Intelligent cable control method and system for conveying clean energy

Abstract

The invention provides an intelligent cable control method and system for conveying clean energy, and the method comprises the steps: obtaining historical energy data of a grid connection point of an intelligent cable, carrying out the dynamic graph structure modeling according to the historical energy data, and obtaining a time-space correlation data set; determining a multi-order space-time weight matrix according to the space-time correlation data set and a preset multi-order Chebyshev polynomial; based on the multi-order space-time weight matrix, performing self-attention enhancement construction to obtain a multi-dimensional dynamic space-time attention feature matrix; determining a space-time dynamic bearing matrix according to a plurality of preset gated convolution capture scales and the multi-dimensional dynamic space-time attention feature matrix; according to the space-time dynamic load matrix, evaluating and predicting the dynamic load of the grid-connected point to obtain a short-term load prediction result and a long-term load prediction result; and generating a load control instruction of the grid connection point of the intelligent cable based on the short-term and long-term load prediction results.

Description

Technical Field

[0001] The present application relates to the field of clean energy technology, and in particular to an intelligent cable control method and system for transmitting clean energy. Background Art

[0002] At present, with the rapid development of renewable energy such as solar energy and wind energy, smart grids and smart cables are playing an increasingly important role in the transmission of clean energy. In order to adapt to the characteristics of renewable energy, smart cables for renewable energy have emerged. This cable designed for renewable energy systems such as solar energy and wind energy needs to have a higher level of intelligence. However, in practical applications, how to achieve dynamic load adjustment of smart cables has become an urgent problem to be solved. This involves the core content of energy management, which requires real-time adjustment of the load capacity of smart cables according to the output characteristics of clean energy and changes in electricity demand. These control methods are aimed at improving the transmission efficiency and safety of clean energy, but they also bring technical challenges. Summary of the invention

[0003] The present application provides an intelligent cable control method and system for transmitting clean energy, which are used to improve the transmission efficiency and safety of clean energy.

[0004] In a first aspect, an embodiment of the present application provides an intelligent cable control method for transmitting clean energy, the method comprising: Obtaining historical energy data of the grid connection point of the smart cable, performing dynamic graph structure modeling based on the historical energy data, and obtaining a time-space related data set; Determining a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial; Based on the multi-order spatiotemporal weight matrix, self-attention enhancement construction is performed to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix; Determine a spatiotemporal dynamic bearing matrix according to a plurality of preset gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix; According to the spatiotemporal dynamic load matrix, the dynamic load of the grid connection point is evaluated and predicted to obtain a short-term load prediction result and a long-term load prediction result; Based on the short-term and long-term load prediction results, a load control instruction for the grid connection point of the smart cable is generated.

[0005] In a second aspect, an embodiment of the present application provides an intelligent cable control system for transmitting clean energy, the device comprising: A data processing module, used to obtain historical energy data of the grid connection point of the smart cable, perform dynamic graph structure modeling based on the historical energy data, and obtain a time-space related data set; A weighted fusion module, used for determining a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial; A feature discovery module, used to perform self-attention enhancement construction based on the multi-order spatiotemporal weight matrix to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix; A dynamic tracking module, used to determine a spatiotemporal dynamic bearing matrix according to a plurality of preset gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix; A result prediction module, used to evaluate and predict the dynamic load of the grid connection point according to the spatiotemporal dynamic load matrix, and obtain a short-term load prediction result and a long-term load prediction result; The cable control module is used to generate a load control instruction for the grid connection point of the smart cable based on the short-term and long-term load prediction results.

[0006] In a third aspect, an embodiment of the present application provides an electronic device, the electronic device comprising a memory and a processor; The memory is used to store computer programs; The processor is used to execute the computer program and implement the intelligent cable control method for transmitting clean energy as described in any one of the embodiments of the present application when executing the computer program.

[0007] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements an intelligent cable control method for transporting clean energy as described in any one of the embodiments of the present application.

