Power grid forward-looking risk assessment method and device, and medium

Abstract

The present invention relates to the technical field of power grid dispatching optimization. Disclosed are a power grid forward-looking risk assessment method and device, and a medium, solving the problems in the prior art of unstable wind speed, low fan efficiency, difficult power grid dispatching and poor safety of wind power generation. In the present invention, a target fan power time series is constructed on the basis of fan power data of a historical time period, then an extreme learning machine model, a long short-term memory neural network model and a temporal convolutional network model are trained, and then a short-term wind power prediction result is obtained, improving the accuracy and the flexibility of the short-term wind power prediction result. In an offline phase of a power grid, a generative adversarial network model is constructed on the basis of a power grid section data set, and at least one power grid forward-looking scenario series is generated; and in an online operation stage of the power grid, a power grid forward-looking operation scenario series is obtained on the basis of preset risk assessment indexes and current power grid operation scenario information, enhancing the accuracy and timeliness of power grid operation risk assessment.

Description

A method, apparatus and medium for forward-looking risk assessment of power grids Technical Field

[0001] This invention relates to the field of power grid dispatch optimization technology, and in particular to a method, device and medium for forward-looking risk assessment of power grids. Background Technology

[0002] In recent years, with the rapid development of the national grid economy and power grid construction, the power system, especially the UHV AC / DC projects, has been continuously put into operation, and the 220kV and above main grid has been continuously strengthened, resulting in an increasingly complex power grid structure. In addition, as a clean and renewable energy source, wind energy is of great significance for promoting energy structure adjustment and reducing environmental pollution. However, the large-scale grid connection of wind power increases the risk of grid operation. Therefore, it is necessary to improve the accuracy and reliability of wind power forecasting, and then perceive and assess the forward-looking operational risks of the power grid.

[0003] Currently, single prediction models are often used to forecast short-term wind power. However, these models often fail to account for the complex interactions between various factors, resulting in low accuracy and flexibility in short-term wind power predictions. Research methods and operational application technologies for power grid operation risk perception and identification primarily rely on risk indicators for cross-sectional assessment and contingency control. A power grid fault outage model has been constructed, and the risk transfer process is characterized through power flow tracing. Risk node sets are then optimized and solved by selecting cross-sectional risk indicators. However, under extreme weather conditions and the multiple risk coupling effects on the power grid, the accuracy and timeliness of power grid operation risk assessment need to be improved.

[0004] Therefore, there is an urgent need for a power grid operation risk assessment method, device, computer equipment, and storage medium to improve the accuracy of short-term wind power forecast results and enhance the accuracy and timeliness of power grid operation risk assessment. Summary of the Invention

[0005] To address the problems of unstable wind speeds, low turbine efficiency, difficulties in grid dispatching, and poor security in existing wind power technologies, this invention provides a method, device, and medium for forward-looking risk assessment of the power grid, which can enhance the accuracy and timeliness of power grid operational risk assessment. The specific technical solution is as follows:

[0006] A method for forward-looking risk assessment of power grids includes the following steps:

[0007] Acquire power grid operation status data, meteorological monitoring data, and historical wind turbine power data; construct a wind turbine power time series based on historical wind turbine power data;

[0008] The missing data in the wind turbine power time series is obtained, and the missing data is filled in by a piecewise linear regression model to obtain the target wind turbine power time series.

[0009] Based on the target wind turbine power time series, train an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model respectively.

[0010] A multi-objective prediction model is trained based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model.

[0011] The Pareto optimal solution set of the combined weight vector of the multi-objective prediction model is obtained by using the multi-objective Harris Eagle algorithm.

[0012] The deep deterministic strategy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model based on the Pareto optimal solution set.

[0013] Based on each model in the multi-objective prediction model and its corresponding dynamic optimal weights, the prediction results of wind power are obtained.

[0014] A power grid section dataset is generated based on the power grid operation status data, the meteorological monitoring data, and the wind power prediction results.

[0015] During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset.

[0016] The generative adversarial network model is used to obtain at least one type of power grid operation risk data;

[0017] Based on the at least one type of power grid operation risk data, a power grid operation risk scenario set is generated, wherein the power grid operation risk scenario set includes at least one power grid operation risk scenario.

[0018] Using a time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid forward-looking scenario sequence;

[0019] During the online operation phase of the power grid, pre-set risk assessment indicators and current power grid operation scenario information are obtained;

[0020] Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

[0021] Preferably, obtaining the missing data in the wind turbine power time series includes:

[0022] The wind turbine power data that meets the missing value preset condition in the wind turbine power time series is determined as the missing value data; the missing value preset condition includes that the value of the wind turbine power data is zero, and the change value of the wind turbine power data relative to the wind turbine power data of the previous time series is greater than a preset threshold.

[0023] Preferably, before imputing the missing data using a piecewise linear regression model, the method further includes:

[0024] The missing data is interpolated and filled to obtain the intermediate wind turbine power time series;

[0025] An initial piecewise linear regression model is trained based on the intermediate wind turbine power time series to optimize the parameters of the initial piecewise linear regression model until the trained piecewise linear regression model is obtained.

[0026] Preferably, training the initial piecewise linear regression model based on the intermediate wind turbine power time series includes:

[0027] Obtain the exponentially weighted moving average and moving median features of the intermediate wind turbine power time series;

[0028] The intermediate wind turbine power time series is processed according to the BiLSTM model to obtain the meta-features of the intermediate wind turbine power time series;

[0029] The initial piecewise linear regression model is trained based on the meta-features, the exponentially weighted moving average features, and the moving median features to optimize the parameters of the initial piecewise linear regression model until the trained piecewise linear regression model is obtained.

[0030] Preferably, based on the target wind turbine power time series, an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained respectively, including:

[0031] The CEEMDAN sequence decomposition method is used to decompose the historical wind power time series data to obtain multiple first subsequence data.

[0032] Using the multiple first subsequence data, an Extreme Learning Machine model, a Long Short-Term Memory Neural Network model, and a Temporal Convolutional Network model are trained simultaneously.

[0033] Preferably, the prediction results of short-term wind power are obtained based on each model and its corresponding dynamic optimal weight in the multi-objective prediction model, including:

[0034] Obtain time-series data of the target wind power;

[0035] The time series data of the target wind power is decomposed using the CEEMDAN sequence decomposition method to obtain multiple second subsequence data.

[0036] By using each model in the multi-objective prediction model, the multiple second sub-sequence data are simultaneously traversed and predicted to obtain the corresponding first prediction result, second prediction result and third prediction result;

[0037] Based on the corresponding dynamic optimal weights of each model, the first prediction result, the second prediction result, and the third prediction result are weighted and calculated to obtain the prediction result of short-term wind power.

[0038] Preferably, training a multi-objective prediction model based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model includes:

[0039] Obtain time-series data of the test wind power;

[0040] The time series data of the tested wind power was decomposed using the CEEMDAN sequence decomposition method to obtain multiple third subsequence data.

[0041] Based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model, the multiple third subsequence data are simultaneously traversed and tested to obtain the corresponding first test result, second test result, and third test result.

[0042] The mean square error and standard deviation of the actual short-term wind power are obtained between the first test result, the second test result and the third test result, respectively.

[0043] A multi-objective prediction model is trained using the minimum mean squared error and the minimum standard deviation as optimization objectives.

[0044] Preferably, the weights and biases of the multi-objective prediction model are iteratively updated using the action vectors of the agent, including:

[0045] Based on the principle of minimum average absolute percentage error, static weight coefficients are selected from the Pareto optimal solution set using the action vector of the agent.

[0046] The reference error is obtained based on the static weighting coefficient and the actual short-term wind power.

[0047] Analyze the correlation between the minimum mean square error, the minimum standard deviation, and the reference error;

[0048] Based on the correlation, obtain the reward function;

[0049] The weights and biases of the multi-objective prediction model are iteratively updated using the reward function.

[0050] A power grid operation risk assessment device, the device comprising:

[0051] The data acquisition module is used to acquire power grid operation status data and meteorological monitoring data;

[0052] The data integration module is used to generate a power grid section dataset based on the power grid operation status data and the meteorological monitoring data.

[0053] The modeling module is used to construct a generative adversarial network model based on the power grid cross-section dataset during the offline phase of the power grid; and to obtain at least one type of power grid operation risk data through the generative adversarial network model.

[0054] The processing module is configured to generate a set of power grid operation risk scenarios based on the at least one type of power grid operation risk data, wherein the set of power grid operation risk scenarios includes at least one power grid operation risk scenario.

[0055] The simulation and evaluation module is used to perform simulation calculations on at least one power grid operation risk scenario using time-series operation simulation methods, and generate at least one power grid forward-looking scenario sequence.

