Method and device for evaluating risk of power transmission section over-limit based on meteorological large model

Abstract

The application provides a power transmission section tide flow over-limit risk assessment method and device based on a meteorological large model, which comprises the following steps: based on future meteorological data predicted by a meteorological large model, constructing a power prediction model and a power price prediction model based on a TimesNet-TCN-CBAM hybrid model framework, and outputting regional predicted power in a future medium and long term and regional predicted power price in a future medium and long term; constructing a power price-tide flow sample tensor; based on a linearized direct current tide flow formula, calculating predicted tide flow of each power transmission section in a future time period t, and calculating a tide flow over-limit probability; when the tide flow over-limit probability exceeds a threshold value, triggering a preventive control strategy to be executed. The application directly converts medium and long term power price-power prediction results into quantitative assessment and real-time control instructions of section tide flow over-limit risk, forms an integrated technical solution from market signals to physical power grid safety, and significantly reduces power transmission congestion and power failure risks caused by power price fluctuations.

Description

Technical Field

[0001] This invention belongs to the field of new energy power technology, and in particular relates to a method and device for assessing the risk of power flow exceeding the limit of transmission sections based on a large meteorological model. Background Technology

[0002] Transmission section refers to the transmission cross-section formed by a group of interconnected transmission lines (or transformers) in a power grid. With the deepening of the unified electricity market, the volume of inter-provincial medium- and long-term transactions has increased exponentially, and transmission section congestion has changed from an occasional event to a regular risk. The current practice adopted by power dispatching agencies to handle the contradiction between inter-provincial medium- and long-term transaction volume and transmission section congestion is to have dispatchers perform static safety checks on D-2 using the Power System Analysis and Synthesis Program (PSASP) or the large-scale power grid steady-state and dynamic simulation analysis software (BPA). After converting the medium- and long-term contract volume according to the peak-valley coefficient, an offline power flow calculation is performed. If the check fails, the contract volume can only be manually reduced. In 2023, the deviation assessment cost of a certain power grid reached 740 million yuan.

[0003] The fundamental flaw in this static model lies in treating the four factors—weather, power, electricity price, and power flow—as three separate, independent processes. Specifically: (1) The meteorological big model is only responsible for outputting weather data, and only provides wind speed data or light data.

[0004] (2) Power forecasting and electricity price forecasting each use different models to generate curves, and the existing forecasting models are mostly single models, such as traditional statistical models or simple machine learning models. These models have obvious limitations in dealing with complex nonlinear relationships and multivariate coupling problems, and cannot effectively capture the complex relationship between meteorological factors and power and electricity prices. In addition, the generalization ability of existing models is insufficient, and models that have not undergone sufficient cross-validation are prone to overfitting or underfitting when faced with new data or environmental changes. At the same time, existing methods are slow to respond when dealing with real-time data updates and dynamic environmental changes. For example, rapid changes in meteorological conditions or adjustments to electricity market rules may cause existing models to fail.

[0005] (3) During the scheduling phase, an offline power flow check is performed based on the static load, and the check program only recognizes the fixed load curve.

[0006] Therefore, due to the lack of a long-term coupling mechanism of "electricity price-power-power flow", the sensitivity matrix is ​​fixed in the long term, there is no closed-loop control in the medium and long term, it is impossible to translate meteorological disturbances into the probability of exceeding the limit of the transmission section in advance, and it is even more impossible to start automated preventive control in the prediction stage. Summary of the Invention

[0007] This invention proposes a method and device for assessing the risk of power flow exceeding the limit of a transmission section based on a meteorological big model. It embeds real-time assessment of the risk of power flow exceeding the limit and closed-loop control into the medium- and long-term electricity price forecast link, maps the weather disturbances output by the meteorological big model into the probability of power flow exceeding the limit of the transmission section in real time, and provides closed-loop control instructions that can be executed automatically.

[0008] To achieve the above objectives, the technical solution of the present invention is implemented as follows: A method for assessing power flow exceedance risk at transmission sections based on a large meteorological model includes: S1. Based on future meteorological data predicted by a large meteorological model, output the regional power forecast for the medium and long term through a power prediction model. The power prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; S2. Based on the predicted regional power and the future meteorological data, output the predicted regional electricity price for the medium and long term through the electricity price prediction model. The electricity price prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; S3. Obtain power grid topology data and the limit capacity (INTC) of each transmission section over a certain future time span, and construct a price-power flow sample tensor with time period t as the time granularity. , ,in The node admittance matrix is ​​obtained from power grid topology data. S4. Using linearized DC power flow formula ;in This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector; The vector represents the deviation between the active power flow of each transmission section and the reference historical average in the future time period t; S is the sensitivity matrix, which is obtained by fitting the price-power flow sample tensor of the historical time period. S5. Based on the linearized DC power flow formula, calculate the predicted power flow for the future time period t for each transmission section, and calculate the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

[0009] Furthermore, the TimesNet-TCN-CBAM hybrid model framework includes cascaded TimesNet modules, TCN modules, and CBAM modules; The TimesNet module receives the input time series data and processes it sequentially through multiple TimesBlock layers: each TimesBlock identifies the main period of the sequence through Fast Fourier Transform, uses Softmax to adaptively aggregate different period weights, reshapes the one-dimensional sequence of the time series data into a two-dimensional tensor, extracts period features through two-dimensional convolution, and then outputs enhanced time series features through residual connection. The TCN module receives the enhanced temporal features, employs multi-layer dilated causal convolution, and combines residual connections to capture long sequence dependencies, outputting long temporal features; The CBAM module sequentially uses the channel attention submodule and the spatial attention submodule to generate attention weights for the long-term features in the channel and spatial dimensions, performs adaptive weighting and refinement, and finally outputs the prediction result.