[0008] An embodiment of the present application provides a smart cable control method for transmitting clean energy, the method comprising: obtaining historical energy data of a grid connection point of the smart cable, performing dynamic graph structure modeling based on the historical energy data, and obtaining a time-space related data set; determining a multi-order time-space weight matrix based on the time-space related data set and a preset multi-order Chebyshev polynomial; performing self-attention enhancement construction based on the multi-order time-space weight matrix to obtain a multi-dimensional dynamic time-space attention feature matrix; determining a time-space dynamic load matrix based on a preset plurality of gated convolution capture scales and a multi-dimensional dynamic time-space attention feature matrix; evaluating and predicting the dynamic load of the grid connection point based on the time-space dynamic load matrix to obtain a short-term load prediction result and a long-term load prediction result; and generating a load control instruction for the grid connection point of the smart cable based on the short-term and long-term load prediction results. Through the above method, through dynamic graph structure modeling and spatiotemporal related data set generation, the dynamic correlation between grid points is accurately captured, and the volatility of clean energy power generation is effectively adapted. The combination of multi-order Chebyshev polynomials and self-attention mechanism enhances the modeling ability of multi-scale spatial relationships and long-range time dependencies, improves the accuracy of load prediction, and through the application of multi-scale gated convolution, the model considers both short-term and long-term time characteristics, and provides a comprehensive decision-making basis for intelligent cable control, achieving more balanced and efficient energy transmission, and realizing the intelligence of the entire process from data collection to control command generation, significantly improving the efficiency and reliability of clean energy transmission, showing significant advantages in dynamic adaptability, prediction accuracy and system reliability, and providing strong technical support for the intelligent transmission of clean energy. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying any creative work.

[0010] Figure 1 A schematic flow chart of a smart cable control method for transmitting clean energy provided in an embodiment of the present application; Figure 2 A schematic block diagram of an intelligent cable control system for transmitting clean energy provided in an embodiment of the present application. DETAILED DESCRIPTION

[0011] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0012] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0013] It should also be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0014] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0015] See also Figure 1 , Figure 1 The embodiment of the present application also provides a schematic flow chart of a smart cable control method for transmitting clean energy. The specific steps of the method include: S101-S106.

[0016] S101. Obtain historical energy data of the grid connection point of the smart cable, perform dynamic graph structure modeling based on the historical energy data, and obtain a time-space related data set.

[0017] It should be noted that the Point of Common Coupling (PCC) refers to the place where the smart cables of the distributed generation system (such as wind power generation, photovoltaic power generation, etc.) are connected to the public power grid. The grid connection point is the key point for the exchange of electric energy between the distributed generation system and the power grid. Distributed optical fiber sensors are installed on the smart cables to collect data such as temperature, stress, vibration, etc. on the cables during the power transmission process. The connection point between the smart cable and the power grid is the grid connection point. At this grid connection point, the distributed generation system, such as a wind farm or a solar farm, can inject electric energy into the grid and absorb electric energy from the grid. Voltage and frequency regulation equipment, protection equipment, electric energy metering equipment, and communication equipment are installed at the grid connection point to ensure the coordinated operation of the wind farm and the power grid.

[0018] Distributed power generation systems, such as wind farms or solar farms, have power generation efficiencies that are affected by the weather and have different patterns in different periods. When analyzing their power generation patterns, it is necessary to combine short-term patterns to analyze the impact of weather factors, and it is also necessary to analyze long-term patterns to analyze the impact of the transmission network structure between the distributed system and the power grid.

[0019] Exemplarily, by constructing a new graph structure model, the dynamic spatial correlation attributes between the grid connection points are directly extracted from the historical electrical flow data of the grid connection points. This graph structure is dynamically established based on multi-source data, and the multi-dimensional attributes between the grid connection points (such as temperature, stress, vibration, current, voltage and power factor) are used to determine the dynamic similarity of the power transmission law of the grid connection points. The generated dynamic graph structure attributes are defined as a spatiotemporal perception distance data set, which is used to reflect the time dependence and spatial correlation between the grid connection points in the power grid. This spatiotemporal perception distance data set is used in the next processing step to dynamically adjust the transmission power of the smart cable and monitor load changes.

[0020] S102: Determine a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial.

[0021] Exemplarily, a new spatial correlation module is designed based on multi-order Chebyshev polynomials using a spatiotemporal correlation dataset. This module processes STAD data through a dynamic graph convolutional neural network (GCN) to mine the spatial correlation between grid points at different spatial scales. Specifically, the dynamic graph structure model introduces multi-scale neighborhoods in GCN to capture the interdependence of grid points at different spatial distances, and then accurately adjusts the spatial weights of the Chebyshev polynomials. This module generates a multi-order spatiotemporal weight matrix by analyzing the correlation at different neighborhood levels, providing a basis for subsequent temporal correlation calculations.

[0022] S103. Based on the multi-order spatiotemporal weight matrix, self-attention enhancement is constructed to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix.