[0056] The data acquisition module is also used to acquire preset risk assessment indicators and current power grid operation scenario information during the online operation phase of the power grid.

[0057] The online application module is used to obtain the power grid operation risk assessment results based on the preset risk assessment indicators and the current power grid operation scenario information, and to match and obtain the corresponding power grid forward operation scenario sequence.

[0058] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the power grid forward risk assessment method as described above.

[0059] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0060] This invention constructs a wind turbine power time series based on historical wind turbine power data; it obtains missing data in the wind turbine power time series and fills in the missing data using a piecewise linear regression model to obtain the target wind turbine power time series; it then trains an Extreme Learning Machine model, a Long Short-Term Memory Neural Network model, and a Temporal Convolutional Network model respectively; subsequently, it trains a multi-objective prediction model; it uses the multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model; it employs a deep deterministic policy gradient method to obtain the dynamic optimal weights of each model in the multi-objective prediction model based on the Pareto optimal solution set; and based on each model in the multi-objective prediction model and its corresponding dynamic optimal weights, it obtains the short-term wind power prediction results. This improves the accuracy and flexibility of short-term wind power prediction results. During the offline phase of the power grid, a generative adversarial network (GAN) model is constructed based on the power grid cross-section dataset. This GAN model is used to acquire at least one type of power grid operation risk data, generating a set of power grid operation risk scenarios. A time-series simulation method is then employed to extrapolate and calculate at least one power grid operation risk scenario, generating at least one sequence of forward-looking power grid scenarios. During the online phase of the power grid, based on preset risk assessment indicators and current power grid operation scenario information, power grid operation risk assessment results are obtained, and corresponding forward-looking power grid operation scenario sequences are matched and acquired. This enhances the accuracy and timeliness of power grid operation risk assessment. Attached Figure Description

[0061] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0062] Figure 1 is an application environment diagram of the power grid operation risk assessment method in one embodiment;

[0063] Figure 2 is a flowchart illustrating a power grid operation risk assessment method in one embodiment;

[0064] Figure 3 is a flowchart illustrating the power grid operation risk assessment method in another embodiment;

[0065] Figure 4 is a flowchart illustrating the power grid operation risk assessment method in the most detailed embodiment;

[0066] Figure 5 is a schematic diagram of the IEEE 39-node power grid risk assessment in one embodiment;

[0067] Figure 6 is a structural block diagram of a power grid operation risk assessment device in one embodiment;

[0068] Figure 7 is an internal structure diagram of a computer device in one embodiment. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0071] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0072] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0073] The wind power prediction method provided in this application embodiment can be applied to the application environment shown in Figure 1. The terminal 102 communicates with the server 104 via a network. A data storage system can store the data that the server 104 needs to process. The data storage system can be integrated onto the server 104, or it can be located in the cloud or on another network server.

[0074] Server 104 acquires power grid operation status data, meteorological monitoring data, and historical wind power time series data. Based on the historical wind power time series data, it trains an Extreme Learning Machine (ELM) model, a Long Short-Term Memory (LSTM) neural network (LSTM) model, and a temporal convolutional network (TCNN) model, respectively. Based on these models, it trains a multi-objective prediction model. Using the multi-objective Harris-Eagle algorithm, it obtains the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model. Using the deep deterministic policy gradient method, it obtains the dynamic optimal weights of each model in the multi-objective prediction model based on the Pareto optimal solution set. Finally, based on each model and its corresponding dynamic optimal weights, it obtains the wind power prediction results. Based on power grid operation status data, meteorological monitoring data, and wind power prediction results, a power grid cross-section dataset is generated. During the offline phase of the power grid, a generative adversarial network (GAN) model is constructed based on the power grid cross-section dataset. Through the GAN model, at least one type of power grid operation risk data is obtained. Based on this data, a power grid operation risk scenario set is generated, containing at least one power grid operation risk scenario. A time-series simulation method is used to extrapolate and calculate at least one power grid operation risk scenario, generating at least one power grid forward-looking scenario sequence. During the online phase of the power grid, preset risk assessment indicators and current power grid operation scenario information are obtained. Based on these indicators, power grid operation risk assessment results are obtained, and corresponding power grid forward-looking operation scenario sequences are matched and obtained.

[0075] The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle systems. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster consisting of multiple servers.

[0076] In an exemplary embodiment, as shown in FIG2, a power grid operation risk assessment method is provided. Taking the application of this method to the server in FIG1 as an example, the method includes the following steps S202 to S228. Wherein:

[0077] Step S202: Construct a wind turbine power time series based on historical wind turbine power data.

[0078] Among them, wind turbine power data refers to the power generated by the wind turbines in the wind farm, which can also be called wind power output data. In some embodiments, during the operation of the wind turbines in the wind farm, wind turbine power data is collected at preset time intervals, such as collecting wind turbine power data once every 15 minutes, in order to construct a wind turbine power time series.

[0079] In some embodiments, the constructed fan power time series can be represented as {P1, P2, ..., P...} n}, where P is the power value and n represents the number of elements in the sequence. For example, a wind turbine power time series could be {100, 102, 98, 100, 98, ...}. Since the power of each wind turbine in the wind turbine power time series is collected at preset time intervals, taking a collection every 15 minutes as an example, this wind turbine power time series can be represented by a corresponding time feature sequence {T1, T2, ..., T}. n}, P n With T n Corresponding, for example {2021-01-01 00:00, 2021-01-01 00:15, 2021-01-01 00:30, ...}.

[0080] Step S204: Obtain missing data in the wind turbine power time series, and fill in the missing data through a piecewise linear regression model to obtain the target wind turbine power time series;

[0081] Missing data refers to data that was not collected at the time of acquisition. This is understandable, as data collection may be affected by external factors, such as sudden power outages or shutdowns. Therefore, the wind turbine power time series data may contain some missing values, requiring automatic search and completion. Since missing data is generally stored as 0, it may be confused with normal 0-value wind turbine power data during the actual search process, necessitating accurate differentiation.

[0082] In some embodiments, piecewise linear regression models can be used to capture local features and patterns in the target wind turbine power time series. It is understood that this piecewise linear regression model does not simply perform linear regression on the entire wind turbine power time series; it can also divide the time series into different segments based on the local characteristics of the data and perform linear regression within each segment. This approach makes the model more adaptable to changes and fluctuations in the data, thereby enabling more accurate prediction of the target wind turbine power.

[0083] By using the above method and employing a piecewise linear regression model to fill in the missing data, a complete time series of the target wind turbine power was obtained. This data filling process helps to preserve local features in the data, thereby improving the accuracy of filling in missing values ​​and making the final time series more complete and reliable.

[0084] Step S206: Based on the time series data of historical wind power, train the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model respectively.

[0085] Specifically, based on historical wind power time series data, the following are the general steps for training wind power prediction using three models: Limited Memory Machine (ELM), Long Short-Term Memory (LSTM), and Time Convolutional Neural Network (TCN).

[0086] Extreme Learning Machine (ELM) model:

[0087] Prepare the dataset: Divide the historical wind power time series data into training and test sets, and normalize the data.

[0088] Constructing the ELM model: ELM is a single-hidden-layer feedforward neural network that randomly initializes the weight matrix and bias vector. Wind power data is input into the input layer, and the hidden layer output is calculated using the randomized weight matrix and bias vector.

[0089] Training the model: Adjust the hidden layer weight matrix and bias vector using the least squares method or regularization to make the model output as close as possible to the real wind power value.

[0090] Model evaluation: Use test set data to evaluate the model's predictive performance, such as calculating metrics like root mean square error (RMSE).

[0091] Long Short-Term Memory (LSTM) Neural Network Model:

[0092] Prepare the dataset: Transform the historical wind power time series data into input and target sequences, and normalize the data.

[0093] Building an LSTM model: LSTM is a recurrent neural network suitable for sequential data. It has memory units and gating mechanisms, which can capture long-term dependencies in the sequence. The model consists of an input layer, an LSTM layer, and an output layer.

[0094] Model training: By using the backpropagation algorithm and gradient descent optimizer, the weights and biases of the model are adjusted according to the input sequence and the target sequence, so that the model can accurately predict wind power.

[0095] Model evaluation: Use test set data to evaluate the model's predictive performance, such as calculating metrics like root mean square error (RMSE).

[0096] Temporal Convolutional Network (TCN) model:

[0097] Prepare the dataset: Convert the historical wind power time series data into input and target sequences, and perform normalization processing.

[0098] Building a TCN model: TCN is a sequence modeling method based on convolutional neural networks that captures long-term dependencies in sequences through one-dimensional convolutional layers and residual blocks. The model consists of an input layer, convolutional layers, residual blocks, and an output layer.