[0010] Furthermore, step S1 includes: Historical meteorological and power data for the region are acquired and cleaned and preprocessed. By using historical meteorological data and historical power data, a power prediction model based on the TimesNet-TCN-CBAM hybrid model framework is trained to establish a nonlinear mapping relationship between meteorology and power. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The future meteorological data is input into the power prediction model, which outputs the regional power prediction for the medium and long term.

[0011] Furthermore, step S2 includes: Acquire historical meteorological data, historical power data, and historical electricity price data for the region, and perform cleaning and preprocessing. Based on historical meteorological data, historical power data, and historical electricity price data, an electricity price prediction model built on the TimesNet-TCN-CBAM hybrid model framework was trained to establish a nonlinear mapping relationship between electricity price and meteorological and power data. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The regional power forecast for the medium and long term is obtained through the power prediction model. The future meteorological data and the regional predicted power are input into the electricity price prediction model, and the regional predicted electricity price for the medium and long term is output.

[0012] Furthermore, in step S3, CIM / XML files for a certain future time span are obtained in real time from the energy management system of the power grid dispatch center, power grid topology data is exported from the CIM / XML files, and node admittance matrix is ​​obtained based on the power grid topology data.

[0013] Furthermore, in step S4, the fitting of the sensitivity matrix S includes: Construct a price-power flow sample tensor for historical time periods. Based on the regional predicted power, nodal admittance matrix, and ultimate capacity of each transmission section, select valid samples. Using time period t as the time granularity, take paired samples from all time periods. , }; This represents the vector of deviations between predicted and actual electricity prices in the regional electricity price-power flow sample tensor for historical periods. The measured deviation vector between the active power flow of each transmission section and the reference historical average during a historical period; The sensitivity matrix S is fitted using least squares or ridge regression: .

[0014] Furthermore, the sensitivity matrix S is updated hourly with new samples, and if the power grid topology undergoes N-1 maintenance, it is triggered to recalculate immediately.

[0015] Furthermore, in step S5, calculating the predicted power flow for the future time period t includes: ;in This represents the power flow of transmission section ij under the measured or reference operating mode at the current moment; This represents the sensitivity matrix of the transmission section ij. This represents the predicted power flow of transmission section ij in the future time period t; transmission section ij refers to the transmission section between grid node i and node j.

[0016] Furthermore, in step S5, the prevention and control strategy includes: Level 1 second-level control: Sends reactive power regulation to the SVC / STATCOM at the beginning of the current transmission section. , ;in This represents the reactive power regulation gain coefficient; This indicates the current measured or predicted active power flow at the transmission section; V represents the equivalent line reactance of the current transmission section; V represents the bus voltage at the beginning of the current transmission section. This indicates that the allowed power flow is set to 95% of the maximum capacity; Level 2 minute-level control: After Level 1 second-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, the output of the sending or receiving unit is adjusted via AGC rescheduling command. The objective function is: ;in This represents the probability of power flow exceeding the limit at the transmission section of the power grid node i and node j, which is the probability of power flow exceeding the limit at the current transmission section. Level 3 hourly control: After Level 2 minute-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, a power reduction / transfer suggestion will be pushed to the medium- and long-term trading order system to automatically reduce the power of medium- and long-term contracts for the next day. , where parameters It is determined by the amount of risk.

[0017] Another aspect of this invention proposes a power flow over-limit risk assessment device for transmission sections based on a large meteorological model, comprising: Power Prediction Module: Based on future meteorological data predicted by a large meteorological model, this module outputs medium- to long-term regional power forecasts through a power prediction model. The power prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Electricity price forecasting module: Based on the predicted regional power and the future meteorological data, it outputs the predicted regional electricity price for the medium and long term through an electricity price forecasting model. The electricity price prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Sample Tensor Module: Acquires power grid topology data over a certain future time span, as well as the limit capacity (INTC) of each transmission section, and constructs a price-power flow sample tensor with time interval t as the time granularity. , ,in The node admittance matrix is ​​obtained from power grid topology data. Linearized DC power flow module: Employs linearized DC power flow formula ;in This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector; The vector represents the deviation between the active power flow of each transmission section and the reference historical average in the future time period t; S is the sensitivity matrix, which is obtained by fitting the price-power flow sample tensor of the historical time period. Over-limit calculation and control module: Based on the linearized DC power flow formula, it calculates the predicted power flow for the future time period t for each transmission section and calculates the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

[0018] Compared with the prior art, the present invention has the following beneficial effects: This invention proposes an integrated modeling method for medium- and long-term electricity price, power, and cross-sectional power flow, which for the first time seamlessly extends medium- and long-term power-price forecasting to real-time closed-loop control of transmission cross-sectional power flow exceedance risk. By coupling the future (e.g., 24-168h) electricity price-power curves output by the meteorological big data model and the TimesNet-TCN-CBAM hybrid model with the real-time topology of the power grid and the cross-sectional limit capacity, a long-term quantitative assessment of the probability of transmission cross-sectional power flow exceedance is achieved.