[0023] Exemplarily, based on the multi-order spatiotemporal weight matrix, an adaptive spatiotemporal attention module is constructed to further enhance the dynamic perception ability of the model. The self-attention module is improved according to the characteristics of graph convolution, and automatically adapts the spatial weights of the input of each order Chebyshev polynomial. Combined with the multi-head self-attention mechanism, the module captures the dynamic association of the cable connection points at different time scales, enhances the model's ability to capture complex spatial topology and dynamic time relationships, and finally generates a multi-dimensional dynamic spatiotemporal attention feature matrix, which provides spatiotemporal characteristic support for the subsequent cable load prediction and control strategy formulation.

[0024] S104, determining a spatiotemporal dynamic bearing matrix according to a plurality of preset gated convolution capture scales and a multi-dimensional dynamic spatiotemporal attention feature matrix.

[0025] Exemplarily, based on a multi-dimensional dynamic spatiotemporal attention feature matrix, the dynamic receptive fields of multiple gated convolution capture scales are fused to capture the interdependence of the grid connection points of the smart cable at multiple time scales. The above process is completed by a gated convolutional network, which effectively handles load fluctuations and emergencies during power transmission by combining short-term and long-term time features. Thus, the spatiotemporal features on multiple time scales can be fused into a unified dynamic feature matrix, thereby generating a load prediction value containing time dependency and forming a spatiotemporal dynamic load matrix, providing an accurate basis for subsequent dynamic regulation.

[0026] S105. Evaluate and predict the dynamic load of the grid connection point according to the spatiotemporal dynamic load matrix to obtain short-term load prediction results and long-term load prediction results.

[0027] For example, based on the spatiotemporal dynamic load matrix, a load forecasting model is constructed to analyze and predict the load conditions of each grid connection point in different time periods. The model uses the LSTM network to process the time dependence of each grid connection point in power transmission, and combines GCN to efficiently model spatial characteristics to achieve accurate prediction of future short-term and long-term loads. The load forecasting model continuously optimizes model parameters through multiple iterative training to adapt to changes under different load conditions. By analyzing the deviation between the predicted value and the actual data, the model further corrects the load forecasting results to provide support for the next step of dynamic load regulation.

[0028] S106: Generate a load control instruction for the grid connection point of the smart cable based on the short-term load prediction result and the long-term load prediction result.

[0029] For example, based on the short-term load forecast results and the long-term load forecast results, a dynamic load adjustment strategy for smart cables is formulated by designing a multi-objective optimization method. An improved particle swarm optimization algorithm is used to take multi-objective constraints such as cable safety load capacity and grid stability as conditions to generate a load control scheme that meets dynamic needs. At the same time, combined with the adaptive optimization mechanism, the reinforcement learning method is used to continuously adjust the control strategy parameters according to the actual load data to dynamically adapt to changes in external conditions. Finally, the load control instructions are executed through a distributed control system to achieve real-time monitoring, risk warning and dynamic control of energy transmission, ensuring the efficient transmission of clean energy and the safety and reliability of the cable system.

[0030] The embodiment of the present application provides a smart cable control method for transmitting clean energy, the method comprising: obtaining historical energy data of a grid connection point of the smart cable, performing dynamic graph structure modeling based on the historical energy data, and obtaining a time-space related data set; determining a multi-order time-space weight matrix based on the time-space related data set and a preset multi-order Chebyshev polynomial; performing self-attention enhancement construction based on the multi-order time-space weight matrix to obtain a multi-dimensional dynamic time-space attention feature matrix; determining a time-space dynamic load matrix based on a preset plurality of gated convolution capture scales and a multi-dimensional dynamic time-space attention feature matrix; evaluating and predicting the dynamic load of the grid connection point based on the time-space dynamic load matrix to obtain a short-term load prediction result and a long-term load prediction result; and generating a load control instruction for the grid connection point of the smart cable based on the short-term load prediction result and the long-term load prediction result. Through the above method, through dynamic graph structure modeling and spatiotemporal related data set generation, the dynamic correlation between grid points is accurately captured, and the volatility of clean energy power generation is effectively adapted. The combination of multi-order Chebyshev polynomials and self-attention mechanism enhances the modeling ability of multi-scale spatial relationships and long-range time dependencies, improves the accuracy of load prediction, and through the application of multi-scale gated convolution, the model considers both short-term and long-term time characteristics, and provides a comprehensive decision-making basis for intelligent cable control, achieving more balanced and efficient energy transmission, and realizing the intelligence of the entire process from data collection to control command generation, significantly improving the efficiency and reliability of clean energy transmission, showing significant advantages in dynamic adaptability, prediction accuracy and system reliability, and providing strong technical support for the intelligent transmission of clean energy.