[0099] Model training: By using the backpropagation algorithm and gradient descent optimizer, the weights and biases of the model are adjusted according to the input sequence and the target sequence, so that the model can accurately predict wind power.

[0100] In training the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model, some parameters are set for each model based on empirical values, for example:

[0101] ELM model: The number of neurons in the hidden layer is set to 16;

[0102] LSTM model; number of hidden layer neurons set to 100; learning rate set to 0.001; maximum number of epochs set to 100; number of convolutional kernels set to 64.

[0103] TCN model: Convolution kernel number set to 64; training batch number set to 32; number of iterations set to 50; inactivation rate set to 0.2.

[0104] Step S208: Train a multi-objective prediction model based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model.

[0105] Specifically, the entire multi-objective model is trained using backpropagation and optimization algorithms, with the goal of minimizing the model's error on the target variable. Gradient descent or other optimization algorithms can be used for model training. The trained multi-objective prediction model is then evaluated using a test dataset. The evaluation metric can be chosen appropriately based on the type of target variable, such as root mean squared error (RMSE) or mean absolute error (MAE). Once the model is trained and passes evaluation, it can be used to predict new target variable values.

[0106] The expression for the multi-objective prediction model is as follows:

[0107] In the formula, y i It is the measured power at time i. Let w represent the prediction of a single model at time i, respectively. i,1 ,w i,2 ,wi,3 It is the combined weight at time i.

[0108] Step S210: Use the multi-objective Harris Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the multi-objective prediction model.

[0109] Specifically, the multi-objective Harris Eagle algorithm is a commonly used method for solving multi-objective optimization problems. It transforms the multi-objective prediction task into a multi-objective optimization problem, where each objective variable can be obtained from a different prediction model. An objective function can be defined to measure the optimization objective of each objective. This objective function may include metrics such as prediction error, generalization ability, and complexity. The feasible region of the Pareto optimal solution set is determined. The feasible region determines the range of values ​​for the weight vector, and constraints can be set according to actual needs. A flock of eagles is generated using a random initialization method. Each eagle represents a possible weight vector, and through continuous iteration and optimization, the flock gradually converges to the optimal solution. The fitness of each eagle is calculated; fitness is evaluated based on the value of the objective function, usually according to the definition of Pareto optimality. The position of each eagle is updated according to the selected update strategy. The update strategy can adopt standard search strategies in the Harris Eagle algorithm, such as exploration and tracking, population search, and encounter and merging. The flock is checked to see if it has converged to a stable Pareto optimal solution set. If the stopping condition has not been met, iterative optimization continues. The final stable group is the Pareto optimal solution set, which contains the optimal solution of multiple weight vector combinations.

[0110] The multi-objective optimization is defined as follows: min f(x)=[f1(x),f2(x),...f n (x)]

[0111] Where g i (x), h j Given that g(x) and h(x) are constraints, find the minimum value of f(x) under the condition that g(x) and h(x) are satisfied.

[0112] Given vector and

[0113] If and only if hour,

[0114] and It is the Pareto optimal solution, if The set of Pareto optimal solutions is denoted as .

[0115] Furthermore, P f ={f(x)|x∈Pset This is called the Pareto front. All Pareto optimal solutions constitute the Pareto optimal solution set. These solutions, mapped through the objective function, form the Pareto optimal front or Pareto front surface of the problem. That is, the objective function value corresponding to the Pareto optimal solution is the Pareto optimal front.

[0116] Step S212: Using the deep deterministic strategy gradient method, the dynamic optimal weights of each model in the multi-objective prediction model are obtained based on the Pareto optimal solution set.

[0117] Among them, Deep Deterministic Policy Gradient (DDPG) is a commonly used reinforcement learning algorithm that can be used to learn dynamic optimal weights.

[0118] Specifically, the state space and action space of the problem are defined. In a multi-objective prediction model, the state space can be indicators such as the prediction error and generalization ability of each prediction model, while the action space can be the individual weights in the weight vector. A DDPG network consisting of two neural networks is constructed: an Actor network for learning the action policy and a Critic network for evaluating the value of the action. The Actor network outputs the action vectors of each weight in the current state, and the Critic network outputs a Q-value to evaluate the value of the action based on the Actor's action vectors and state information. The parameters of the Actor and Critic networks are randomly initialized. According to the DDPG policy, the Actor network selects and executes actions to obtain the reward signal and the next state. Using these samples, the parameters of the Critic network are updated, and the loss function of the Critic network is minimized using gradient descent. Simultaneously, the parameters of the Actor network are updated based on the output of the Critic network, and the objective function of the Actor network is maximized using gradient ascent. This process of sampling, optimizing, and updating network parameters is repeated until the network converges or reaches the preset number of iterations. After DDPG training is complete, the dynamic optimal weights of each model in the multi-objective prediction model can be obtained through the Actor network. The Actor network is used to input the current state and output a weight vector, which is then used as the weights of the model.

[0119] Step S214: Obtain the wind power prediction results based on each model in the multi-objective prediction model and its corresponding dynamic optimal weights.

[0120] Based on the current wind input and the characteristics of the prediction models, each prediction model is used to obtain its own prediction results. A corresponding dynamic optimal weight is applied to the prediction results of each model. Each prediction result is multiplied by its corresponding dynamic optimal weight to obtain a weighted prediction result. The weighted prediction results are then summed to obtain the final wind power prediction result.

[0121] The aforementioned wind power prediction method employs a combination of multiple prediction models, taking into account the complex interactions between different models and thus improving the accuracy of short-term wind power predictions. Using multiple prediction models and adjusting the dynamic optimal weights allows for flexible adjustment of the weights of each model based on actual conditions, enhancing the adaptability and flexibility of the prediction method. By using the multi-objective Harris-Eagle algorithm to obtain the Pareto optimal solution set of the combined weight vector of the prediction models, multiple prediction strategies can be provided for selection while maintaining prediction accuracy. Furthermore, by using the deep deterministic policy gradient method, the optimal weights of each model can be dynamically adjusted based on the real-time Pareto optimal solution set, further improving the accuracy and adaptability of the prediction. This method can improve the accuracy and flexibility of short-term wind power prediction results, providing important support for wind energy utilization and power system stability.

[0122] In an exemplary embodiment, based on historical wind power time-series data, an Extreme Learning Machine model, a Long Short-Term Memory Neural Network model, and a Temporal Convolutional Network model are trained, respectively, including:

[0123] The CEEMDAN sequence decomposition method is used to decompose the time series data of historical wind power to obtain multiple first subsequence data. Using multiple second subsequence data, an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained simultaneously.

[0124] Specifically, the sequence decomposition method CEEMDAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise) is used to decompose the historical wind power time series data into multiple first subsequence data. CEEMDAN can effectively decompose the original sequence into multiple subsequences, each representing a different frequency component or mode. The multiple first subsequence data obtained from the decomposition are used as input, and appropriate training, validation, and test sets are set according to requirements to ensure that the data partitioning is reasonable.

[0125] Compared to traditional decomposition methods (EMD, VMD, and WPD), CEEMDAN overcomes the problems of mode aliasing in Empirical Mode Decomposition and random residual noise in Aggregated Empirical Mode Decomposition. By adding Gaussian white noise at each stage of decomposition before performing EMD decomposition, the problem of residual noise is solved, enabling sufficient decomposition of time series with complex variation characteristics such as wind power.

[0126] In this embodiment, ELM, LSTM, and TCN models are trained based on the multiple first subsequence data decomposed by the CEEMDAN method. This allows for the selection of the most suitable model as needed, and wind power prediction based on historical wind power data.

[0127] Step S216: Generate a power grid section dataset based on the power grid operation status data, meteorological monitoring data, and wind power prediction results.

[0128] Specifically, the collected power grid operation status data and meteorological monitoring data are preprocessed and integrated, including data cleaning, noise reduction, and correction, to ensure data accuracy and consistency. Specific cross-sectional areas and parameters are selected based on the specific purpose of the power grid analysis and assessment. These parameters can be power grid components within a specific area or specific power grid characteristics, such as transmission line length and substation capacity. Using the preprocessed and integrated data, a power grid cross-sectional dataset is generated and labeled and categorized. Labeling can be based on indicators such as power grid operation risk level and failure probability for subsequent risk assessment and analysis. Next, the generated power grid cross-sectional dataset is validated to ensure data integrity and accuracy. This may include comparison and comparative analysis with actual observation data.

[0129] Specifically, the power grid section dataset described in step S202 is represented as: N = [n1, n2, ... n i ,...,n m ];

[0130] In the formula, n i Let s be the feature information vector of the i-th cross section. mi For the m-th electrical measurement information of the i-th section, w mi This represents the m-th meteorological measurement information for the i-th cross section.

[0131] Step S218: During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset.

