period as input, the graph convolution layer is used to extract spatial structural features; Step 2: Input the graph representation sequence extracted from each time period into the recurrent neural network to learn the temporal dependency; Step 3: Using the historical time-varying relationship network and the corresponding power grid operation indicators, the prediction model is trained through supervised learning methods; Step 4: Select a loss function and optimize it according to the prediction target; Step 5: Predict the operating status, risk changes, and early warning signals of the power grid in several time periods in the future.

10. The method for identifying multi-temporal and spatial supply and demand risks of a power system based on scenario simulation according to claim 9, characterized in that: The model correction unit adjusts the power grid model parameters based on the prediction model, specifically: the prediction model predicts the supply and demand situation and key indicators in the next period or several future periods, collects the actual power grid operation data provided by the real-time collection department, compares the predicted value with the actual measured value, calculates the error, determines the power grid operation adjustment parameters according to actual needs, establishes a mapping relationship based on the prediction output and the error input, and converts the prediction results into parameter adjustment instructions.

Images