Offshore wind power submarine cable transmission power probability safety assessment method and system

By employing multidimensional data governance and deep reinforcement learning models, the problem of insufficient data processing and prediction in the safety assessment of offshore wind power transmission power via submarine cables was solved, achieving high-precision safety assessment and dynamic adaptability, thereby improving the reliability and safety of offshore wind power systems.

CN121920853APending Publication Date: 2026-04-24GUANGXI POWER GRID CORP +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202511866197.1
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-12-11
Publication Date
2026-04-24

AI Technical Summary

Technical Problem

Existing technologies for assessing the power transmission capacity of offshore wind power cables suffer from inaccurate data preprocessing, insufficient prediction models, and a lack of real-time updates to safety boundaries. These issues result in insufficient assessment accuracy and dynamic adaptability, making it difficult to cope with the dynamic operating conditions of complex offshore environments.

Method used

We employ multidimensional data governance, attention mechanism-bidirectional long short-term memory neural network and deep reinforcement learning model, combined with multidimensional data governance, dynamic prediction modeling and adaptive decision optimization, to remove outlier data through improved Z-score and cluster detection, fill in blank values ​​with K-nearest neighbor algorithm, construct Att-BiLSTM model for wind power output prediction, and dynamically characterize submarine cable power safety margin through deep reinforcement learning framework of PPO algorithm.

Benefits of technology

This achievement represents a breakthrough in both accuracy and robustness for the safety assessment of offshore wind power transmission power via submarine cables. It reduces data noise, improves the accuracy of wind power output prediction, and dynamically characterizes the safety margin of submarine cable power, providing full-chain technical support for high-reliability grid connection and transmission safety monitoring of offshore wind power.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN121920853A_ABST
    Figure CN121920853A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of power system reliability evaluation, in particular to an offshore wind power submarine cable transmission power probability safety evaluation method and system. According to the method, multi-dimensional data management, dynamic prediction modeling and self-adaptive decision optimization are creatively fused, so that double breakthrough of safety assessment precision and robustness is realized. Abnormal data removal, blank value filling and normalization are carried out on the data in the data level, data noise is reduced, and key features are completely improved; in a prediction level, an Att-BiLSTM model is constructed, and wind power output prediction is realized through bidirectional time sequence feature extraction and attention weight dynamic distribution; in the aspect of probability safety assessment, a deep reinforcement learning framework based on a PPO algorithm is designed, the submarine cable power safety margin is dynamically described, full-chain technical support is provided for offshore wind power high-reliability grid connection and power transmission safety monitoring, and the method has wide industrial application prospects.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of power system reliability assessment technology, and in particular to a method and system for probabilistic security assessment of offshore wind power transmission power via submarine cables. Background Technology

[0002] As the global energy structure accelerates its transition to cleaner energy, offshore wind power, as an important form of renewable energy, has become a key strategic focus for many countries in its large-scale development and efficient grid connection. However, offshore wind power output is affected by a complex interplay of environmental factors, including the spatiotemporal volatility of wind speed, the dynamic impact of waves on turbine foundations, and equipment performance degradation due to salt spray corrosion. These factors combined result in highly random and nonlinear characteristics in the power transmitted via submarine cables. Statistics show that the annual output fluctuation of a typical offshore wind farm can reach 40%-60% of its rated capacity, and the instantaneous power fluctuation rate can exceed 70% under extreme weather conditions, leading to frequent over-limit risks for the transmission system. Traditional assessment methods, based on static safety margin thresholds or simplified probabilistic models, are ill-suited to the dynamic operating conditions in the complex offshore environment.

[0003] Existing technologies for assessing the safety of submarine cable power transmission power suffer from systemic flaws: Data preprocessing relies on a single anomaly detection algorithm (such as the 3σ criterion), making it difficult to identify environmentally related faults like salt spray corrosion; linear interpolation for filling missing values ​​ignores the nonlinear relationship between wind speed and power, leading to data errors amplified by over 30%; prediction models using structures like unidirectional LSTM fail to quantify the chain effect of wave height, vibration, and power attenuation, while static safety boundaries do not consider real-time parameters such as cable aging, resulting in a 22% misjudgment rate in summer and annual wind curtailment losses exceeding 8 million kWh; uncertainty modeling assumes wind power errors follow a single Gaussian distribution, ignoring the bimodal characteristics of typhoon seasons and the spatiotemporal correlation between source and load (such as tidal-wind power anti-peak coupling), causing a 35% deviation in scenario generation and severely overestimating system safety margins. These problems collectively lead to insufficient assessment accuracy and dynamic adaptability, hindering reliable grid connection of offshore wind power.

[0004] Therefore, a method and system for probabilistic safety assessment of offshore wind power transmission power via submarine cables is needed. Summary of the Invention

[0005] To address the problems in existing technologies, this invention provides a method and system for probabilistic safety assessment of offshore wind power submarine cable transmission power. By innovatively integrating multi-dimensional data governance, dynamic predictive modeling, and adaptive decision optimization, it achieves a dual breakthrough in both the accuracy and robustness of safety assessment. The specific technical solution is as follows: A method for probabilistic security assessment of offshore wind power transmission power via submarine cables includes the following steps: Step S1: Collect measured offshore wind power data and perform data preprocessing, including removing outliers, filling in blanks, and normalization. Measured offshore wind power data includes temperature, wind speed, and offshore wind power output data. The preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y The power output of offshore wind power; Step S2: Construct a training model of attention mechanism-bidirectional long short-term memory neural network, and train it with preprocessed measured data to obtain the optimal regression model for predicting offshore wind power output. Step S3: Collect wind power data predicted from environmental data, and construct an uncertainty dataset by combining wind power prediction error and load fluctuation data; Step S4: Construct a deep reinforcement learning model to characterize the power transmission safety space, and train the deep reinforcement learning model with an uncertainty dataset to characterize the space that can reflect the safe operation of the submarine cable transmission power. Step S5 involves large-scale sampling of source-load data with randomness, evaluation within a space reflecting the safe operation of submarine cable transmission power, and provision of safety indicators for offshore wind power transmission power of submarine cables.