[0019] Based on the DC power flow linearization formula and the normal error assumption, this invention uses the dynamic sensitivity matrix obtained by rolling regression to automatically calculate the power flow over-limit probability of the transmission section every hour in the future (e.g., 1-7 days), and triggers multi-level control when the power flow over-limit probability exceeds the threshold.

[0020] The multi-level control proposed in this invention includes second-level SVC / STATCOM reactive power regulation commands, minute-level AGC unit rescheduling commands, and hour-level automatic reduction suggestions for medium- and long-term contract power volumes; it forms a closed loop with the power flow over-limit prediction scheme of this invention, and preventive control is initiated in the prediction stage.

[0021] This invention fills the gap in the prediction and automatic control of power flow over-limit risks in medium- and long-term power transmission sections, and has significant technological advancements and broad economic and social benefits. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the embodiments will be briefly described below. The drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a schematic diagram of the TimeNet-TCN-CBAM hybrid model framework structure of the present invention; Figure 2 This is a schematic diagram of the method flow of Embodiment 1 of the present invention; Figure 3 This is a flowchart of the provincial power forecast and provincial electricity price forecast in Embodiment 1 of the present invention; Figure 4 This is the risk assessment and closed-loop control process for power transmission section exceeding limits in Embodiment 1 of the present invention. Detailed Implementation

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0025] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0026] Example 1: The power prediction model and electricity price prediction model involved in this embodiment are both built based on the TimesNet-TCN-CBAM hybrid model framework. The TimesNet-TCN-CBAM hybrid model framework will be explained first below.

[0027] The TimesNet-TCN-CBAM hybrid model framework, as a deep learning prediction model, comprises cascaded TimesNet, TCN, and CBAM modules. By integrating TimesNet's ability to extract periodic features, TCN's ability to capture long-sequence dependencies, and CBAM's (Convolutional Block Attention Module) attention enhancement mechanism, it achieves a fusion of multi-timescale feature extraction, enhanced feature representation, and efficient computation. The TimesNet-TCN-CBAM hybrid model framework optimizes the extraction of periodic features from time series through the TimesNet module's adaptive periodicity recognition and feature transformation. Simultaneously, it utilizes TCN to process long-term temporal information, and the CBAM module enhances the channel and spatial representation of features. Supporting end-to-end training, it is suitable for high-frequency, large-scale time series prediction tasks, exhibiting significant generalization ability and flexibility. This model design not only improves the accuracy of time series prediction but also enhances the model's adaptability to changes at different time scales, making it perform excellently in various time series analysis tasks. Figure 1 The image shows the framework structure of the TimesNet-TCN-CBAM hybrid model, explained in detail below: (1) TimesNet module: The TimesNet module is a deep learning model specifically designed for time series forecasting. It effectively addresses the periodicity and long-term dependencies in time series data by combining periodic feature extraction and residual connections. The core idea of ​​TimesNet is to transform a one-dimensional time series into a two-dimensional tensor and then use two-dimensional convolutions to capture these periodic variations. This transformation allows the model to simultaneously handle intra-period variations (column direction) and inter-period variations (row direction), thus providing a more comprehensive understanding of the dynamic characteristics of the time series. Period estimation is a crucial step in TimesNet. The model first estimates the main period of the time series using Fast Fourier Transform (FFT) or other period detection methods. Then, based on the estimated period length, the one-dimensional time series is reshaped into a two-dimensional tensor. This reshaping not only helps capture periodic features but also reduces the number of model parameters and computational complexity. The reshaped two-dimensional tensor is then fed into two-dimensional convolutional layers. These convolutional layers can be standard convolutional layers or more parameter-efficient Inception blocks or other variants. Two-dimensional convolutional layers can simultaneously capture intra-period variations and inter-period variations, thus extracting time series features more effectively. After features from multiple periods are extracted, TimesNet uses an adaptive aggregation mechanism to weight and combine these features. This typically involves using the Softmax function to calculate the weights for different periods, and then summing the features using weighted methods. This adaptive aggregation approach allows the model to dynamically adjust the weights based on the actual periodicity of the time series, thereby improving the model's prediction accuracy.

[0028] The TimesNet module receives input time-series data and processes it sequentially through multiple TimesBlock layers. Each TimesBlock in the TimesNet module contains residual connections. This design can alleviate the vanishing or exploding gradient problem in deep networks, while enhancing the network's learning ability.

[0029] The TimesNet module receives input time-series data and processes it sequentially through multiple layers of TimesBlock. For example... Figure 1 The TimesBlock structure is illustrated in the diagram. Each TimesBlock identifies the main period of the time series data sequence through Fast Fourier Transform, then changes its shape and reshapes the one-dimensional time series sequence into a two-dimensional tensor based on the period length, splitting the dimensions of change within and between periods. Relying on efficient initial blocks (lightweight convolutional units), it mines the basic time series features under the two-dimensional structure with low computational overhead. Then, it restores the shape, converting the two-dimensional feature tensor back into a one-dimensional data form. At the same time, it uses Softmax to adaptively aggregate the weights of different periods, and then outputs enhanced time series features through residual connections.