[0031] In order to more clearly introduce the technical solution of the present application, the technical solution of the present application will be introduced through specific embodiments below. It should be noted that the specific embodiments are used to expand the technical solution of the present application, but are not intended to limit the present application.

[0032] In some embodiments, historical energy data of the grid connection points of the smart cable are obtained, and dynamic graph structure modeling is performed based on the historical energy data to obtain a time-space related data set, including: vectorizing the historical energy data to obtain an energy vector sequence; performing normalization calculation based on the energy vector sequence to obtain an energy probability distribution sequence; based on the energy probability distribution sequence, performing cosine distance calculation on the energy vectors between the grid connection points to obtain a conversion cost matrix; using the conversion cost matrix and the energy probability distribution sequence, minimizing the time-space perception distance between the grid connection points to obtain a time-space correlation matrix; sparsifying and binarizing the time-space correlation matrix to obtain a time-space correlation graph structure matrix; generating a time-space related data set based on the time-space correlation graph structure matrix.

[0033] Exemplarily, the multi-day historical energy data of each grid-connected point is organized into an energy vector sequence, and the historical vector sequence is vectorized and normalized. By calculating the Euclidean norm of each energy vector and normalizing it to a probability distribution value, an energy probability distribution sequence is obtained, which retains the key features of energy distribution. Using cosine distance as the cost function, the similarity of electrical flow vectors between different grid-connected points and different dates is calculated. The cosine distance calculation can cleverly quantify the energy similarity between grid-connected points, laying the foundation for subsequent spatiotemporal correlation analysis. The minimization calculation of spatiotemporal perception distance further reveals the dynamic correlation pattern between grid-connected points, while the sparsification and binarization processing effectively reduces the data complexity and highlights the most critical spatiotemporal correlation. The resulting spatiotemporal correlation data set not only contains rich temporal and spatial information, but also has high interpretability and computational efficiency. This method significantly improves the understanding and modeling capabilities of the dynamic characteristics of clean energy transmission networks, and provides a more accurate and reliable data basis for subsequent load forecasting and control decisions, thereby helping to optimize energy distribution, improve transmission efficiency, and enhance the stability and reliability of the entire intelligent cable system.

[0034] In some embodiments, based on the spatiotemporal association graph structure matrix, a spatiotemporal correlation data set is generated, including: performing multi-head self-attention calculation on the spatiotemporal association graph structure matrix to obtain a dynamic time dependency feature matrix; reshaping the spatiotemporal dependency relationship between the grid points according to the dynamic time dependency feature matrix to obtain a reshaped spatiotemporal dependency feature; performing residual connection and layer normalization processing on the reshaped spatiotemporal dependency feature to obtain a temporal attention output matrix; adaptively encoding the spatial dependency according to the temporal attention output matrix to obtain a spatial attention weight coefficient; fusing the spatial attention weight coefficient and the spatiotemporal association graph structure matrix to obtain a revised spatial attention representation; aggregating the spatiotemporal correlation data according to the revised spatial attention representation to obtain a spatiotemporal correlation data set.

[0035] Exemplarily, the spatiotemporal correlation graph structure matrix is ​​projected into the query, key and value spaces respectively to obtain the Q matrix, K matrix and V matrix. For each attention head, the attention scores of the Q matrix, K matrix and V matrix are calculated. The output results of each attention head are concatenated and linearly transformed to obtain the dynamic time-dependent feature matrix. The dynamic time-dependent feature matrix is ​​reshaped into a three-dimensional tensor. This step separates the time and space dimensions for subsequent processing. The reshaped spatiotemporal dependency features are residually connected to the original input. This step helps to alleviate the gradient vanishing problem and preserve the original information. The temporal attention output matrix is ​​transposed and mapped to a high-dimensional space through a one-dimensional convolution, and the spatial attention weight coefficient is calculated after adding the position encoding information. The spatial attention weight coefficient is element-wise multiplied with the spatiotemporal correlation graph structure matrix, and then weighted by a learnable parameter. This step fuses prior knowledge with the learned attention. The input features are weighted summed using the modified spatial attention representation. Finally, the output is residually connected to the original input and layer normalized to obtain a spatiotemporal correlation dataset.