[0006] Preferably, the removal of outlier data in step S1 specifically includes identifying and removing outliers using an improved Z-score method and detecting outliers in multidimensional data based on clustering. The identification and removal of outliers using the improved Z-score method is as follows: For time series data X={ x 1, x 2,…, x n} Calculate the moving median of the time series data X. M t and moving absolute median M ad The anomaly detection threshold is: ; If the i-th data point x i If the value exceeds this range, it is considered an outlier and removed. The following are specific examples of outlier detection based on clustering in multidimensional data: Clustering algorithms are used to identify outliers in sparse regions of multidimensional data. The formula is as follows: ; In the formula: D core For data core density, xData collected from actual sampling; ϵ The neighborhood radius is defined as the density of samples below a threshold, which are considered anomalies. The multidimensional data includes wind speed, power, and temperature.

[0007] Preferably, in step S1, filling in the blank values ​​specifically uses the K-nearest neighbor algorithm, which selects the nearest neighbor sample for weighted filling based on Euclidean distance. The calculation formula is as follows: ; in, These are the missing items to be filled. Let be the corresponding feature value of the j-th nearest neighbor sample; For the first j The weights of the nearest neighbor samples are calculated based on the reciprocal of the Euclidean distance. k The number of nearest neighbor samples selected; Indicates the data value after filling. The corresponding feature value of the j-th nearest neighbor sample The Euclidean distance; This is the data value after filling in the missing parts.

[0008] Preferably, step S2 specifically includes the following steps: (1) Constructing a training model for an attention-based bidirectional long short-term memory neural network, including an input layer, an LSTM layer, an output layer, and an attention mechanism; specifically as follows: 1) The input layer receives normalized multidimensional time-series data P X ∈R T×D , T The total length of the time series. D For feature dimensions; 2) The LSTM layer extracts temporal features bidirectionally. The hidden states of the forward and backward LSTMs are h→ t and h← t After piecing them together, we get: ; in, H h is the number of hidden units in a single-layer LSTM; where: h t for t The hidden state vector at time step; This represents the hidden state of the forward LSTM at time t; This represents the hidden state of the backward LSTM at time t. 3) The attention mechanism uses trainable weight parameters W a ∈R 2H×H and v a ∈R H The attention weights are calculated as follows: ; The weighted vector is: ; In the formula: The attention-weighted context vector; α t for t Attention weight at any given moment; h t for t The hidden state vector at time step; The output layer is mapped to the predicted value through a fully connected layer, resulting in: ; in, W o ∈R H×1 Let b be the weight matrix. o For bias terms of fully connected layers; (2) The optimal regression model for predicting offshore wind power output was obtained by training the preprocessed measured data. The Huber loss function was used during training, as follows: ; In the formula: y This represents the actual power output of offshore wind power. ŷ The predicted offshore wind power output value is given by the model. d For the threshold; The optimizer uses the AdamW optimizer, which introduces weight decay to prevent overfitting, as detailed below: ; In the formula: l The attenuation coefficient; i t , i t+1 The first t , t+ Model parameters for one iteration; or The learning rate; and These are the first-order moment estimates and second-order moment estimates after bias correction, respectively; e adam It is a numerically stable term; During regularization, a Dropout layer is added after the BiLSTM layer to randomly drop out neurons at a ratio of k. Training is complete when the evaluation index of the regression model for predicting offshore wind power output reaches a preset threshold, and the optimal regression model for predicting offshore wind power output is obtained.

[0009] Preferably, step S3 specifically includes the following steps: (1) Wind power output prediction error modeling: Wind power prediction errors are caused by the superposition of multiple Gaussian distributions, with the following probability density function: ; In the formula: p ( e (This represents the wind power prediction error) e The probability density function; K The number of Gaussian components; π k For the first k The mixture weights of Gaussian components satisfy Σ π k =1; N (ε| m k , s 2k) is the first k Gaussian components m k The mean, s 2k is the variance; Based on historical error data, the distribution is estimated using the Gaussian kernel function: ; In the formula: Let be the estimated probability density function; N This represents the number of historical error samples. e i For the first i One historical error sample; h For bandwidth parameters; (2) Load fluctuation modeling: Seasonal decomposition is used to break down load data into trend terms. T t Seasonal items S t and residuals R t The details are as follows: ; In the formula: P Load,t for t The load value at time t; the residuals follow a normal distribution; (3) Use the trained optimal regression model for predicting offshore wind power output to predict wind power output in future periods. Randomly sample wind power prediction errors from the wind power output prediction error model. e t Random wind power output is generated, as follows: ; A large number of scenarios are generated to form an uncertainty dataset. P Load , Pseason g ].