[0030] After convolutional layers, ReLU activation function is typically applied to introduce non-linearity, and Dropout layers are used to prevent overfitting. These techniques help improve the model's generalization ability and prediction accuracy.

[0031] The TimesNet module effectively addresses the periodicity and long-term dependencies in time series data through its unique periodic feature extraction and residual connection techniques. Its design not only improves the model's predictive accuracy but also maintains low computational complexity and parameter count, making it an ideal choice for processing complex time series data.

[0032] (2) Temporal Convolutional Neural Network (TCN) module: TCN is a unique neural network architecture primarily used for time series modeling, consisting of multiple layers of extended causal convolutional structures and residual connections. The core difference between its extended causal convolution and traditional convolution lies in the fact that, as the convolutional kernel slides, it allows sampling of input data at specific intervals. This interval is called the dilation factor, which typically increases exponentially with the number of network layers. For example, when the dilation factor d=1, the convolutional kernel samples all data points consecutively; while when d=2, the kernel samples only every second data point. In this way, extended causal convolution can effectively expand the network's receptive field while maintaining low computational complexity and parameter count. Furthermore, the "causal" property ensures that when predicting the output at the current time point, the network can only rely on current and past data, completely eliminating interference from future information and thus avoiding data leakage. This design makes TCN suitable for handling tasks with a clear temporal order.

[0033] The architecture of TCN typically consists of multiple stacked residual blocks. Each residual block contains two layers of extended causal convolutions, each followed by a weight normalization, ReLU activation function, and dropout layer.

[0034] In this embodiment, the TCN module receives the enhanced temporal features, employs multi-layer dilated causal convolution, and combines residual connections to capture long-sequence dependencies, outputting long-sequence features. Overall, TCN, with its unique dilated causal convolution structure and residual connections, performs excellently in handling long sequences and complex temporal dependencies, while also being computationally efficient and requiring few parameters.

[0035] (3) Attention Mechanism Module (CBAM Module): CBAM is an attention mechanism module used to enhance the representation capabilities of convolutional neural networks. It consists of a Channel Attention Module (CAM) and a Spatial Attention Module (SAM). CBAM can sequentially generate attention maps in the channel and spatial dimensions, and combine them with the input feature map through element-wise multiplication to achieve adaptive feature refinement.

[0036] In time series data prediction, the CBAM module can effectively enhance the feature extraction capability of the network. The TimesNet-TCN-CBAM model utilizes TimesNet modules and LSTM to extract temporal features from the data. By adding the CBAM module to the network structure, its feature extraction capability can be effectively improved. Furthermore, CBAM has potential in handling nonlinear and complex patterns in data; by strengthening the model's focus on specific features, it helps reveal underlying patterns and regularities. In time series prediction tasks, CBAM can be used to enhance the sensitivity of LSTM networks to features. This LSTM network architecture design incorporating CBAM typically adds a CBAM module after the LSTM layers to achieve channel-level weighting of features at each time point in the time series. This design allows the network to dynamically adjust its attention during sequence analysis, strengthening important time point information and ignoring unimportant or noisy points. The key to the CBAM mechanism's improvement in time series prediction accuracy lies in providing the model with dynamically weighted feature representations. This mechanism allows the model to more flexibly adjust its focus on different time points when processing sequence data.

[0037] In this embodiment, the CBAM module sequentially uses the channel attention submodule and the spatial attention submodule to generate attention weights for the long-term features in the channel and spatial dimensions, performs adaptive weighted refinement, and finally outputs the prediction result (e.g., the regional predicted power or regional predicted electricity price in this embodiment).

[0038] Based on the above description of the TimesNet-TCN-CBAM hybrid model framework, this embodiment proposes a method for assessing power flow exceedance risks at transmission sections based on a large meteorological model, such as... Figure 2 As shown, it includes: S1. Based on the future meteorological data predicted by the meteorological big data model, the power prediction model outputs the regional power prediction for the medium and long term. The power prediction model is constructed based on the TimesNet-TCN-CBAM hybrid model framework.

[0039] In this embodiment, the power forecast for the medium to long term is performed at the provincial level. This step specifically includes: Acquire and integrate historical meteorological and historical power data for the region, including historical meteorological elements of all cities and individual stations in the province, as well as historical power data for the entire province; Cleaning and preprocessing of historical meteorological and power data, including missing value imputation, outlier detection, and data standardization; By using historical meteorological data and historical provincial power data, a power prediction model based on the TimesNet-TCN-CBAM hybrid model framework is trained to minimize prediction error, establish a nonlinear mapping relationship between meteorology and power, and evaluate the model's generalization ability through cross-validation to ensure prediction accuracy.

[0040] The future meteorological elements of various cities and individual stations are forecasted using a large meteorological model, and the future meteorological data of various cities and individual stations are obtained; for example, this step uses the Kuangming large meteorological model.

[0041] The future meteorological data of various cities and individual stations predicted by the Kuangming meteorological big data model are input into the power prediction model, and the predicted power of the whole province in the medium and long term is output.