[0036] In some embodiments, a multi-order spatiotemporal weight matrix is ​​determined based on a spatiotemporal related data set and a preset multi-order Chebyshev polynomial, including: performing Laplace matrix decomposition processing on the spatiotemporal related data set to obtain an eigenvector matrix and an eigenvalue matrix; parameterizing the Chebyshev polynomials according to the eigenvector matrix and the eigenvalue matrix to obtain multi-order Chebyshev polynomial coefficients; performing convolution operations on the multi-order Chebyshev polynomial coefficients and the spatiotemporal related data set to obtain an initial spatiotemporal feature representation; recursively expanding the multi-order Chebyshev polynomials according to the initial spatiotemporal feature representation to obtain a multi-order spectral convolution kernel; performing adaptive weight allocation processing on the multi-order spectral convolution kernel and the spatiotemporal related data set to obtain an adaptive multi-order spatiotemporal weight coefficient; performing nonlinear transformation processing on the spatiotemporal related data set according to the adaptive multi-order spatiotemporal weight coefficient to obtain a multi-order spatiotemporal weight matrix.

[0037] For example, let the spatiotemporal correlation dataset be , where N is the number of nodes / space units, T is the time step, and F is the feature dimension. Then, the Laplacian matrix is: , where A∈R N×N , A is the adjacency matrix, which represents the connection relationship between nodes; D∈R N×N , D is the degree matrix, D ii =∑ j A ij .

[0038] Perform eigendecomposition on the Laplacian matrix: , where U∈R N×N , U is the eigenvector matrix, Λ∈RN×N , Λ is the eigenvalue matrix.

[0039] Chebyshev polynomial of order k: , recursively defined as: ,in, , is the maximum eigenvalue in the eigenvalue matrix Λ, and I is the identity matrix. Multi-order Chebyshev polynomial coefficients The calculation formula is: After convolution, the initial spatiotemporal features are expressed as: According to the initial spatiotemporal feature representation, the multi-order Chebyshev polynomials are recursively expanded, and the multi-order spectral graph convolution kernel is obtained as follows: . An adaptive weight function is introduced to obtain the adaptive multi-order spatiotemporal weight coefficients, specifically: ,in, is an activation function (such as Softmax), and MLP is a multi-layer perceptron. Then, through nonlinear activation functions such as ReLU or Sigmoid, a multi-order spatiotemporal weight matrix is ​​obtained, which is: .

[0040] In this way, the accuracy of describing the spatiotemporal relationship in the intelligent cable network is improved, and the computational complexity is significantly reduced, providing a more reliable and efficient mathematical model for subsequent load forecasting and control decisions, thereby helping to optimize the energy distribution strategy and improve the overall performance and stability of the system.

[0041] In some embodiments, based on a multi-order spatiotemporal weight matrix, a self-attention enhancement construction is performed to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix, including: performing multi-head linear projection processing on the multi-order spatiotemporal weight matrix to obtain a query matrix, a key matrix and a value matrix; performing dot product operations on the spatiotemporal dependencies according to the query matrix and the key matrix to obtain an original attention score matrix; scaling and normalizing the original attention score matrix to obtain an attention weight matrix; performing weighted summation processing on the value matrix according to the attention weight matrix to obtain an initial self-attention output; performing residual connection and layer normalization processing on the initial self-attention output and the multi-order spatiotemporal weight matrix to obtain an enhanced self-attention feature; performing nonlinear transformation and multi-dimensional aggregation processing on the spatiotemporal dependencies according to the enhanced self-attention feature to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix.

[0042] Exemplarily, multi-head linear projection processing enables the model to capture information from multiple angles at the same time, enhancing the richness of feature expression. The dot product operation of the query matrix and the key matrix accurately quantifies the correlation between different elements, providing a basis for the subsequent attention mechanism. Through scaling and normalization, the attention weight matrix can more stably reflect the importance of each spatiotemporal element, and dynamically adjust the value matrix in the weighted sum to generate a more accurate initial self-attention output. The introduction of residual connection and layer normalization not only retains the original information, but also improves the training stability of the network and avoids the gradient vanishing problem. Through nonlinear transformation and multi-dimensional aggregation, the model can capture more complex spatiotemporal dependency patterns, and the generated multi-dimensional dynamic spatiotemporal attention feature matrix has higher accuracy and robustness. Overall, this method enhances the generalization ability of the model, provides solid technical support for load prediction and control optimization of intelligent cable systems, and helps to achieve efficient energy management and stable power transmission.

[0043] In some embodiments, a spatiotemporal dynamic bearing matrix is ​​determined based on a plurality of preset gated convolution capture scales and a multi-dimensional dynamic spatiotemporal attention feature matrix, including: S1041-S1045.