[0010] Preferably, step S4 specifically includes the following steps: (1) The state space S used space This includes electrical measurement parameters, environmental parameters, and equipment status parameters. Specifically, the electrical measurement parameters include submarine cable transmission power. P cable Node voltage V Phase angle i Uncertain loads and wind power P Load 、 Pseason g Environmental parameters include wind speed. v wind Equipment status parameters include the aging coefficient of the submarine cable. α aging ; The motion space A used space Including adjusting the incremental Δ of wind power P ∈[−Δ P max , Δ P max ],satisfy: ; In the formula: P cable,t This represents the transmission power of the submarine cable at time t. (2) The reward function includes security rewards. r safe Economic rewards r econ Equipment lifespan bonus r life The details are as follows: ; In the formula: w 1 represents the safety reward weighting coefficient; P mid This is the reference value for the rated transmission power of the submarine cable; s safe This is the sensitivity variance coefficient; ; In the formula: w 2 represents the economic reward weighting coefficient; R cable This refers to the economic coefficient for losses. ; In the formula:w 3 represents the weighting coefficient for equipment lifespan bonus; Δ P This refers to power fluctuations. The total reward is: ; In the formula, r t for t Instant rewards for each moment; (3) Dominance function AGAE t The generalized dominance method is used to estimate the equilibrium bias and variance, as follows: ; In the formula: AGAE t for t Generalized advantage estimate at time; d t+1 This refers to timing difference error; c Discount factor; l GAE These are GAE parameters used to balance bias and variance; d t = r t +γ V ( s t+1 )− V ( s t ), V ( s t ) is the state s t Value function estimation; l Number of steps; Smooth policy updates are ensured through KL divergence constraints, as detailed below: ; Where: KL( π θold || π θ The KL divergence between the old and new strategies is used to measure the difference in strategy distributions. π θold The parameters are the old strategy parameters before the update; π θ For the updated new strategy parameters; d KL The constraint threshold for KL divergence.

[0011] Preferably, step S5 specifically includes the following steps: (1) Scene generation through Monte Carlo sampling, specifically including: Uncertainty dataset [P Load , Pseason g Using [[wind power output, load demand]] as the sampling source, N scenarios are uniformly generated across [wind power output, load demand] through Latin hypercube sampling; the sampling density is increased for high-risk areas, with the following weights: ; In the formula: x i ∈[ P Load , Pseason g ], p ( x ) represents the true joint probability distribution of source-charge power; q ( x ) represents the assumed distribution of uniform sampling; (2) Conduct dynamic security boundary assessment, including real-time security margin, security probability, and limit violation risk indicators, as detailed below: ① The calculation method for real-time safety margin is as follows: The safety threshold is dynamically adjusted based on the output of the DRL model, as follows: ; In the formula: This represents the dynamic safe power limit under state s; Let be the relevant safety margin function under state s, which is obtained by the trained DRL model under the current state s.

[0012] ②The calculation method for the probability of safety is as follows: For N sampled scenarios, count the number of scenarios that satisfy the safety constraints: ; In the formula: I(●) is an indicator function, which is 1 when the condition is met, and 0 otherwise; ③ The over-limit risk indicators include average over-limit power and maximum over-limit power, as detailed below: Average over-limit power The calculation is as follows: ; In the formula: N exceed This refers to the number of scenarios that exceed the limit.

[0013] Maximum over-limit power The calculation is as follows: .

[0014] A probabilistic safety assessment system for offshore wind power submarine cable transmission power, employing the method described herein, includes: The data processing module is used to collect measured offshore wind power data and perform data preprocessing, including removing outliers, filling in blank values, and normalization. The measured offshore wind power data includes temperature, wind speed, and offshore wind power output data. The preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y The power output of offshore wind power; The prediction model building module is used to build a training model of attention mechanism-bidirectional long short-term memory neural network, and to train the optimal regression model for predicting offshore wind power output by combining it with preprocessed measured data. The dataset construction module is used to collect wind power data for environmental data prediction and construct an uncertainty dataset by combining wind power prediction error and load fluctuation data. The reinforcement learning model building module is used to build a deep reinforcement learning model that characterizes the power transmission safety space. The deep reinforcement learning model is trained in combination with an uncertainty dataset to characterize the space that can reflect the safe operation of submarine cable transmission power. The safety assessment module is used to perform large-scale sampling of source-load data with randomness, conduct assessments in a space that reflects the safe operation of submarine cable transmission power, and provide safety indicators for the transmission power of offshore wind power submarine cables.

[0015] A computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to perform the aforementioned method for probabilistic security assessment of offshore wind power submarine cable transmission power.

[0016] A processor for running a program, wherein the program executes the aforementioned method for probabilistic security assessment of offshore wind power transmission power via submarine cables.

[0017] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention achieves a dual breakthrough in the accuracy and robustness of safety assessment by innovatively integrating multi-dimensional data governance, dynamic predictive modeling, and adaptive decision optimization. At the data level, outlier removal, gap filling, and normalization reduce data noise and significantly enhance key features. At the prediction level, an Att-BiLSTM model is constructed, using bidirectional temporal feature extraction and dynamic attention weight allocation to predict wind power output. In probabilistic safety assessment, a deep reinforcement learning framework based on the PPO algorithm is designed to dynamically characterize the power safety margin of submarine cables, providing full-chain technical support for high-reliability grid connection and transmission safety monitoring of offshore wind power, and possessing broad prospects for industrial application. Attached Figure Description

[0018] 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.

[0019] Figure 1 This is a flowchart of the method of the present invention.

[0020] Figure 2 This is a graph showing wind power data.

[0021] Figure 3 This is a diagram of a typical 220kV offshore transmission system.

[0022] Figure 4 The model prediction results and error graph are shown.

[0023] Figure 5 To enhance the learning and training process diagram.

[0024] Figure 6 This is a diagram showing the results of a safety assessment.

[0025] Figure 7 This is a system schematic diagram of the present invention. Detailed Implementation

[0026] 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.

[0027] 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.

[0028] 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.

[0029] 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.

[0030] Example 1: like Figure 1 As shown in the figure, this embodiment provides a method for probabilistic security assessment of offshore wind power transmission power via submarine cables, including the following steps: Step S1: Collect measured offshore wind power data and perform data preprocessing, including outlier removal, missing value filling, and normalization, to ensure data quality meets the requirements for subsequent model training. Measured offshore wind power data includes temperature, wind speed, and offshore wind power output data; the preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y This refers to the power output of offshore wind power.