[0042] S2. Based on the predicted regional power and the future meteorological data, output the predicted regional electricity price for the medium and long term through the electricity price prediction model, which is constructed based on the TimesNet-TCN-CBAM hybrid model framework.

[0043] This step specifically includes: Acquire and integrate historical meteorological data from all cities and individual stations across the province, as well as historical power data and historical electricity price data for the entire province; Cleaning and preprocessing of historical meteorological data, historical power data, and historical electricity price data, including missing value imputation, outlier detection, and data standardization; The electricity price prediction model, based on the TimesNet-TCN-CBAM hybrid model framework, was trained using historical electricity price information, historical meteorological data, and historical provincial power data. The model was also cross-validated to ensure accuracy.

[0044] The Kuangming meteorological big data model is used to forecast the future meteorological data of various cities and individual stations, and the future meteorological data of various cities and individual stations are obtained. The future meteorological data of individual stations in various cities and prefectures predicted by the Kuangming meteorological big data model, as well as the future medium- and long-term predicted power of the whole province output by the power prediction model, are all input into the electricity price prediction model, and the future medium- and long-term predicted electricity price of the whole province is output.

[0045] Figure 3 The diagram shows a flowchart of the medium- to long-term power forecast and electricity price forecast for the entire province as described in steps S1 and S2 above.

[0046] S3. Obtain power grid topology data and the limit capacity INTC of each transmission section over a certain future time span, and construct the electricity price-power flow sample tensor with time period t as the time granularity.

[0047] Based on the medium- and long-term predicted power of the entire province obtained in step S1 And the projected medium- to long-term provincial electricity prices obtained in step S2 Then, perform secondary data fusion.

[0048] The power grid topology data (CIM / XML) for the next 24-168 hours (where h represents hours) and the maximum capacity (INTC) of each section are obtained in real time from the provincial EMS; the electricity price is predicted using a 1-hour time granularity. Predicted power Concatenated with topology parameters to form the electricity price-power flow sample tensor .

[0049] in This represents the node admittance matrix, or simply the admittance matrix. Its function is to package the electrical transport capacity between all buses (nodes) in the power grid into a single square matrix.

[0050] In this embodiment, Based on CIM / XML data obtained from the provincial EMS, the dimension = number of nodes × number of nodes, generally stored sparsely. Integrating Electricity Prices - Trend Sample Tensor The purpose is to explain "what the current grid structure looks like, which lines are strong, and which are weak", so that sensitivity calculation and over-limit probability assessment are based on the real-time grid structure, rather than a fixed typical topology.

[0051] S4. A linearized DC power flow formula is adopted.

[0052] The linearized DC power flow formula is: ; in: This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector has a dimension of N×1 (N is the number of electricity price nodes).

[0053] The vector represents the deviation between the active power flow of each transmission section in the future time period t and the reference historical average, with a dimension of M×1 (M is the number of transmission sections).

[0054] S is the sensitivity matrix, obtained by fitting the historical electricity price-power flow sample tensor. The historical electricity price-power flow sample tensor is the training set that implicitly drives sensitivity updates. The method for obtaining S includes: concatenating the electricity price-power flow sample tensor for the past 90 days using the method described in step S3. The concatenated data serves as the input for updating the sensitivity matrix S. This describes the current power grid structure; the INTC vector indicates the limits of each transmission section. INTC is used to filter valid samples, retaining only nodes that are electrically close to the target transmission section. This describes the range of electricity prices and power levels for the next hour. It utilizes the electricity price-power flow sample tensor from the past 90 days. Take paired samples at 1 point every hour { , The sensitivity matrix S is obtained by fitting the data using ordinary least squares (or ridge regression). The resulting sensitivity matrix S has N×M dimensions, and the elements in the matrix represent how many MW the cross-sectional power flow increases for every 1 yuan / MWh increase in the nodal electricity price.

[0055] After the calculation is completed, S is updated hourly with new samples to ensure that the mapping relationship between medium- and long-term electricity price fluctuations and cross-sectional power flow remains consistent with the current grid structure. The operation mode (INTC) remains consistent. If an N-1 maintenance occurs in the power grid topology, an immediate recalculation is triggered to ensure that S is consistent with the real-time network.

[0056] S5. Based on the linearized DC power flow formula, calculate the predicted power flow for the future time period t for each transmission section, and calculate the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

[0057] For any transmission section ij, calculate the predicted power flow for the future time period t: The aforementioned transmission section ij refers to the transmission section connecting power grid node i and node j, i.e., the set of lines connecting power grid node i and node j. This can be a single circuit, multiple parallel circuits, or a series branch network. In calculations, this group of lines is usually combined into an equivalent branch network, using the total ultimate capacity. As its limit-crossing criterion, the sensitivity of section ij and predicting trends These are all equivalent quantities for this set of lines.

[0058] ; in: This represents the power flow (MW) of transmission section ij under the measured or reference operating mode at the current moment, serving as the starting point.

[0059] Assume power flow prediction error That is, assuming that the power flow prediction error follows a mean of 0 and a variance of 0. It follows a normal distribution. The variance is estimated online from the sample variance of the "predicted value - measured value" of the power flow at the transmission section over the past 90 days, and updated hourly for calculating the probability of power flow exceeding limits. For transmission section ij, the variance is... .

[0060] Power flow exceedance probability of transmission section ij : ; in It is the standard normal cumulative distribution function.