[0044] S1041. Perform multi-scale gated unit convolution processing on the multi-dimensional dynamic spatiotemporal attention feature matrix to obtain a multi-scale temporal feature representation.

[0045] Exemplarily, according to three preset gated Tanh units with different receptive field sizes, a convolution operation is performed on the multi-dimensional dynamic spatiotemporal attention feature matrix to obtain three groups of initial time features; the three groups of initial time features are activated by tanh function and sigmoid function respectively to obtain three groups of activated time features; a feature splicing operation is performed on the three groups of activated time features to obtain spliced ​​multi-scale time features; and a convergence layer processing is performed on the spliced ​​multi-scale time features to obtain a multi-scale time feature representation.

[0046] S1042: Adaptively weight the spatiotemporal dependency relationship according to the multi-scale temporal feature representation to obtain an adaptive spatiotemporal weight coefficient.

[0047] Exemplarily, self-attention calculation is performed on the multi-scale time feature representation to obtain the attention score of the time dimension; weighted summation is performed on the multi-dimensional dynamic spatiotemporal attention feature matrix according to the attention score of the time dimension to obtain the time weighted feature; self-attention calculation of the spatial dimension is performed on the time weighted feature to obtain the attention score of the spatial dimension; weighted summation is performed on the time weighted feature according to the attention score of the spatial dimension to obtain an adaptive spatiotemporal weight coefficient.

[0048] S1043. Perform Chebyshev polynomial expansion processing on the adaptive spatiotemporal weight coefficients and the multi-dimensional dynamic spatiotemporal attention feature matrix to obtain a multi-order spectral graph convolution kernel.

[0049] Exemplarily, a Laplace matrix is ​​constructed based on the adaptive spatiotemporal weight coefficients to obtain eigenvalues ​​and eigenvectors; the eigenvalues ​​and eigenvectors are parameterized by Chebyshev polynomials to obtain polynomial coefficients; the multidimensional dynamic spatiotemporal attention feature matrix is ​​recursively expanded based on the polynomial coefficients to obtain a K-order spectral graph convolution kernel.

[0050] S1044, performing nonlinear transformation processing on the spatiotemporal dependency according to a multi-order spectral graph convolution kernel to obtain a spatiotemporal dynamic feature matrix; Exemplarily, a convolution operation is performed on a multi-order spectral graph convolution kernel and a multi-dimensional dynamic spatiotemporal attention feature matrix to obtain an initial spatiotemporal feature; a nonlinear activation function is performed on the initial spatiotemporal feature to obtain an activated spatiotemporal feature. The activated spatiotemporal feature is subjected to residual connection and layer normalization to obtain a spatiotemporal dynamic feature matrix.

[0051] S1045. Perform multi-head self-attention fusion processing on the spatiotemporal dynamic feature matrix to obtain the spatiotemporal dynamic bearing matrix.

[0052] Exemplarily, according to a preset number of attention heads, a linear projection is performed on the spatiotemporal dynamic feature matrix to obtain multiple groups of query, key and value matrices; dot product attention calculations are performed on the multiple groups of query, key and value matrices respectively to obtain multi-head attention outputs; the multi-head attention outputs are concatenated and linearly transformed to obtain fused attention features; and a weighted summation is performed on the spatiotemporal dynamic feature matrix according to the fused attention features to obtain a spatiotemporal dynamic carrying matrix.

[0053] In some embodiments, the dynamic load of the grid connection point is evaluated and predicted based on the spatiotemporal dynamic load matrix to obtain short-term load prediction results and long-term load prediction results, including: S1051-S1056.

[0054] S1051. Perform time series decomposition processing on the spatiotemporal dynamic load matrix to obtain trend components, seasonal components and residual components.

[0055] Exemplarily, a moving average filter is applied to the spatiotemporal dynamic load matrix to obtain an initial trend estimate. A detrended sequence is calculated based on the initial trend estimate to obtain a seasonal-residual mixed sequence. A periodic analysis is performed on the seasonal-residual mixed sequence to extract the seasonal pattern and obtain a seasonal component. The trend component and the seasonal component are subtracted from the original sequence to obtain a residual component.

[0056] S1052. Perform multi-scale feature extraction processing according to the trend component, the seasonal component and the residual component to obtain a multi-scale spatiotemporal feature representation.