[0132] Specifically, in the offline phase of the power grid, constructing a generative adversarial network (GAN) model based on the power grid cross-section dataset refers to utilizing the generative capabilities of the GAN model to train a GAN model that uses the power grid cross-section dataset as input, enabling it to learn and generate virtual power grid cross-section data with similar characteristics.

[0133] Generative Adversarial Networks (GANs) are neural network models consisting of a generator and a discriminator. They are trained adversarially, enabling the generator to gradually produce virtual data that resembles real data, while the discriminator attempts to accurately distinguish between real and generated data. This adversarial training gradually improves the generator's performance, and the generated data becomes increasingly closer to the distribution of real data.

[0134] Building upon the existing power grid cross-section dataset, a GAN model can be used to generate more virtual power grid cross-section data. The GAN model can simulate new synthetic datasets by learning the correlations and distributions between power grid cross-section data. To train this GAN model, a suitable loss function needs to be defined to evaluate the difference between the generator-generated data and the real data, and the model's parameters need to be updated using a backpropagation algorithm.

[0135] The constructed GAN model can be used to generate virtual power grid cross-section data similar to the existing power grid cross-section dataset. This virtual data can supplement the deficiencies of existing datasets, expand the training set size, and improve the model's generalization ability. Simultaneously, the generated virtual data can also be used to simulate various risk scenarios during power grid operation, for risk assessment and decision support.

[0136] By constructing a generative adversarial network (GAN) model and training it using a power grid cross-section dataset, a model capable of generating virtual power grid cross-section data similar to real power grid data can be obtained. This model can learn the distribution and characteristics of power grid data to generate data with similar features and statistical attributes, providing more data resources for power grid operation risk assessment and related applications.

[0137] Step S220: Obtain at least one type of power grid operation risk data by generating an adversarial network model.

[0138] Specifically, a new batch of power grid operation risk data is generated using a pre-trained GAN model. The generated data will have similar characteristics and distribution to the training data. Feature extraction is then performed on the generated data and the original data to obtain high-dimensional risk feature vectors. Various feature extraction methods can be used, such as principal component analysis (PCA) or other dimensionality reduction techniques.

[0139] Step S222: Based on at least one type of power grid operation risk data, generate a set of power grid operation risk scenarios. The set of power grid operation risk scenarios contains at least one power grid operation risk scenario. Calculate and screen high-dimensional risk features using the K-Means clustering method.

[0140] Specifically, a common scenario for understanding grid operation risks is power load fluctuation. During grid operation, load fluctuations can lead to power system overload, voltage fluctuations, and other potential risks. Below is a collection of example scenarios that include grid operation risks caused by load fluctuations:

[0141] Scenario 1: Voltage drop during high load periods

[0142] Description: During periods of high load, the power system faces enormous load demand that exceeds its supply capacity. This causes a drop in supply voltage, which may lead to equipment failure or power outages for users.

[0143] Features: High load, supply voltage drop, potential user equipment failure or power outage risk.

[0144] Solutions: Increase generation capacity, improve power system dispatch and management strategies, and optimize the power supply network.

[0145] Scenario 2: Sudden increase in load

[0146] Description: Due to certain events (such as sudden weather, equipment failure, etc.), the power system suffers a sudden increase in load that exceeds the system's capacity, which may lead to overload and voltage drop.

[0147] Characteristics: Sudden increase in load, power system capacity overload, potential overload and voltage drop risk.

[0148] Solution: Implement load management strategies, respond promptly to load changes, and improve the dispatchability of power supply and distribution equipment.

[0149] Scenario 3: Uneven load distribution

[0150] Description: Due to uneven load distribution or power supply system failure, some areas or lines may carry excessive loads, which may cause some areas of the power grid to be overloaded while other parts are underloaded.

[0151] Characteristics: Uneven load distribution, overload in some areas or on some lines, and light load in other areas.

[0152] Solutions: Optimize load distribution, improve power supply network planning, and expand power transmission and distribution capacity in some areas.

[0153] These risk scenarios provide examples of the risks that load fluctuations may cause during power grid operation and their potential solutions. Note that actual power grid risk scenarios may be more complex and need to be determined based on specific power grid data and circumstances.

[0154] The extracted high-dimensional risk feature vectors are input into the K-Means clustering algorithm to divide the risk data into different clusters. K-Means is a commonly used unsupervised learning algorithm that can automatically cluster similar data points together. Calculating and filtering high-dimensional risk features using the K-Means clustering method can help identify and analyze risk patterns and anomalies in power grid operation scenarios. This can provide valuable information for power system managers to take appropriate measures to address potential risk problems.

[0155] Step S224: Using a time-series simulation method, at least one power grid operation risk scenario is simulated and calculated to generate at least one power grid forward-looking scenario sequence.

[0156] Specifically, time-series simulation is a commonly used tool for power grid risk assessment. It simulates power grid operation to extrapolate and calculate a sequence of future power grid scenarios. First, one or more power grid operation risk scenarios are selected, such as load surges or transmission line faults, and the necessary simulation parameters are determined, such as the probability of fault occurrence and repair time. Next, time-series simulation is used to perform power grid operation simulation calculations according to the determined risk scenarios and parameters. The simulation can consider load changes, generator start-up and shutdown, fault occurrence and repair, etc. Based on the simulation results, the state of the power grid at different points in time is obtained, allowing analysis of information regarding the grid's reliability, stability, and vulnerability. Potential problems can be identified, and future power grid operation can be predicted. Based on the analysis results, a sequence of future power grid scenarios is generated. These sequences can help power grid operators formulate reasonable operating strategies to cope with possible future risk scenarios.

[0157] Step S226: During the online operation phase of the power grid, obtain preset risk assessment indicators and current power grid operation scenario information.

[0158] Specifically, during the online operation phase of the power grid, real-time power grid operation data, including parameters such as power load, generator output, voltage, and current, is acquired through monitoring equipment such as sensors and monitoring devices. This data can be used to assess the operating status and performance of the power grid. Relevant environmental data, such as weather data, temperature, and humidity, is also acquired. This data can affect the operation of the power grid, for example, wind intensity in a wind power generation scenario. A series of risk assessment indicators are developed based on the safety and reliability requirements of the power grid. These indicators can be used to assess the risk level of the power grid under different scenarios, such as voltage stability, frequency stability, and capacity utilization. These indicators can be set and modified according to actual conditions. Historical data and fault records are used to analyze and review the operation of the power grid to obtain information on past power grid operation scenarios. This information can be used to improve the accuracy of risk assessment and identify potential problems and risks. Based on the preset risk assessment indicators and the collected power grid operation scenario information, corresponding risk assessment models and algorithms are used to calculate and assess the risks of the power grid. These models and algorithms can be selected and adjusted according to actual needs and scenarios to achieve accurate risk assessment.

[0159] Step S228: Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

[0160] Specifically, by utilizing collected power grid operation scenario information and pre-set risk assessment indicators, and employing corresponding risk assessment models and algorithms, the operational risks of the power grid are evaluated. Based on the pre-set indicators and assessment results, the current risk level of the power grid is determined. Based on the risk assessment results, corresponding forward-looking power grid operation scenario sequences are matched. These sequences can be pre-planned strategies and measures to address power grid operation conditions under different risk levels. For example, in the event of a sudden increase in power load, there can be corresponding generator start-up and shutdown strategies; in the event of transmission line faults, there can be corresponding line switching and repair strategies. Based on the forward-looking power grid operation scenario sequences, the power grid operation strategies are further analyzed and optimized. This may include adjusting generator output, optimizing power load dispatching, and developing backup power strategies to improve the safety and reliability of the power grid.

[0161] In the aforementioned power grid operation risk assessment method, a power grid cross-section dataset is generated by acquiring power grid operation status data and meteorological monitoring data. Then, a generative adversarial network (GAN) model is used to generate power grid operation risk data. This method leverages the advantages of GAN models to generate more accurate power grid operation risk data, further improving the accuracy of risk assessment. The power grid operation risk data generated using the GAN model can generate various power grid operation risk scenarios. Then, time-series simulation methods are used to extrapolate and calculate these risk scenarios, generating a power grid forward-looking scenario sequence. This allows for better evaluation of the power grid's operation under different risk scenarios and early prediction of potential problems. During the online operation phase of the power grid, preset risk assessment indicators and current power grid operation scenario information are acquired. Based on this information, power grid operation risk assessment results are obtained and matched with corresponding power grid forward-looking operation scenario sequences. This process can be performed in real time, providing real-time risk assessment results to help decision-makers make rapid adjustments during operation. In summary, the new power grid operation risk assessment method utilizes GAN models and time-series simulation methods to improve the accuracy and timeliness of power grid risk assessment, making power grid operation more stable and reliable.