[0031] Outlier removal specifically includes identifying and removing outliers using an improved Z-score method and detecting outliers in multidimensional data based on clustering: ① The improved Z-score method is used for anomaly detection in single-dimensional time series data. It is suitable for single feature data with time-series characteristics, such as temperature, wind speed, and offshore wind power output. Details are as follows: For time series data X={ x 1, x 2,…, x n} (where X can be a temperature series, wind speed series, or power series), calculate the moving median of the time series data X. M t and moving absolute median M ad The anomaly detection threshold is: ; If the i-th data point x i If the value exceeds this range, it is considered an outlier and removed. ② Based on DBSCAN clustering, outliers in multidimensional data are detected by fusing the three-dimensional features of wind speed, power, and temperature for joint anomaly detection. N ∈ℝ n×3 (Including wind speed, power, and temperature) The DBSCAN clustering algorithm was used to mark samples in sparse regions as outliers, using the following formula: ; In the formula: Dcore For data core density, x This refers to the multidimensional data collected from actual sampling. e The neighborhood radius is defined as the density of samples below a certain threshold, which are considered abnormal.

[0032] Specifically, filling in blank values ​​uses the K-nearest neighbor algorithm, which selects the nearest neighbor sample for weighted filling based on Euclidean distance. The calculation formula is as follows: ; in, These are the missing items to be filled. Let be the corresponding feature value of the j-th nearest neighbor sample; For the first j The weights of the nearest neighbor samples are calculated based on the reciprocal of the Euclidean distance. k The number of nearest neighbor samples selected; Indicates the data value after filling. The corresponding feature value of the j-th nearest neighbor sample The Euclidean distance; This is the data value after filling in the missing parts.

[0033] Based on the above output process, the multidimensional data is normalized using the following method to avoid the influence of the dimensions of different parameters on the model's prediction accuracy.

[0034] Z-score standardization: Standardizes temperature data.

[0035] ; In the formula: m The mean, s The standard deviation is denoted as .

[0036] For features with clearly defined physical boundaries, such as wind power output and submarine cable transmission power, a maximum-minimum normalization method is used to map the data to the [0, 1] interval, as shown in the formula: ; In the formula: x max and x min These are the maximum and minimum values ​​of the feature, respectively.

[0037] Quantile transformation: For features with long-tailed distributions, such as wind speed, quantile transformation is used to make them conform to a uniform distribution.

[0038] ; in, F The distribution function of the measured offshore wind power output data is fitted. G This is the original empirical distribution function.

[0039] Step S2 involves constructing a training model for an attention mechanism-bidirectional long short-term memory neural network, and training it using preprocessed measured data to obtain the optimal regression model for predicting offshore wind power output. This specifically includes the following steps: (1) Construct a training model of an attention mechanism (Attention, Att)-Bidirectional Long Short-Term Memory Neural Network (BiLSTM), including an input layer, an LSTM layer, an output layer, and an attention mechanism; divide the preprocessed wind power data samples into training and testing sets in an 8:2 ratio. Details are as follows: 1) The input layer receives normalized multidimensional time-series data P X ∈R T×D , T The total length of the time series. D For feature dimensions; 2) The LSTM layer extracts temporal features bidirectionally. The hidden states of the forward and backward LSTMs are h→ t and h← t After piecing them together, we get: ; in, H h is the number of hidden units in a single-layer LSTM; where: h t for t The hidden state vector at time step; This represents the hidden state of the forward LSTM at time t; This represents the hidden state of the backward LSTM at time t. 3) The attention mechanism uses trainable weight parameters W a ∈R 2H×H and v a ∈R H Calculate attention weights, where W a This is the weight matrix of the attention layer, used to divide 2 H Bidirectional hidden state of a dimension h t Mapped to H 3D space; v a This is the attention score vector, used to calculate the importance score at each time step. These two parameters are automatically learned and optimized through backpropagation during model training. Specifically: ; The weighted vector is: ; In the formula: The attention-weighted context vector;α t for t Attention weight at any given moment; h t for t The hidden state vector at time step; The output layer is mapped to the predicted value through a fully connected layer, resulting in: ; in, W o ∈R H×1 Let b be the weight matrix. o For bias terms of fully connected layers; (2) The optimal regression model for predicting offshore wind power output was obtained by training the preprocessed measured data. The Huber loss function was used during training, which is robust to outliers, as follows: ; In the formula: y This represents the actual power output of offshore wind power. ŷ The predicted offshore wind power output value is given by the model. d The threshold parameter is used when the prediction error is less than... d When using squared loss, greater than d Linear loss is used at this time.

[0040] The optimizer uses the AdamW optimizer, which introduces weight decay to prevent overfitting, as detailed below: ; In the formula: l The attenuation coefficient; i t , i t+1 The first t , t+ Model parameters for one iteration; or The learning rate; and These are the first-order moment estimates and second-order moment estimates after bias correction, respectively; e adam This is a numerically stable term to prevent division by zero errors.

[0041] During regularization, a Dropout layer is added after the BiLSTM layer to randomly drop neurons at a ratio of k=0.15; Training is complete when the evaluation index of the regression model for predicting offshore wind power output reaches a preset threshold, and the optimal regression model for predicting offshore wind power output is obtained.

[0042] In this embodiment, to evaluate the predictive performance of the models, the root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE) are used to evaluate the predictive performance of various models.

[0043] (1) Root mean square error (RMSE): Represents the degree of dispersion between the actual value and the predicted value.

[0044] ; In the formula: P Gi and P Pi These are the actual offshore wind power value and the predicted value, respectively.

[0045] (2) Mean absolute error (MAE): represents the magnitude of the average error between the actual value and the predicted value.