[0061] when When the threshold of 5% is adjustable, a risk event is triggered, and the transmission section ij should implement a prevention and control strategy.

[0062] The generation and implementation of the prevention and control strategy include: (1) Level 1 control (second level): Send reactive power regulation ΔQ to the SVC / STATCOM at the beginning of the transmission section. The instruction format follows the IEC61850 GOOSE message. ; in: It is the reactive power regulation gain coefficient (dimensionless per-unit value), which is used to map the active power over-limit deviation into the reactive power that needs to be compensated. Here it is taken as 0.8 (per-unit value).

[0063] : The reactive power regulation amount (unit: Mvar, positive value is reactive power generation, negative value is reactive power absorption) that needs to be sent to the SVC / STATCOM at the head end of the current transmission section (i.e., the transmission section ij that triggered the risk event marker).

[0064] : Current measured or predicted active power flow (MW) at the transmission section.

[0065] Set the allowable power flow to 95% of the maximum capacity, leaving a 5% margin.

[0066] Equivalent line reactance (Ω) of the current transmission section.

[0067] V: Current bus voltage at the beginning of the transmission section (kV), taken as the rated value or real-time PMU value.

[0068] This formula uses the voltage displacement generated by reactive power compensation to indirectly suppress the active power flow, thereby pulling the over-limit portion back to the safe area.

[0069] (2) Secondary control (minute level): After the primary control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, the output of the sending or receiving unit is adjusted through the AGC rescheduling instruction. The objective function is ; (3) Level 3 control (hourly): After Level 2 control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, a power reduction / transfer suggestion will be pushed to the medium and long-term trading order system to automatically reduce the power of medium and long-term contracts for the next day. , ,parameter The percentage of contracted electricity that needs to be reduced is determined by the risk level. Based on the probability of power flow exceeding limits, the percentage of contracted electricity that needs to be reduced is automatically calculated according to a preset risk-reduction mapping curve. This will allow the medium- and long-term contract electricity volume for the following day to be... The level has been lowered to a safe level.

[0070] For example, the specific approach is as follows: Real-time reading Its value is a decimal between 0 and 1; Using piecewise linear functions: ; ; That is: the probability of trend exceeding the limit is not reduced if it is below 5%; it increases linearly between 5% and 20%; and it is reduced to 100% (full transfer) above 20%.

[0071] Calculated Then, the system automatically sends a JSON message to the medium- and long-term transaction order server: The reason code "cross-section blockage risk" is attached.

[0072] The proposed power reduction / transfer suggestions enable hourly closed-loop operation without human intervention.

[0073] This embodiment also includes a feature that compares the predicted and measured power flow values ​​at the transmission section every 15 minutes to update the data. And rolling correction ; If the root mean square error (RMSE) of two consecutive predictions within 30 minutes exceeds 3%, then the error samples are sent back to the TimesNet-TCN-CBAM electricity price prediction module to trigger online incremental learning and achieve model adaptation.

[0074] At the same time, the timestamp of each over-limit risk event, the name of the transmission section, The control actions and actual results are written into the blockchain evidence storage subsystem to ensure that scheduling decisions are traceable.

[0075] like Figure 4 The diagram shows the risk assessment and closed-loop control process for transmission section exceeding limits, covered by steps S3-S5.

[0076] This embodiment directly transforms medium- and long-term electricity price-power forecast results into quantitative assessments and real-time control commands for cross-sectional power flow over-limit risks, forming an integrated technical solution from market signals to physical grid security, significantly reducing the risks of transmission congestion and power outages caused by electricity price fluctuations.

[0077] Example 2: This embodiment proposes a power flow over-limit risk assessment device based on a large meteorological model, comprising: Power Prediction Module: Based on future meteorological data predicted by a large meteorological model, this module outputs medium- to long-term regional power forecasts through a power prediction model. The power prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Electricity price forecasting module: Based on the predicted regional power and the future meteorological data, it outputs the predicted regional electricity price for the medium and long term through an electricity price forecasting model. The electricity price prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Sample Tensor Module: Acquires power grid topology data over a certain future time span, as well as the limit capacity (INTC) of each transmission section, and constructs a price-power flow sample tensor with time interval t as the time granularity. , ,in The node admittance matrix is ​​obtained from power grid topology data. Linearized DC power flow module: Employs linearized DC power flow formula ;in This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector; The vector represents the deviation between the active power flow of each transmission section and the reference historical average in the future time period t; S is the sensitivity matrix, which is obtained by fitting the price-power flow sample tensor of the historical time period. Over-limit calculation and control module: Based on the linearized DC power flow formula, it calculates the predicted power flow for the future time period t for each transmission section and calculates the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

[0078] The TimesNet-TCN-CBAM hybrid model framework includes cascaded TimesNet, TCN, and CBAM modules. The TimesNet module receives the input time series data and processes it sequentially through multiple TimesBlock layers: each TimesBlock identifies the main period of the sequence through Fast Fourier Transform, uses Softmax to adaptively aggregate different period weights, reshapes the one-dimensional sequence of the time series data into a two-dimensional tensor, extracts period features through two-dimensional convolution, and then outputs enhanced time series features through residual connection. The TCN module receives the enhanced temporal features, employs multi-layer dilated causal convolution, and combines residual connections to capture long sequence dependencies, outputting long temporal features; The CBAM module sequentially uses the channel attention submodule and the spatial attention submodule to generate attention weights for the long-term features in the channel and spatial dimensions, performs adaptive weighting and refinement, and finally outputs the prediction result.