[0162] In an exemplary embodiment, as shown in Figure 3, the steps employ a time-series simulation method to extrapolate and calculate at least one power grid operation risk scenario, generating at least one power grid prospective scenario sequence, including steps S302 to S306. Wherein: the time-series simulation method includes Monte Carlo decision trees;

[0163] Step S302: Use the weighted cross-entropy function as the loss function of the Monte Carlo decision tree, and obtain the evaluation value of at least one power grid operation risk scenario through the Monte Carlo decision tree.

[0164] Step S304: Divide the evaluation values ​​of at least one power grid operation risk scenario and obtain the processing results.

[0165] Step S306: A parallel inference architecture is adopted, and the processing results are subjected to multi-threaded parallel inference calculations using Monte Carlo decision trees to generate at least one power grid prospective scenario sequence. Multi-threaded inference calculations are provided for multi-branch scenarios, outputting a power grid operation inference scenario sequence under different objectives such as renewable energy consumption, economic optimization, and safety constraints.

[0166] Specifically, a weighted cross-entropy function can be used as the loss function for Monte Carlo decision trees to evaluate power grid operation risk scenarios. Monte Carlo decision trees are a probability-based decision tree model that fully considers uncertainties and weights. When constructing a Monte Carlo decision tree, the nodes and boundary conditions can be set according to the characteristics of the power grid and the risk scenarios to be evaluated. These conditions can include power load levels, generator status, transmission line parameters, etc., ensuring the rationality and operability of the decision tree. The Monte Carlo method is used to construct and optimize the decision tree. Through random sampling and simulation evaluation using the Monte Carlo method, a sampled set of power grid operation risk scenarios and operational information for each scenario are obtained, and risk assessment values ​​under different decision paths are calculated.

[0167] Specifically, a weighted cross-entropy function is used to calculate the risk assessment value, and the result is adjusted according to the weights. This ultimately yields one or more power grid operation risk scenarios with high assessment values. For power grid operation risk scenarios with high assessment values, further processing can be performed based on actual needs and priorities. For example, the assessment values ​​can be divided into different priorities for better response and control.

[0168] Specifically, a parallel inference architecture is adopted, and Monte Carlo decision trees are used to perform multi-threaded parallel inference calculations on the processing results to generate at least one power grid forward-looking scenario sequence. These forward-looking scenario sequences can be used to formulate power grid emergency response strategies, adjust power load scheduling, and so on, to address different power grid operation risks.

[0169] In this embodiment, by evaluating Monte Carlo decision trees and performing multi-threaded parallel inference calculations, a probability- and weight-based forward-looking scenario sequence for the power grid can be provided to assist in decision-making and planning measures, thereby improving the safety and reliability of power grid operation.

[0170] In an exemplary embodiment, the evaluation values ​​of at least one power grid operation risk scenario are divided to obtain the processing result, including:

[0171] By using a preset evaluation value calculation formula, the maximum expected evaluation value for each power grid operation risk scenario is obtained.

[0172] Based on maximizing the expected evaluation value, at least one power grid operation risk scenario is divided and processed to obtain a processing result with multiple power grid operation risk scenario sequences.

[0173] The formula for calculating the evaluation value is as follows:

[0174] In the formula, Ε is the expectation operator; T is the visit step size; Q(v t ) for accessing scene v t Reward value; N(v)p N(v) represents the number of times the parent scene is visited; c ) represents the number of times the sub-scene is accessed; V pen This is the penalty term for violating the constraints; γ is a constant coefficient, and γ > 0.

[0175] In this embodiment, each power grid operation risk scenario is simulated and calculated according to a given evaluation value calculation formula. Based on the expectation operator and other relevant parameters, the maximum simulated expected evaluation value for each scenario is calculated. Based on the maximum simulated expected evaluation value, each power grid operation risk scenario is divided into segments. The power grid operation risk scenario can be divided into multiple segments based on different ranges of evaluation values ​​or set thresholds, with each segment representing a different risk level or processing priority. Based on the segmentation results, the power grid operation risk scenario sequences from each segment are combined to form a processing result with multiple power grid operation risk scenario sequences. Each scenario sequence can represent a different risk level, allowing corresponding countermeasures to be adjusted for different risk levels.

[0176] During the segmentation process, appropriate evaluation value ranges and thresholds, as well as corresponding penalty items, can be set according to specific needs and relevant constraints. This allows for the division of power grid operation risk scenarios into multiple segments based on different evaluation values ​​and constraints, facilitating better risk management and the planning of corresponding control strategies.

[0177] In an exemplary embodiment, the loss function of a Monte Carlo decision tree is:

[0178] In the formula, m represents the number of scenes; when the current scene is located in the target scene set, then l i If the value is 1, then l is not in the target scene set. i p is 0; 1i p 2i These are the probability values ​​that the current scene belongs to the target scene set; r key The weight representing the percentage of scenes located in the target scene set; The percentage weight of the number of scenes that are not located in the target scene set.

[0179] In this embodiment, the first part of the loss function measures the model's ability to correctly classify a scene as belonging to the target scene set, while the second part measures the model's ability to correctly classify a scene as not belonging to the target scene set. By balancing these two parts, the loss function weights the penalties for correct and incorrect classifications. In Monte Carlo decision trees, the optimization algorithm attempts to minimize the loss function to find the optimal decision tree model. Based on specific weights and scene probability values, the algorithm can determine the optimal decision point and splitting strategy to minimize misclassification and missing target scenes, thereby improving the model's prediction accuracy and decision-making performance.

[0180] In an exemplary embodiment, the evaluation values ​​of at least one power grid operation risk scenario are divided to obtain the processing result, including:

[0181] Constraints are set to apply to the Monte Carlo decision tree. The constrained Monte Carlo decision tree is then used to partition the evaluation values ​​of at least one power grid operation risk scenario, yielding the processing results. The constraints include an evaluation penalty term, expressed as: ∑V pen,i =V pen,bl +V pen,op +V pen,pf ;

[0182] In the formula, V pen,bl V pen,op V pen,pf These are operational constraint penalties, safety and stability penalties, and power balance penalties, respectively.

[0183] In this embodiment, the optimization algorithm can use this objective function, combined with other constraints, to adjust the partitioning strategy and decision points of the Monte Carlo decision tree to minimize the value of the objective function. Through this constraint and optimization method, a processing result can be obtained that, while satisfying the constraints, better classifies and evaluates the target power grid operation risk scenarios.

[0184] In one exemplary embodiment, obtaining a preset risk value assessment index includes:

[0185] Obtain the power grid's loss of load risk value, voltage over-limit risk value, and line over-limit risk value;

[0186] Based on the underload risk value, voltage over-limit risk value, and line over-limit risk value, a preset risk value assessment index is obtained according to a preset weighting ratio; wherein, the expression for the underload risk value is:

[0187] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S;C (E k ) represents the system state E k The severity of the consequences of load loss; P ci ND represents the load shedding amount at load node i, and ND represents the number of load nodes;

[0188] The expression for the voltage over-limit risk value is:

[0189] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; v (E k ) represents the system state E k The severity of the consequences of exceeding the voltage limit, where NI is the number of system nodes;

[0190] The expression for the risk value of exceeding the limit on the line is:

[0191] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; s (E k ) represents the system state E k The severity of the consequences of voltage exceeding the limit; NL is the number of system nodes, P Li P is the active power flowing through branch i. Limax Let be the maximum active power allowed to flow through branch i.

[0192] In this embodiment, the technical effect of obtaining the power grid's load shedding risk value, voltage over-limit risk value, and line over-limit risk value, and calculating the preset risk value assessment index based on these risk values, is to improve the stability and reliability of the power grid.

[0193] The risk values ​​of power grid load shedding, voltage over-limit, and line over-limit are important indicators in power grid operation, reflecting the stability and reliability of the power grid under specific conditions. By calculating and analyzing these risk values, potential problems in the power grid can be identified in a timely manner, and corresponding measures can be taken to solve them, thereby reducing the probability of power grid failures and improving the stability and reliability of the power grid.

[0194] The preset risk assessment indicators are calculated based on the actual operation of the power grid and historical data, and can comprehensively reflect the risk level of the power grid in different aspects. By comparing the preset risk assessment indicators with the actual risk values, the operating status and reliability of the power grid can be evaluated, providing a reference for the operation management and maintenance of the power grid, thereby better ensuring the safe and stable operation of the power grid.

[0195] In summary, obtaining the risk values ​​of power grid load shedding, voltage over-limit, and line over-limit, and calculating the preset risk value assessment indicators based on preset weight ratios, can improve the stability and reliability of the power grid and provide important reference for its operation, management, and maintenance.

[0196] In the most specific embodiment, referring to FIG4, FIG4 is a flowchart of the most specific embodiment.