[0046] ; (3) Mean Absolute Percentage Error (MAPE): Measures the percentage error between the predicted value and the actual value.

[0047] ; By combining the above data and evaluation indicators, the model is trained and evaluated to obtain the optimal offshore wind power output prediction model.

[0048] Step S3 involves collecting wind power data predicted from environmental data and constructing an uncertainty dataset by combining wind power prediction errors and load fluctuation data. This specifically includes the following steps: 1) Wind power output prediction error modeling: This invention employs a Gaussian mixture model (GMM), where the wind power prediction error is a superposition of multiple Gaussian distributions, with the following probability density function: ; In the formula: p ( e (This represents the wind power prediction error) e The probability density function; K The number of Gaussian components; π k For the first k The mixture weights of Gaussian components satisfy Σ π k =1; N (ε| m k , s 2k) is the first k Gaussian components mk The mean, s 2k is the variance; Nonparametric kernel density estimation (KDE): Based on historical error data, the distribution is estimated using a Gaussian kernel function. ; In the formula: Let be the estimated probability density function; N This represents the number of historical error samples. e i For the first i One historical error sample; h The bandwidth parameter controls the smoothness of the estimation. This invention takes... h =1.06 σ^N −1 / 5 ; 2) Load fluctuation modeling Seasonal decomposition is used to break down load data into trend terms. T t Seasonal items S t and residuals R t The details are as follows: ; In the formula: P Load,t for t Load value at time; residual term R t It follows a normal distribution.

[0049] (3) The specific steps for generating a dataset on uncertainty in new energy processing based on the above model are as follows: ① Use a trained optimal regression model (Att-BiLSTM model) to predict wind power output in future periods. ; ② Randomly sample wind power prediction errors from the wind power output prediction error model based on the GMM model. e t Generate random wind power output: ; Based on the above conditions, a massive number of group scenarios are generated, forming an uncertainty dataset. P Load , Pseason g ].

[0050] Step S4: Construct a deep reinforcement learning model to characterize the power transmission safety space. Train the deep reinforcement learning model using an uncertainty dataset to characterize the space that reflects the safe operation of the submarine cable's power transmission. The random wind power output generated in S3 will be used as the basis for this process. Pseason g and random load demand P Load As external input to the DRL model, it is used to construct the state space and simulate power grid operating conditions. Specifically, it includes the following steps: (1) The state space S used space This includes electrical measurement parameters, environmental parameters, and equipment status parameters. Specifically, the electrical measurement parameters include submarine cable transmission power. P cable Node voltage V Phase angle I, Uncertain loads and wind power P Load 、 Pseason g Environmental parameters include wind speed, etc. v wind Equipment status parameters include the aging coefficient of the submarine cable. α aging ; The motion space A used space Including adjusting the incremental Δ of wind power P ∈[−Δ P max , Δ P max ], Δ P max To achieve the maximum amplitude in a single step, the following conditions must be met: ; In the formula: P cable,t This represents the transmission power of the submarine cable at time t. (2) The reward function includes security rewards. r safe Economic rewards r econ Equipment lifespan bonus r life The security bonus specifically uses an exponential function to enhance sensitivity in the limit violation penalty; the economic bonus is that transmission loss cost is proportional to the square of power; and the equipment lifespan bonus specifically reduces the impact of frequent power fluctuations on the lifespan of submarine cables. Details are as follows: ; In the formula: w 1 represents the safety reward weighting coefficient; P mid This is the reference value for the rated transmission power of the submarine cable; s safe The coefficient of variation is used to control the steepness of the penalty. ; In the formula:w 2 represents the economic reward weighting coefficient; R cable This refers to the economic coefficient for losses. ; In the formula: w 3 represents the weighting coefficient for equipment lifespan bonus; The total reward is: ; In the formula, r t for t Instant rewards for each moment; (3) Dominance function AGAE t The generalized dominance method is used to estimate the equilibrium bias and variance, as follows: ; In the formula: AGAE t for t Generalized advantage estimate at time; d t+1 This refers to timing difference error; c The discount factor is set to 0.99 in this invention. l GAE This is the GAE smoothing parameter, used to balance bias and variance; d t = r t +γ V ( s t+1 )− V ( s t ), V ( s t ) is the state s t Value function estimation; l Number of steps; Smooth policy updates are ensured through KL divergence constraints, as detailed below: ; Where: KL( π θold || π θ The KL divergence between the old and new strategies is used to measure the difference in strategy distributions. π θold The parameters are the old strategy parameters before the update; π θ For the updated new strategy parameters; d KLThe constraint threshold for KL divergence is set to 0.01 in this invention to limit the magnitude of each policy update and prevent training instability.

[0051] Step S5 involves large-scale sampling of source-load data with randomness, evaluating it within a space reflecting the safe operation of submarine cable transmission power, and providing safety indicators for the transmission power of offshore wind power submarine cables. Specifically, this includes the following steps: (1) Scene generation through Monte Carlo sampling, specifically including: ① Basic scenario generation: Utilizing the uncertainty dataset generated by S3 [ P Load , Pseason g Using [[wind power output, load demand]] as the sampling source, N scenarios are uniformly generated in [[wind power output, load demand]] through Latin hypercube sampling; ② Importance Sampling: Increase sampling density in high-risk areas (such as extreme wind speeds and extreme loads), with sampling weights as follows: ; In the formula: x i ∈[ P Load , Pseason g ], p ( x ) represents the true joint probability distribution of source-charge power; q ( x ) represents the assumed distribution of uniform sampling; w i The larger the value, the higher the probability that the scenario will occur in the real distribution, and the more samples will be taken.