[0057] For example, a long short-term memory network is applied to the trend component to extract long-term trend features. Fourier transform is used for the seasonal component to extract periodic features. Wavelet transform is applied to the residual component to extract multi-scale residual features. The long-term trend features, periodic features and multi-scale residual features are fused to obtain a multi-scale spatiotemporal feature representation.

[0058] S1053. Perform a spatiotemporal attention mechanism on the multi-scale spatiotemporal feature representation to obtain attention-weighted features.

[0059] Exemplarily, a temporal attention module is constructed to calculate the importance weights of different time steps. A spatial attention module is constructed to calculate the spatial dependencies of different grid points. The temporal attention and spatial attention results are combined to obtain a spatiotemporal attention matrix. The spatiotemporal attention matrix is ​​used to weight the multi-scale spatiotemporal feature representation to obtain the attention weighted feature.

[0060] S1054. Construct a deep residual network based on the attention weighted features to obtain a short-term load prediction model.

[0061] For example, a multi-layer residual block is designed, each of which contains a convolutional layer, a batch normalization layer, and an activation function. Skip connections are added between residual blocks to form a deep residual network structure. The attention-weighted features are used as input and forward propagated through the deep residual network. A fully connected layer is added at the end of the network to output the short-term load prediction result.

[0062] S1055. Perform long short-term memory network processing on the attention weighted features to obtain a long-term load prediction model.

[0063] For example, a multi-layer long short-term memory network is constructed, each layer contains multiple LSTM units. A dropout layer is added between LSTM layers to prevent overfitting. The attention weighted features are input into the LSTM network in chronological order. A fully connected layer is added after the LSTM network to output the long-term load forecast results.

[0064] S1056. Perform integrated learning processing according to the short-term load forecasting model and the long-term load forecasting model to obtain a short-term load forecasting result and a long-term load forecasting result.

[0065] Exemplarily, different subsets of training data are used to train multiple short-term prediction models and long-term prediction models. In the ensemble learning process, the prediction results of each model are weighted averaged to obtain the ensemble prediction result. An adaptive weight adjustment mechanism is introduced to dynamically adjust the weight of each model according to the historical prediction accuracy. Combining the short-term and long-term prediction results, a multi-scale prediction fusion model is constructed to obtain the short-term and long-term load prediction results.

[0066] See also Figure 2 , Figure 2 The embodiment of the present application also provides a schematic block diagram of an intelligent cable control system for delivering clean energy, wherein the intelligent cable control system 200 for delivering clean energy is used to execute the aforementioned intelligent cable control method for delivering clean energy. The intelligent cable control system 200 for delivering clean energy can be configured in a server.

[0067] Among them, the server can be an independent server or a server cluster, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN) and big data and artificial intelligence platforms.

[0068] like Figure 2 As shown, the intelligent cable control system 200 for transporting clean energy includes: a data processing module 201, a weighted fusion module 202, a feature discovery module 203, a dynamic tracking module 204, a result prediction module 205 and a cable control module 206.

[0069] The data processing module 201 is used to obtain the historical energy data of the grid connection point of the smart cable, perform dynamic graph structure modeling based on the historical energy data, and obtain a time-space related data set.

[0070] The weighted fusion module 202 is used to determine a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial.

[0071] The feature discovery module 203 is used to perform self-attention enhancement construction based on a multi-order spatiotemporal weight matrix to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix.

[0072] The dynamic tracking module 204 is used to determine the spatiotemporal dynamic bearing matrix according to the preset multiple gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix.

[0073] The result prediction module 205 is used to evaluate and predict the dynamic load of the grid connection point according to the spatiotemporal dynamic load matrix to obtain short-term load prediction results and long-term load prediction results.

[0074] The cable control module 206 is used to generate a load control instruction for the grid connection point of the smart cable based on the short-term load prediction result and the long-term load prediction result.

[0075] An embodiment of the present application provides an electronic device, which includes a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and implement any one of the intelligent cable control methods for transmitting clean energy in the embodiments of the present application when executing the computer program.

[0076] An embodiment of the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, the processor implements any one of the intelligent cable control methods for transporting clean energy in the embodiments of the present application.

[0077] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. An intelligent cable control method for transmitting clean energy, characterized in that: The method comprises: Obtaining historical energy data of the grid connection point of the smart cable, performing dynamic graph structure modeling based on the historical energy data, and obtaining a time-space related data set; Determining a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial; Based on the multi-order spatiotemporal weight matrix, self-attention enhancement construction is performed to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix; Determine a spatiotemporal dynamic bearing matrix according to a plurality of preset gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix; According to the spatiotemporal dynamic load matrix, the dynamic load of the grid connection point is evaluated and predicted to obtain a short-term load prediction result and a long-term load prediction result; Based on the short-term load prediction result and the long-term load prediction result, a load control instruction of the grid connection point of the smart cable is generated.