[0197] The process involves acquiring power grid operation status data and meteorological monitoring data to generate a power grid cross-section dataset. A generative adversarial network (GAN) model is then constructed to generate a set of power grid operation risk scenarios (currently at iteration number 1). Each risk scenario is then clustered and subjected to discriminant analysis. Monte Carlo sampling is performed, followed by selection using a decision tree with weighted cross-entropy function and evaluation values. Constraints are set for the Monte Carlo method during the simulation calculations, and its accuracy is verified. If the accuracy meets the requirements, parallel simulation calculations are performed to obtain power grid operation risk assessment results and a sequence of forward-looking power grid operation scenarios. If the accuracy does not meet the requirements, constraints are added or removed. After parallel simulation calculations are completed, it is determined whether all scenarios have been traversed. If not, the above steps are iterated again. If completed, a comprehensive power grid risk assessment model is generated based on the results of the parallel simulation calculations and then processed offline. During the online operation phase of the power grid, preset risk assessment indicators and current power grid operation scenario information are acquired. Real-time risk assessment is performed using this model to obtain relatively accurate risk assessment results. These results are then rolled back to the model for deep learning.

[0198] This embodiment is based on the IEEE 39-node system for power grid operation risk assessment testing. The number of nodes is 39, the number of lines is 38, the number of power grid operation simulation scenarios is 10,000, the time resolution of the time series extrapolation calculation is 15 minutes, and the extrapolation time window is 24 hours.

[0199] The test results are shown in Figure 5. The example is a power system risk assessment test based on generative adversarial networks and time-series extrapolation methods. The assessment results are represented in a hierarchical manner. The identified risk alarm lines include 1-4, 11-15, and 25-36, and the risk alarm nodes are 15, 17, and 23. The comprehensive risk assessment risk value is 0.731, which belongs to the network-wide risk alarm level. The calculation time for a single section risk assessment is 1.39 seconds.

[0200] The test results of the embodiments show that the risk assessment method provided by the present invention can accurately and quickly assess the risk lines and nodes of the power grid, solve the comprehensive operation risk level of the power grid, and meet the accuracy and timeliness requirements of risk identification in dispatching operations.

[0201] It should be understood that although the steps in the flowcharts of the above embodiments are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the above embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0202] Based on the same inventive concept, this application also provides a power grid operation risk assessment device for implementing the power grid operation risk assessment method described above. The solution provided by this device is similar to the solution described in the above method; therefore, the specific limitations in one or more power grid operation risk assessment device embodiments provided below can be found in the limitations of the power grid operation risk assessment method described above, and will not be repeated here.

[0203] In an exemplary embodiment, referring to Figures 6 and 7, a power grid operation risk assessment device is provided, including: a data acquisition module 702, used to acquire power grid operation status data and meteorological monitoring data;

[0204] The data integration module 704 is used to generate a power grid section dataset based on power grid operation status data and meteorological monitoring data.

[0205] Modeling module 706 is used to construct a generative adversarial network model based on a power grid section dataset during the offline phase of the power grid; and to obtain at least one type of power grid operation risk data through the generative adversarial network model.

[0206] The processing module 708 is used to generate a set of power grid operation risk scenarios based on at least one type of power grid operation risk data, wherein the set of power grid operation risk scenarios includes at least one power grid operation risk scenario.

[0207] The simulation and evaluation module 710 is used to perform simulation calculations on at least one power grid operation risk scenario using a time-series operation simulation method, and generate at least one power grid forward-looking scenario sequence.

[0208] The data acquisition module 702 is also used to acquire preset risk assessment indicators and current power grid operation scenario information during the online operation phase of the power grid;

[0209] The online application module 712 is used to obtain the power grid operation risk assessment results based on the preset risk assessment indicators and the current power grid operation scenario information, and to match and obtain the corresponding power grid forward operation scenario sequence.

[0210] In an exemplary embodiment, the time-series simulation method includes a Monte Carlo decision tree; the inference and evaluation module 710 is further configured to:

[0211] The weighted cross-entropy function is used as the loss function of the Monte Carlo decision tree, and the evaluation value of at least one power grid operation risk scenario is obtained through the Monte Carlo decision tree.

[0212] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results;

[0213] A parallel inference architecture is adopted, and Monte Carlo decision trees are used to perform multi-threaded parallel inference calculations on the processing results to generate at least one power grid prospective scenario sequence.

[0214] In an exemplary embodiment, the inference and evaluation module 710 is further configured to:

[0215] By using a preset evaluation value calculation formula, the maximum expected evaluation value for each power grid operation risk scenario is obtained.

[0216] Based on maximizing the expected evaluation value, at least one power grid operation risk scenario is divided and processed to obtain a processing result with multiple power grid operation risk scenario sequences.

[0217] The formula for calculating the evaluation value is as follows:

[0218] In the formula, Ε is the expectation operator; T is the visit step size; Q(v t ) for accessing scene v t Reward value; N(v) p N(v) represents the number of times the parent scene is visited; c ) represents the number of times the sub-scene is accessed; V pen This is the penalty term for violating the constraints; γ is a constant coefficient, and γ > 0.

[0219] In an exemplary embodiment, the loss function of a Monte Carlo decision tree is:

[0220] In the formula, m represents the number of scenes; when the current scene is located in the target scene set, then l i If the value is 1, then l is not in the target scene set. i p is 0; 1i p 2i These are the probability values ​​that the current scene belongs to the target scene set; rkey The weight representing the percentage of scenes located in the target scene set; The percentage weight of the number of scenes that are not located in the target scene set.

[0221] In an exemplary embodiment, the deduction and evaluation module 710 is further configured to: set constraints, constrain the Monte Carlo decision tree, and use the constrained Monte Carlo decision tree to partition the evaluation values ​​of at least one power grid operation risk scenario to obtain a processing result, wherein the constraints include an evaluation penalty term, the expression of which is: ∑V pen,i =V pen,bl +V pen,op +V pen,pf ;

[0222] In the formula, V pen,bl V pen,op V pen,pf These are operational constraint penalties, safety and stability penalties, and power balance penalties, respectively.

[0223] In one exemplary embodiment, the data acquisition module 702 is further configured to:

[0224] Obtain the power grid's loss of load risk value, voltage over-limit risk value, and line over-limit risk value;

[0225] Based on the underload risk value, voltage over-limit risk value, and line over-limit risk value, a preset risk value assessment index is obtained according to a preset weighting ratio; wherein, the expression for the underload risk value is:

[0226] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; C (E k ) represents the system state E k The severity of the consequences of load loss; P ci ND represents the load shedding amount at load node i, and ND represents the number of load nodes;

[0227] The expression for the voltage over-limit risk value is:

[0228] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; v (E k ) represents the system state E k The severity of the consequences of exceeding the voltage limit, where NI is the number of system nodes;

[0229] The expression for the risk value of exceeding the limit on the line is:

[0230] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; s (E k ) represents the system state E k The severity of the consequences of voltage exceeding the limit; NL is the number of system nodes, P Li P is the active power flowing through branch i. Limax Let be the maximum active power allowed to flow through branch i.

[0231] Each module in the aforementioned power grid operation risk assessment device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0232] In an exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram is shown in Figure 7. The computer device includes a processor, memory, input / output interfaces (I / O), and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is connected to the system bus via the I / O interfaces. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database of the computer device stores power grid operation status data and meteorological monitoring data. The I / O interfaces of the computer device are used for information exchange between the processor and external devices. The communication interface of the computer device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements a power grid operation risk assessment method.

[0233] Those skilled in the art will understand that the structure shown in Figure 7 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0234] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0235] Acquire power grid operation status data and meteorological monitoring data;

[0236] Based on the power grid operation status data and the meteorological monitoring data, a power grid section dataset is generated;

[0237] During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset.

[0238] The generative adversarial network model is used to obtain at least one type of power grid operation risk data;

[0239] Based on the at least one type of power grid operation risk data, a power grid operation risk scenario set is generated, wherein the power grid operation risk scenario set includes at least one power grid operation risk scenario.

[0240] Using a time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid forward-looking scenario sequence;

[0241] During the online operation phase of the power grid, pre-set risk assessment indicators and current power grid operation scenario information are obtained;

[0242] Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

[0243] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0244] The time-series simulation method includes Monte Carlo decision trees; using the time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid prospective scenario sequence, including:

[0245] The weighted cross-entropy function is used as the loss function of the Monte Carlo decision tree, and the evaluation value of at least one power grid operation risk scenario is obtained through the Monte Carlo decision tree.

[0246] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results;

[0247] A parallel inference architecture is adopted, and Monte Carlo decision trees are used to perform multi-threaded parallel inference calculations on the processing results to generate at least one power grid prospective scenario sequence.