[0052] (2) Conduct dynamic security boundary assessment, including real-time security margin, security probability, and limit violation risk indicators, as detailed below: ① The calculation method for real-time safety margin is as follows: The safety threshold is dynamically adjusted based on the output of the DRL model, as follows: ; In the formula: This represents the dynamic safe power limit under state s; Let be the relevant safety margin function under state s, which is obtained by the trained DRL model under the current state s.

[0053] ②The calculation method for the probability of safety is as follows: For N sampled scenarios, count the number of scenarios that satisfy the safety constraints: ; In the formula: I(●) is an indicator function, which is 1 when the condition is met, and 0 otherwise; ③ The over-limit risk indicators include average over-limit power and maximum over-limit power, as detailed below: Average over-limit power The calculation is as follows: ; In the formula: N exceed This refers to the number of scenarios that exceed the limit.

[0054] Maximum over-limit power The calculation is as follows: .

[0055] Take mathematical data of offshore wind power in Europe, such as Figure 2 As shown, place it in the following way: Figure 3 In the transmission network shown, the line is a submarine cable. First, the collected raw offshore wind power output data, combined with other collected features, is preprocessed into multidimensional data. An anomaly detection mechanism using Z-score and DBSCAN clustering is employed for data formatting and cleaning. Then, multivariate KNN imputation and quantile transformation techniques are used to repair and improve the multidimensional feature data and wind power output data, followed by normalization. The preprocessed data is then used for offshore wind power prediction. The prediction results and errors will be... Figure 4 As shown, by Figure 4 (a) It can be seen that the prediction curve of the Att-BiLSTM model is in high agreement with the actual wind power output curve, and it can accurately track the trend of change even in the range of violent power fluctuations. Figure 4 (b) Error analysis shows that the prediction error is mainly concentrated within ±10kW, which is much less than 1% compared to the overall waveform error; the root mean square error (RMSE) is 3.47MW, the mean absolute error (MAE) is 6.12MW, and the mean absolute percentage error (MAPE) is 0.25%, which is 27.6% lower than the RMSE of the traditional LSTM model, demonstrating the superiority of the attention mechanism in capturing key temporal features. Therefore, the prediction method proposed in this invention can effectively predict offshore wind power output.

[0056] Based on the trained model and collected data, predicted offshore wind power output is generated. A random error, calculated as the mean of the predicted values ​​plus 0.15 times the variance of the predicted values, is added to the predicted data. Uncertain load demand is generated similarly using collected load demand data. Node voltage, phase angle, and submarine cable aging coefficient are considered as the overall state of the DRL (Depth-Range Load Management) system. The DRL's action space is constructed in the form of incremental wind power output. A hierarchical reward mechanism is implemented to maximize submarine cable transmission power, allowing the DRL to gradually approach the safe operating boundary of offshore wind power output at maximum transmission power during training. The convergence graph of the training process is shown in the figure. Figure 5As shown in the figure, the reward value of the PPO algorithm increases rapidly in the first 300 training rounds, from an initial -500 to -80, indicating that the agent quickly learns the basic safety control strategy. Between rounds 320 and 830, the reward value steadily increases and converges around -8, with the fluctuation range decreasing from ±80 to ±15, and the convergence standard deviation being approximately 12.3, indicating good training stability. The final convergence reward value is around -7.8, verifying the PPO algorithm's good adaptability in characterizing complex power transmission safety spaces. This leads to the conclusion that using PPO to characterize the power transmission safety space in this invention provides advantages for characterizing the safety space of submarine cable transmissions.

[0057] By combining the collected actual data with the trained DRL model, the optimal offshore wind power output and transmission power can be calculated simultaneously. Data was constructed using 20,000 discrete samples and Gaussian noise of dB=30. The resulting safety assessment of submarine cable transmission power is shown below. Figure 6 As shown in the figure, the safety assessment results are as follows: Figure 6 As shown, the safety probability density distribution is mainly located between 150 and 180 MW, with peak values ​​at two edges, and a transmission limit of 500 MW, indicating that the system mainly operates within a safe range. Compared to traditional assessment methods, the method used in this invention effectively improves assessment performance. The results verify that the method proposed in this invention can accurately assess the safety status of submarine cable transmission, providing a reliable basis for offshore wind power grid connection decisions. The method proposed in this invention can effectively assess the situation of submarine cable transmission.

[0058] Example 2: like Figure 7 As shown, based on the same inventive concept as Embodiment 1, this embodiment provides a probabilistic security assessment system for offshore wind power submarine cable transmission power, and the method described includes: The data processing module is used to collect measured offshore wind power data and perform data preprocessing, including removing outliers, filling in blank values, and normalization. The measured offshore wind power data includes temperature, wind speed, and offshore wind power output data. The preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y The power output of offshore wind power; The prediction model building module is used to build a training model of attention mechanism-bidirectional long short-term memory neural network, and to train the optimal regression model for predicting offshore wind power output by combining it with preprocessed measured data. The dataset construction module is used to collect wind power data for environmental data prediction and construct an uncertainty dataset by combining wind power prediction error and load fluctuation data. The reinforcement learning model building module is used to build a deep reinforcement learning model that characterizes the power transmission safety space. The deep reinforcement learning model is trained in combination with an uncertainty dataset to characterize the space that can reflect the safe operation of submarine cable transmission power. The safety assessment module is used to perform large-scale sampling of source-load data with randomness, conduct assessments in a space that reflects the safe operation of submarine cable transmission power, and provide safety indicators for the transmission power of offshore wind power submarine cables.

[0059] Example 3: Based on the same inventive concept as Embodiment 1, this embodiment provides a computer-readable storage medium, which includes a stored program, wherein, when the program is executed, it controls the device where the computer-readable storage medium is located to execute the aforementioned method for probabilistic safety assessment of offshore wind power submarine cable transmission power.