[0079] The power prediction module includes: Historical meteorological and power data for the region are acquired and cleaned and preprocessed. By using historical meteorological data and historical power data, a power prediction model based on the TimesNet-TCN-CBAM hybrid model framework is trained to establish a nonlinear mapping relationship between meteorology and power. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The future meteorological data is input into the power prediction model, which outputs the regional power prediction for the medium and long term.

[0080] The electricity price forecasting module includes: Acquire historical meteorological data, historical power data, and historical electricity price data for the region, and perform cleaning and preprocessing. Based on historical meteorological data, historical power data, and historical electricity price data, an electricity price prediction model built on the TimesNet-TCN-CBAM hybrid model framework was trained to establish a nonlinear mapping relationship between electricity price and meteorological and power data. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The regional power forecast for the medium and long term is obtained through the power prediction model. The future meteorological data and the regional predicted power are input into the electricity price prediction model, and the regional predicted electricity price for the medium and long term is output.

[0081] In the sample tensor module, CIM / XML files for a certain future time span are obtained in real time from the power grid dispatch center's energy management system. Power grid topology data is exported from the CIM / XML files, and the node admittance matrix is ​​obtained based on the power grid topology data.

[0082] In the linearized DC power flow module, the fitting of the sensitivity matrix S includes: Construct a price-power flow sample tensor for historical time periods. Based on the regional predicted power, nodal admittance matrix, and ultimate capacity of each transmission section, select valid samples. Using time period t as the time granularity, take paired samples from all time periods. , }; This represents the vector of deviations between predicted and actual electricity prices in the regional electricity price-power flow sample tensor for historical periods. The measured deviation vector between the active power flow of each transmission section and the reference historical average during a historical period; The sensitivity matrix S is fitted using least squares or ridge regression: .

[0083] The sensitivity matrix S is updated hourly with new samples. If the power grid topology undergoes N-1 maintenance, it is triggered to recalculate immediately.

[0084] In the over-limit calculation and control module, the calculation of the predicted power flow for the future time period t includes: ;in This represents the power flow of transmission section ij under the measured or reference operating mode at the current moment; This represents the sensitivity matrix of the transmission section ij. This represents the predicted power flow of transmission section ij in the future time period t; transmission section ij refers to the transmission section between grid node i and node j.

[0085] The prevention and control strategies include: Level 1 second-level control: Sends reactive power regulation to the SVC / STATCOM at the beginning of the current transmission section. , ;in This represents the reactive power regulation gain coefficient; This indicates the current measured or predicted active power flow at the transmission section; V represents the equivalent line reactance of the current transmission section; V represents the bus voltage at the beginning of the current transmission section. This indicates that the allowed power flow is set to 95% of the maximum capacity; Level 2 minute-level control: After Level 1 second-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, the output of the sending or receiving unit is adjusted via AGC rescheduling command. The objective function is: ;in This represents the probability of power flow exceeding the limit at the transmission section of the power grid node i and node j, which is the probability of power flow exceeding the limit at the current transmission section. Level 3 hourly control: After Level 2 minute-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, a power reduction / transfer suggestion will be pushed to the medium- and long-term trading order system to automatically reduce the power of medium- and long-term contracts for the next day. , where parameters It is determined by the amount of risk.

[0086] The power flow overload risk assessment device based on a meteorological large model proposed in this embodiment can achieve the same technical effect as the power flow overload risk assessment method based on a meteorological large model described in Embodiment 1.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model, characterized in that, include: S1. Based on future meteorological data predicted by a large meteorological model, output the regional power forecast for the medium and long term through a power prediction model. The power prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; S2. Based on the predicted regional power and the future meteorological data, output the predicted regional electricity price for the medium and long term through the electricity price prediction model. The electricity price prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; S3. Obtain power grid topology data and the limit capacity (INTC) of each transmission section over a certain future time span, and construct a price-power flow sample tensor with time period t as the time granularity. , ,in The node admittance matrix is ​​obtained from power grid topology data. S4. Using linearized DC power flow formula ;in This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector; The vector represents the deviation between the active power flow of each transmission section and the reference historical average in the future time period t; S is the sensitivity matrix, which is obtained by fitting the price-power flow sample tensor of the historical time period. S5. Based on the linearized DC power flow formula, calculate the predicted power flow for the future time period t for each transmission section, and calculate the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

2. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 1, characterized in that, The TimesNet-TCN-CBAM hybrid model framework includes cascaded TimesNet, TCN, and CBAM modules; The TimesNet module receives the input time series data and processes it sequentially through multiple TimesBlock layers: each TimesBlock identifies the main period of the sequence through Fast Fourier Transform, uses Softmax to adaptively aggregate different period weights, reshapes the one-dimensional sequence of the time series data into a two-dimensional tensor, extracts period features through two-dimensional convolution, and then outputs enhanced time series features through residual connection. The TCN module receives the enhanced temporal features, employs multi-layer dilated causal convolution, and combines residual connections to capture long sequence dependencies, outputting long temporal features; The CBAM module sequentially uses the channel attention submodule and the spatial attention submodule to generate attention weights for the long-term features in the channel and spatial dimensions, performs adaptive weighting and refinement, and finally outputs the prediction result.

3. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 2, characterized in that, Step S1 includes: Historical meteorological and power data for the region are acquired and cleaned and preprocessed. By using historical meteorological data and historical power data, a power prediction model based on the TimesNet-TCN-CBAM hybrid model framework is trained to establish a nonlinear mapping relationship between meteorology and power. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The future meteorological data is input into the power prediction model, which outputs the regional power prediction for the medium and long term.

4. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 2, characterized in that, Step S2 includes: Acquire historical meteorological data, historical power data, and historical electricity price data for the region, and perform cleaning and preprocessing. Based on historical meteorological data, historical power data, and historical electricity price data, an electricity price prediction model built on the TimesNet-TCN-CBAM hybrid model framework was trained to establish a nonlinear mapping relationship between electricity price and meteorological and power data. By using large meteorological models, we can forecast the medium- and long-term meteorological data of the region and obtain the future meteorological data of the region. The regional power forecast for the medium and long term is obtained through the power prediction model. The future meteorological data and the regional predicted power are input into the electricity price prediction model, and the regional predicted electricity price for the medium and long term is output.

5. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 1, characterized in that, In step S3, CIM / XML files for a certain future time span are obtained in real time from the energy management system of the power grid dispatch center. Power grid topology data is exported from the CIM / XML files, and the node admittance matrix is ​​obtained based on the power grid topology data.

6. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 1, characterized in that, In step S4, the fitting of the sensitivity matrix S includes: Construct a price-power flow sample tensor for historical time periods. Based on the regional predicted power, nodal admittance matrix, and ultimate capacity of each transmission section, select valid samples. Using time period t as the time granularity, take paired samples from all time periods. , }; This represents the vector of deviations between predicted and actual electricity prices in the regional electricity price-power flow sample tensor for historical periods. The measured deviation vector between the active power flow of each transmission section and the reference historical average during a historical period; The sensitivity matrix S is fitted using least squares or ridge regression: .

7. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 6, characterized in that, The sensitivity matrix S is updated hourly with new samples. If the power grid topology undergoes N-1 maintenance, it is triggered to recalculate immediately.

8. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 1, characterized in that, In step S5, calculating the predicted power flow for the future time period t includes: ;in This represents the power flow of transmission section ij under the measured or reference operating mode at the current moment; This represents the sensitivity matrix of the transmission section ij. This represents the predicted power flow of transmission section ij in the future time period t; transmission section ij refers to the transmission section between grid node i and node j.

9. The method for assessing the risk of power flow exceeding limits at transmission sections based on a large meteorological model according to claim 1, characterized in that, In step S5, the prevention and control strategy includes: Level 1 second-level control: Sends reactive power regulation to the SVC / STATCOM at the beginning of the current transmission section. , ;in This represents the reactive power regulation gain coefficient; This indicates the current measured or predicted active power flow at the transmission section; V represents the equivalent line reactance of the current transmission section; V represents the bus voltage at the beginning of the current transmission section. This indicates that the allowed power flow is set to 95% of the maximum capacity; Level 2 minute-level control: After Level 1 second-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, the output of the sending or receiving unit is adjusted via AGC rescheduling command. The objective function is: ;in This represents the probability of power flow exceeding the limit at the transmission section of the power grid node i and node j, which is the probability of power flow exceeding the limit at the current transmission section. Level 3 hourly control: After Level 2 minute-level control takes effect, if the probability of power flow exceeding the limit still exceeds the threshold, a power reduction / transfer suggestion will be pushed to the medium- and long-term trading order system to automatically reduce the power of medium- and long-term contracts for the next day. , where parameters It is determined by the amount of risk.

10. A power flow over-limit risk assessment device for transmission sections based on a large meteorological model, characterized in that, include: Power Prediction Module: Based on future meteorological data predicted by a large meteorological model, this module outputs medium- to long-term regional power forecasts through a power prediction model. The power prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Electricity price forecasting module: Based on the predicted regional power and the future meteorological data, it outputs the predicted regional electricity price for the medium and long term through an electricity price forecasting model. The electricity price prediction model is built on the TimesNet-TCN-CBAM hybrid model framework; Sample Tensor Module: Acquires power grid topology data over a certain future time span, as well as the limit capacity (INTC) of each transmission section, and constructs a price-power flow sample tensor with time interval t as the time granularity. , ,in The node admittance matrix is ​​obtained from power grid topology data. Linearized DC power flow module: Employs linearized DC power flow formula ;in This represents the regional electricity price forecast for future time period t. Compared with the historical average electricity price The deviation vector; The vector represents the deviation between the active power flow of each transmission section and the reference historical average in the future time period t; S is the sensitivity matrix, which is obtained by fitting the price-power flow sample tensor of the historical time period. Over-limit calculation and control module: Based on the linearized DC power flow formula, it calculates the predicted power flow for the future time period t for each transmission section and calculates the probability of power flow exceeding the limit. When the probability of power flow exceeding the limit exceeds the threshold, the prevention and control strategy is triggered.

Images