[0248] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0249] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0250] By using a preset evaluation value calculation formula, the maximum expected evaluation value for each power grid operation risk scenario is obtained.

[0251] Based on maximizing the expected evaluation value, at least one power grid operation risk scenario is divided and processed to obtain a processing result with multiple power grid operation risk scenario sequences.

[0252] The formula for calculating the evaluation value is as follows:

[0253] In the formula, Ε is the expectation operator; T is the visit step size; Q(v t ) for accessing scene v t Reward value; N(v) p N(v) represents the number of times the parent scene is visited; c ) represents the number of times the sub-scene is accessed; V pen This is the penalty term for violating the constraints; γ is a constant coefficient, and γ > 0.

[0254] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0255] The loss function for a Monte Carlo decision tree is:

[0256] In the formula, m represents the number of scenes; when the current scene is located in the target scene set, then l i If the value is 1, then l is not in the target scene set. i p is 0; 1i p 2i These are the probability values ​​that the current scene belongs to the target scene set; r key The weight representing the percentage of scenes located in the target scene set; The percentage weight of the number of scenes that are not located in the target scene set.

[0257] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0258] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0259] Constraints are set to apply to the Monte Carlo decision tree. The constrained Monte Carlo decision tree is then used to partition the evaluation values ​​of at least one power grid operation risk scenario, yielding the processing results. The constraints include an evaluation penalty term, expressed as: ∑V pen,i =V pen,bl +V pen,op +V pen,pf ;

[0260] In the formula, V pen,bl V pen,op V pen,pf These are operational constraint penalties, safety and stability penalties, and power balance penalties, respectively.

[0261] In one embodiment, the processor, when executing a computer program, also performs the following steps:

[0262] Obtain preset risk assessment indicators, including:

[0263] Obtain the power grid's loss of load risk value, voltage over-limit risk value, and line over-limit risk value;

[0264] Based on the underload risk value, voltage over-limit risk value, and line over-limit risk value, a preset risk value assessment index is obtained according to a preset weighting ratio; wherein, the expression for the underload risk value is:

[0265] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; C (E k ) represents the system state E k The severity of the consequences of load loss; P ci ND represents the load shedding amount at load node i, and ND represents the number of load nodes;

[0266] The expression for the voltage over-limit risk value is:

[0267] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; v (E k ) represents the system state E k The severity of the consequences of exceeding the voltage limit, where NI is the number of system nodes;

[0268] The expression for the risk value of exceeding the limit on the line is:

[0269] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; s (E k ) represents the system state E k The severity of the consequences of voltage exceeding the limit; NL is the number of system nodes, P Li P is the active power flowing through branch i. Limax Let be the maximum active power allowed to flow through branch i.

[0270] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:

[0271] Acquire power grid operation status data and meteorological monitoring data;

[0272] Based on the power grid operation status data and the meteorological monitoring data, a power grid section dataset is generated;

[0273] During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset.

[0274] The generative adversarial network model is used to obtain at least one type of power grid operation risk data;

[0275] Based on the at least one type of power grid operation risk data, a power grid operation risk scenario set is generated, wherein the power grid operation risk scenario set includes at least one power grid operation risk scenario.

[0276] Using a time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid forward-looking scenario sequence;

[0277] During the online operation phase of the power grid, pre-set risk assessment indicators and current power grid operation scenario information are obtained;

[0278] Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

[0279] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0280] The time-series simulation method includes Monte Carlo decision trees; using the time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid prospective scenario sequence, including:

[0281] The weighted cross-entropy function is used as the loss function of the Monte Carlo decision tree, and the evaluation value of at least one power grid operation risk scenario is obtained through the Monte Carlo decision tree.

[0282] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results;

[0283] A parallel inference architecture is adopted, and Monte Carlo decision trees are used to perform multi-threaded parallel inference calculations on the processing results to generate at least one power grid prospective scenario sequence.

[0284] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0285] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0286] By using a preset evaluation value calculation formula, the maximum expected evaluation value for each power grid operation risk scenario is obtained.

[0287] Based on maximizing the expected evaluation value, at least one power grid operation risk scenario is divided and processed to obtain a processing result with multiple power grid operation risk scenario sequences.

[0288] The formula for calculating the evaluation value is as follows:

[0289] In the formula, Ε is the expectation operator; T is the visit step size; Q(v t ) for accessing scene v t Reward value; N(v) p N(v) represents the number of times the parent scene is visited; c ) represents the number of times the sub-scene is accessed; V pen This is the penalty term for violating the constraints; γ is a constant coefficient, and γ > 0.

[0290] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0291] The loss function for a Monte Carlo decision tree is:

[0292] In the formula, m represents the number of scenes; when the current scene is located in the target scene set, then l i If the value is 1, then l is not in the target scene set. i p is 0; 1i p 2i These are the probability values ​​that the current scene belongs to the target scene set; r key The weight representing the percentage of scenes located in the target scene set; The percentage weight of the number of scenes that are not located in the target scene set.

[0293] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0294] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0295] Constraints are set to apply to the Monte Carlo decision tree. The constrained Monte Carlo decision tree is then used to partition the evaluation values ​​of at least one power grid operation risk scenario, yielding the processing results. The constraints include an evaluation penalty term, expressed as: ∑Vpen,i =V pen,bl +V pen,op +V pen,pf ;

[0296] In the formula, V pen,bl V pen,op V pen,pf These are operational constraint penalties, safety and stability penalties, and power balance penalties, respectively.

[0297] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0298] Obtain preset risk assessment indicators, including:

[0299] Obtain the power grid's loss of load risk value, voltage over-limit risk value, and line over-limit risk value;

[0300] Based on the underload risk value, voltage over-limit risk value, and line over-limit risk value, a preset risk value assessment index is obtained according to a preset weighting ratio; wherein, the expression for the underload risk value is:

[0301] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; C (E k ) represents the system state E k The severity of the consequences of load loss; P ci ND represents the load shedding amount at load node i, and ND represents the number of load nodes;

[0302] The expression for the voltage over-limit risk value is:

[0303] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; v (E k ) represents the system state E k The severity of the consequences of exceeding the voltage limit, where NI is the number of system nodes;

[0304] The expression for the risk value of exceeding the limit on the line is:

[0305] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; s (E k ) represents the system state E k The severity of the consequences of voltage exceeding the limit; NL is the number of system nodes, P LiP is the active power flowing through branch i. Limax Let be the maximum active power allowed to flow through branch i.

[0306] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:

[0307] Acquire power grid operation status data and meteorological monitoring data;

[0308] Based on the power grid operation status data and the meteorological monitoring data, a power grid section dataset is generated;

[0309] During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset.

[0310] The generative adversarial network model is used to obtain at least one type of power grid operation risk data;

[0311] Based on the at least one type of power grid operation risk data, a power grid operation risk scenario set is generated, wherein the power grid operation risk scenario set includes at least one power grid operation risk scenario.

[0312] Using a time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid forward-looking scenario sequence;

[0313] During the online operation phase of the power grid, pre-set risk assessment indicators and current power grid operation scenario information are obtained;

[0314] Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

[0315] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0316] The time-series simulation method includes Monte Carlo decision trees; using the time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid prospective scenario sequence, including:

[0317] The weighted cross-entropy function is used as the loss function of the Monte Carlo decision tree, and the evaluation value of at least one power grid operation risk scenario is obtained through the Monte Carlo decision tree.

[0318] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results;

[0319] A parallel inference architecture is adopted, and Monte Carlo decision trees are used to perform multi-threaded parallel inference calculations on the processing results to generate at least one power grid prospective scenario sequence.

[0320] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0321] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0322] By using a preset evaluation value calculation formula, the maximum expected evaluation value for each power grid operation risk scenario is obtained.

[0323] Based on maximizing the expected evaluation value, at least one power grid operation risk scenario is divided and processed to obtain a processing result with multiple power grid operation risk scenario sequences.

[0324] The formula for calculating the evaluation value is as follows:

[0325] In the formula, Ε is the expectation operator; T is the visit step size; Q(v t ) for accessing scene v t Reward value; N(v) p N(v) represents the number of times the parent scene is visited; c ) represents the number of times the sub-scene is accessed; V pen This is the penalty term for violating the constraints; γ is a constant coefficient, and γ > 0.

[0326] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0327] The loss function for a Monte Carlo decision tree is:

[0328] In the formula, m represents the number of scenes; when the current scene is located in the target scene set, then l i If the value is 1, then l is not in the target scene set. i p is 0; 1i p 2i These are the probability values ​​that the current scene belongs to the target scene set; r key The weight representing the percentage of scenes located in the target scene set; The percentage weight of the number of scenes that are not located in the target scene set.