[0060] Example 4: Based on the same inventive concept as Embodiment 1, this embodiment provides a processor for running a program, wherein the program executes the aforementioned method for probabilistic security assessment of offshore wind power transmission power via submarine cables.

[0061] Those skilled in the art will recognize that the modules of the various examples described in conjunction 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 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.

[0062] In the embodiments provided by this invention, it should be understood that the division of modules is only a logical functional division. In actual implementation, there may be other division methods, such as multiple modules can be combined into one module, one module can be split into multiple modules, or some features can be ignored.

[0063] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0064] If the integrated module is implemented as a software functional module 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.

[0065] 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 method for probabilistic security assessment of offshore wind power transmission power via submarine cables, characterized in that, Includes the following steps: Step S1: Collect measured offshore wind power data and perform data preprocessing, including removing outliers, filling in blanks, and normalization. Measured offshore wind power data includes temperature, wind speed, and offshore wind power output data. The preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y The power output of offshore wind power; Step S2: Construct a training model of attention mechanism-bidirectional long short-term memory neural network, and train it with preprocessed measured data to obtain the optimal regression model for predicting offshore wind power output. Step S3: Collect wind power data predicted from environmental data, and construct an uncertainty dataset by combining wind power prediction error and load fluctuation data; Step S4: Construct a deep reinforcement learning model to characterize the power transmission safety space, and train the deep reinforcement learning model with an uncertainty dataset to characterize the space that can reflect the safe operation of the submarine cable transmission power. Step S5 involves large-scale sampling of source-load data with randomness, evaluation within a space reflecting the safe operation of submarine cable transmission power, and provision of safety indicators for offshore wind power transmission power of submarine cables.

2. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, Step S1, which removes outlier data, specifically includes identifying and removing outliers using an improved Z-score method and detecting outliers in multidimensional data based on clustering. The improved Z-score method for identifying and removing outliers is as follows: For time series data X={ x 1, x 2,…, x n } Calculate the moving median of the time series data X. M t and moving absolute median M ad The anomaly detection threshold is: ; If the i-th data point x i If the value exceeds this range, it is considered an outlier and removed. The following are specific examples of outlier detection based on clustering in multidimensional data: Clustering algorithms are used to identify outliers in sparse regions of multidimensional data. The formula is as follows: ; In the formula: D core For data core density, x Data collected from actual sampling; ϵ The neighborhood radius is defined, and samples with a density below a threshold are considered abnormal. The multidimensional data includes wind speed, power, and temperature.

3. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, In step S1, filling in the blanks specifically uses the K-nearest neighbor algorithm, which selects the nearest neighbor sample for weighted filling based on Euclidean distance. The calculation formula is as follows: ; in, These are the missing items to be filled. Let be the corresponding feature value of the j-th nearest neighbor sample; For the first j The weights of the nearest neighbor samples are calculated based on the reciprocal of the Euclidean distance. k The number of nearest neighbor samples selected; Indicates the data value after filling. The corresponding feature value of the j-th nearest neighbor sample The Euclidean distance; This is the data value after filling in the missing parts.

4. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, Step S2 specifically includes the following steps: (1) Constructing a training model for an attention-based bidirectional long short-term memory neural network, including an input layer, an LSTM layer, an output layer, and an attention mechanism; specifically as follows: 1) The input layer receives normalized multidimensional time-series data P X ∈R T×D , T The total length of the time series. D For feature dimensions; 2) The LSTM layer extracts temporal features bidirectionally. The hidden states of the forward and backward LSTMs are h→ t and h← t After piecing them together, we get: ; in, H h is the number of hidden units in a single-layer LSTM; where: h t for t The hidden state vector at time step; This represents the hidden state of the forward LSTM at time t; This represents the hidden state of the backward LSTM at time t; 3) The attention mechanism uses trainable weight parameters W a ∈R 2H×H and v a ∈R H The attention weights are calculated as follows: ; The weighted vector is: ; In the formula: The attention-weighted context vector; α t for t Attention weights at each moment; h t for t The hidden state vector at time step; The output layer is mapped to the predicted value through a fully connected layer, resulting in: ; in, W o ∈R H×1 Let b be the weight matrix. o For bias terms of fully connected layers; (2) The optimal regression model for predicting offshore wind power output was obtained by training the preprocessed measured data. The Huber loss function was used during training, as follows: ; In the formula: y This represents the actual power output of offshore wind power. ŷ The predicted offshore wind power output value is given by the model. δ For the threshold; The optimizer uses the AdamW optimizer, which introduces weight decay to prevent overfitting, as detailed below: ; In the formula: λ The attenuation coefficient; θ t , θ t+1 The first t , t+ Model parameters for one iteration; η The learning rate; and These are the first-order moment estimates and second-order moment estimates after bias correction, respectively; ε adam It is a numerically stable term; During regularization, a Dropout layer is added after the BiLSTM layer to randomly drop out neurons at a ratio of k. Training is complete when the evaluation index of the regression model for predicting offshore wind power output reaches a preset threshold, and the optimal regression model for predicting offshore wind power output is obtained.

5. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, Step S3 specifically includes the following steps: (1) Wind power output prediction error modeling: Wind power prediction errors are caused by the superposition of multiple Gaussian distributions, with the following probability density function: ; In the formula: p ( ε (This represents the wind power prediction error) ε The probability density function; K The number of Gaussian components; π k For the first k The mixture weights of Gaussian components satisfy Σ π k =1; N (ε| μ k , σ 2 k) is the first k Gaussian components μ k The mean, σ 2k is the variance; Based on historical error data, the distribution is estimated using the Gaussian kernel function: ; In the formula: The probability density function is the estimated value. N This represents the number of historical error samples. ε i For the first i One historical error sample; h For bandwidth parameters; (2) Load fluctuation modeling: Seasonal decomposition is used to break down load data into trend terms. T t Seasonal items S t and residuals R t The details are as follows: ; In the formula: P Load,t for t The load value at time t; the residuals follow a normal distribution; (3) Use the trained optimal regression model for predicting offshore wind power output to predict wind power output in future periods. Randomly sample wind power prediction errors from the wind power output prediction error model. ε t Random wind power output is generated, as follows: ; A large number of scenarios are generated to form an uncertainty dataset. P Load , Pseason g ]. Based on the above model and S2 predicted wind power data, stochastic wind power output and stochastic load demand are generated. P season g and P Load .

6. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, Step S4 specifically includes the following steps: (1) The state space S used space This includes electrical measurement parameters, environmental parameters, and equipment status parameters. Specifically, the electrical measurement parameters include submarine cable transmission power. P cable Node voltage V Phase angle θ Uncertain loads and wind power P Load 、 Pseason g Environmental parameters include wind speed. v wind Equipment status parameters include the aging coefficient of the submarine cable. α aging ; The motion space A used space This includes adjusting the incremental Δ of wind power. P ∈[−Δ P max , Δ P max ],satisfy: ; In the formula: P cable,t This represents the transmission power of the submarine cable at time t. (2) The reward function includes security rewards. r safe Economic rewards r econ Equipment lifespan bonus r life The details are as follows: ; In the formula: w 1 represents the safety reward weighting coefficient; P mid This is the reference value for the rated transmission power of the submarine cable; σ safe This is the sensitivity variance coefficient; ; In the formula: w 2 represents the economic reward weighting coefficient; R cable This refers to the economic coefficient for losses. ; In the formula: w 3 represents the weighting coefficient for equipment lifespan bonus; Δ P This refers to power fluctuations. The total reward is: ; In the formula, r t for t Instant rewards for each moment; (3) Dominance function AGAE t The generalized dominance method is used to estimate the equilibrium bias and variance, as follows: ; In the formula: AGAE t for t Generalized advantage estimate at time; δ t+1 This refers to timing difference error; γ Discount factor; λ GAE These are GAE parameters used to balance bias and variance; δ t = r t +γ V ( s t+1 )− V ( s t ), V ( s t ) is the state s t Value function estimation; l Number of steps; Smooth policy updates are ensured through KL divergence constraints, as detailed below: ; Where: KL( π θold || π θ The KL divergence between the old and new strategies is used to measure the difference in strategy distributions. π θold The parameters are the old strategy parameters before the update; π θ For the updated new strategy parameters; δ KL The constraint threshold for KL divergence.

7. The method for probabilistic security assessment of offshore wind power submarine cable transmission power according to claim 1, characterized in that, Step S5 specifically includes the following steps: (1) Scene generation through Monte Carlo sampling, specifically including: Uncertainty dataset [ P Load , Pseason g Using [[wind power output, load demand]] as the sampling source, N scenarios are uniformly generated across [wind power output, load demand] through Latin hypercube sampling; the sampling density is increased for high-risk areas, with the following weights: ; In the formula: x i ∈[ P Load , Pseason g ], p ( x ) represents the true joint probability distribution of source-charge power; q ( x () represents the assumed distribution of uniform sampling; (2) Conduct dynamic security boundary assessment, including real-time security margin, security probability, and limit violation risk indicators, as detailed below: ① The calculation method for real-time safety margin is as follows: The safety threshold is dynamically adjusted based on the output of the DRL model, as follows: ; In the formula: This represents the dynamic safe power limit under state s; Let be the relevant safety margin function under state s, which is obtained by the trained DRL model under the current state s. ②The calculation method for the probability of safety is as follows: For N sampled scenarios, count the number of scenarios that satisfy the safety constraints: ; In the formula: I(●) is an indicator function, which is 1 when the condition is met, and 0 otherwise; ③ The over-limit risk indicators include average over-limit power and maximum over-limit power, as detailed below: Average over-limit power The calculation is as follows: ; In the formula: N exceed This refers to the number of scenarios that exceed the limit. Maximum over-limit power The calculation is as follows: 。 8. A probabilistic safety assessment system for offshore wind power submarine cable transmission power, characterized in that, The method described by any one of claims 1 to 7 includes: The data processing module is used to collect measured offshore wind power data and perform data preprocessing, including removing outliers, filling in blank values, and normalization. The measured offshore wind power data includes temperature, wind speed, and offshore wind power output data. The preprocessed wind power data sample is P. X P Y ;where P X This indicates that the data includes temperature, wind speed, and time characteristics; P Y The power output of offshore wind power; The prediction model building module is used to build a training model of attention mechanism-bidirectional long short-term memory neural network, and to train the optimal regression model for predicting offshore wind power output by combining it with preprocessed measured data. The dataset construction module is used to collect wind power data for environmental data prediction and construct an uncertainty dataset by combining wind power prediction error and load fluctuation data. The reinforcement learning model building module is used to build a deep reinforcement learning model that characterizes the power transmission safety space. The deep reinforcement learning model is trained in combination with an uncertainty dataset to characterize the space that can reflect the safe operation of the submarine cable transmission power. The safety assessment module is used to perform large-scale sampling of source-load data with randomness, conduct assessments in a space that reflects the safe operation of submarine cable transmission power, and provide safety indicators for the transmission power of offshore wind power submarine cables.

9. 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 a method for probabilistic safety assessment of offshore wind power transmission power as described in any one of claims 1 to 7.

10. A processor, characterized in that, The processor is used to run a program, wherein the program executes a method for probabilistic safety assessment of offshore wind power transmission power via submarine cables as described in any one of claims 1 to 7.