[0329] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0330] The evaluation values ​​of at least one power grid operation risk scenario are divided and processed to obtain the processing results, including:

[0331] Constraints are set to apply to the Monte Carlo decision tree. The constrained Monte Carlo decision tree is then used to partition the evaluation values ​​of at least one power grid operation risk scenario, yielding the processing results. The constraints include an evaluation penalty term, expressed as: ∑V pen,i =V pen,bl +V pen,op +V pen,pf ;

[0332] In the formula, V pen,bl V pen,op V pen,pf These are operational constraint penalties, safety and stability penalties, and power balance penalties, respectively.

[0333] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:

[0334] Obtain preset risk assessment indicators, including:

[0335] Obtain the power grid's loss of load risk value, voltage over-limit risk value, and line over-limit risk value;

[0336] Based on the underload risk value, voltage over-limit risk value, and line over-limit risk value, a preset risk value assessment index is obtained according to a preset weighting ratio; wherein, the expression for the underload risk value is:

[0337] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; C (E k ) represents the system state E k The severity of the consequences of load loss; P ci ND represents the load shedding amount at load node i, and ND represents the number of load nodes;

[0338] The expression for the voltage over-limit risk value is:

[0339] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E k The probability of S; v (E k ) represents the system state E k The severity of the consequences of exceeding the voltage limit, where NI is the number of system nodes;

[0340] The expression for the risk value of exceeding the limit on the line is:

[0341] In the formula, K represents the number of simulated states; P(E) k ) represents the system state E kThe probability of S; s (E k ) represents the system state E k The severity of the consequences of voltage exceeding the limit; NL is the number of system nodes, P Li P is the active power flowing through branch i. Limax Let be the maximum active power allowed to flow through branch i.

[0342] Those skilled in the art will recognize that the units of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of the invention.

[0343] In the embodiments provided by the present invention, it should be understood that the division of units is only a logical functional division. In actual implementation, there may be other division methods, such as multiple units can be combined into one unit, one unit can be split into multiple units, or some features can be ignored.

[0344] Furthermore, the functional units in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0345] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0346] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A power grid forward looking risk assessment method, characterized in that, Includes the following steps: Acquire power grid operation status data, meteorological monitoring data, and historical wind turbine power data; construct a wind turbine power time series based on historical wind turbine power data; The missing data in the wind turbine power time series is obtained, and the missing data is filled in by a piecewise linear regression model to obtain the target wind turbine power time series. Based on the target wind turbine power time series, an extreme learning machine model, a long short-term memory neural network model, and a temporal convolutional network model are trained respectively. A multi-objective prediction model is trained based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model. The Pareto optimal solution set of the combined weight vector of the multi-objective prediction model is obtained by using the multi-objective Harris Eagle algorithm. The deep deterministic strategy gradient method is used to obtain the dynamic optimal weights of each model in the multi-objective prediction model based on the Pareto optimal solution set. Based on each model in the multi-objective prediction model and its corresponding dynamic optimal weights, the prediction results of wind power are obtained. A power grid section dataset is generated based on the power grid operation status data, the meteorological monitoring data, and the wind power prediction results. During the offline phase of the power grid, a generative adversarial network model is constructed based on the power grid section dataset. The generative adversarial network model is used to obtain at least one type of power grid operation risk data; Based on the at least one type of power grid operation risk data, a power grid operation risk scenario set is generated, wherein the power grid operation risk scenario set includes at least one power grid operation risk scenario. Using a time-series simulation method, at least one power grid operation risk scenario is extrapolated and calculated to generate at least one power grid forward-looking scenario sequence; During the online operation phase of the power grid, pre-set risk assessment indicators and current power grid operation scenario information are obtained; Based on the preset risk assessment indicators and the current power grid operation scenario information, obtain the power grid operation risk assessment results and match them to obtain the corresponding power grid forward operation scenario sequence.

2. The method of grid forward looking risk assessment according to claim 1, characterized in that, Obtain missing data from the wind turbine power time series, including: The wind turbine power data that meets the missing value preset conditions in the wind turbine power time series is identified as missing value data; the missing value preset conditions include that the value of the wind turbine power data is zero, and the change value of the wind turbine power data relative to the wind turbine power data of the previous time series is greater than a preset threshold.

3. The method of grid forward looking risk assessment according to claim 1, characterized in that, Before imputing the missing data using a piecewise linear regression model, the method further includes: The missing data is interpolated and filled to obtain the intermediate wind turbine power time series; An initial piecewise linear regression model is trained based on the intermediate wind turbine power time series to optimize the parameters of the initial piecewise linear regression model until a fully trained piecewise linear regression model is obtained.

4. The method of grid forward risk assessment according to claim 3, wherein, An initial piecewise linear regression model was trained based on the power time series of intermediate wind turbines, including: Obtain the exponentially weighted moving average and moving median features of the intermediate wind turbine power time series; The intermediate wind turbine power time series is processed using the BiLSTM model to obtain the meta-features of the intermediate wind turbine power time series. An initial piecewise linear regression model is trained based on the meta-features, the indexed weighted moving average features, and the moving median features to optimize the parameters of the initial piecewise linear regression model until a fully trained piecewise linear regression model is obtained.

5. The method of grid forward looking risk assessment as claimed in claim 1, wherein, Based on the target wind turbine power time series, an Extreme Learning Machine model, a Long Short-Term Memory Neural Network model, and a Temporal Convolutional Network model are trained respectively, including: The CEEMDAN sequence decomposition method was used to decompose the time series data of historical wind power to obtain multiple first subsequence data. Using the multiple first subsequence data, an Extreme Learning Machine model, a Long Short-Term Memory Neural Network model, and a Temporal Convolutional Network model are trained simultaneously.

6. The method of grid forward looking risk assessment according to claim 1, wherein, Based on each model and its corresponding dynamic optimal weight in the multi-objective prediction model, the prediction results of short-term wind power are obtained, including: Obtain time-series data of the target wind power; The CEEMDAN sequence decomposition method was used to decompose the time series data of the target wind power to obtain multiple second subsequence data. By using each model in the multi-objective prediction model, the multiple second sub-sequence data are simultaneously traversed and predicted to obtain the corresponding first prediction result, second prediction result and third prediction result; Based on the corresponding dynamic optimal weights of each model, the first prediction result, the second prediction result, and the third prediction result are weighted and calculated to obtain the prediction result of short-term wind power.

7. The method of grid forward looking risk assessment according to claim 1, characterized in that, Based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model, a multi-objective prediction model is trained, including: Obtain time-series data of the test wind power; The CEEMDAN sequence decomposition method was used to decompose the time series data of the test wind power to obtain multiple third subsequence data. Based on the Extreme Learning Machine model, the Long Short-Term Memory Neural Network model, and the Temporal Convolutional Network model, the multiple third subsequence data are simultaneously traversed and tested to obtain the corresponding first test result, second test result, and third test result. The mean square error and standard deviation of the actual short-term wind power are obtained between the first test result, the second test result and the third test result, respectively. A multi-objective prediction model is trained using the minimum mean squared error and the minimum standard deviation as optimization objectives.

8. The method of grid forward looking risk assessment according to claim 1, characterized in that, The weights and biases of the multi-objective prediction model are iteratively updated using the agent's action vectors, including: Based on the principle of minimum average absolute percentage error, static weight coefficients are selected from the Pareto optimal solution set using the agent's action vectors. The reference error is obtained based on the static weighting coefficients and the actual short-term wind power. Analyze the correlation between the minimum mean square error, the minimum standard deviation, and the reference error; Obtain the reward function based on the correlation; The weights and biases of the multi-objective prediction model are iteratively updated using a reward function.

9. A power grid operation risk assessment apparatus characterized by comprising: The device includes: The data acquisition module is used to acquire power grid operation status data and meteorological monitoring data; The data integration module is used to generate a power grid section dataset based on power grid operation status data and meteorological monitoring data; The modeling module is used to construct a generative adversarial network model based on a power grid section dataset during the offline phase of the power grid. By generating adversarial network models, at least one type of power grid operation risk data can be obtained; The processing module is used to generate a set of power grid operation risk scenarios based on at least one type of power grid operation risk data, wherein the set of power grid operation risk scenarios contains at least one power grid operation risk scenario; The simulation and evaluation module is used to perform simulation calculations on at least one power grid operation risk scenario using time-series operation simulation methods, and generate at least one power grid forward-looking scenario sequence. The data acquisition module is also used to acquire preset risk assessment indicators and current power grid operation scenario information during the online operation phase of the power grid; The online application module is used to obtain the power grid operation risk assessment results based on preset risk assessment indicators and current power grid operation scenario information, and to match and obtain the corresponding power grid forward operation scenario sequence.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device containing the computer-readable storage medium to perform the power grid forward risk assessment method according to any one of claims 1 to 8.

Images