2. The intelligent cable control method for transmitting clean energy according to claim 1, characterized in that: The obtaining of historical energy data of the grid connection point of the smart cable, and performing dynamic graph structure modeling according to the historical energy data to obtain a time-space related data set includes: Performing vectorization processing on the historical energy data to obtain an energy vector sequence; Performing normalization calculation according to the energy vector sequence to obtain an energy probability distribution sequence; Based on the energy probability distribution sequence, the cosine distance calculation is performed on the energy vectors between the grid connection points to obtain a conversion cost matrix; Using the conversion cost matrix and the energy probability distribution sequence, the time-space perception distance between the grid connection points is minimized to obtain a time-space correlation matrix; The spatiotemporal correlation matrix is ​​subjected to sparse processing and binarization processing to obtain a spatiotemporal correlation graph structure matrix; Based on the spatiotemporal association graph structure matrix, a spatiotemporal correlation data set is generated.

3. The intelligent cable control method for transmitting clean energy according to claim 2, characterized in that: The step of generating a spatiotemporal correlation data set based on the spatiotemporal correlation graph structure matrix includes: Performing multi-head self-attention calculation on the spatiotemporal correlation graph structure matrix to obtain a dynamic time-dependent feature matrix; Reshaping the spatiotemporal dependency relationship between the grid connection points according to the dynamic time dependency feature matrix to obtain a reshaped spatiotemporal dependency feature; Performing residual connection and layer normalization processing on the reshaped spatiotemporal dependency features to obtain a temporal attention output matrix; Adaptively encoding the spatial dependency according to the temporal attention output matrix to obtain a spatial attention weight coefficient; The spatial attention weight coefficient and the spatiotemporal association graph structure matrix are fused to obtain a modified spatial attention representation; The spatiotemporal related data are aggregated according to the modified spatial attention representation to obtain the spatiotemporal related data set.

4. The intelligent cable control method for transmitting clean energy according to claim 1, characterized in that: Determining a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial includes: Performing Laplace matrix decomposition processing on the spatiotemporal correlation data set to obtain an eigenvector matrix and an eigenvalue matrix; Perform parameterization processing on the Chebyshev polynomials according to the eigenvector matrix and the eigenvalue matrix to obtain multi-order Chebyshev polynomial coefficients; Performing a convolution operation on the multi-order Chebyshev polynomial coefficients and the spatiotemporal correlation data set to obtain an initial spatiotemporal feature representation; Recursively expanding the multi-order Chebyshev polynomials according to the initial spatiotemporal feature representation to obtain a multi-order spectral graph convolution kernel; Performing adaptive weight allocation processing on the multi-order spectral graph convolution kernel and the spatiotemporal related data set to obtain adaptive multi-order spatiotemporal weight coefficients; The spatiotemporal correlation data set is subjected to nonlinear transformation processing according to the adaptive multi-order spatiotemporal weight coefficients to obtain the multi-order spatiotemporal weight matrix.

5. The intelligent cable control method for transmitting clean energy according to claim 1, characterized in that: The self-attention enhancement construction is performed based on the multi-order spatiotemporal weight matrix to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix, including: Performing multi-head linear projection processing on the multi-order spatiotemporal weight matrix to obtain a query matrix, a key matrix and a value matrix; Performing a dot product operation on the spatiotemporal dependency according to the query matrix and the key matrix to obtain an original attention score matrix; Scaling and normalizing the original attention score matrix to obtain an attention weight matrix; Performing weighted summation processing on the value matrix according to the attention weight matrix to obtain an initial self-attention output; Performing residual connection and layer normalization processing on the initial self-attention output and the multi-order spatiotemporal weight matrix to obtain an enhanced self-attention feature; The multi-dimensional dynamic spatio-temporal attention feature matrix is ​​obtained by performing nonlinear transformation and multi-dimensional aggregation processing on the spatio-temporal dependency according to the enhanced self-attention feature.

6. The intelligent cable control method for transmitting clean energy according to claim 1, characterized in that: Determining the spatiotemporal dynamic bearing matrix according to the preset multiple gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix includes: Performing multi-scale gated unit convolution processing on the multi-dimensional dynamic spatiotemporal attention feature matrix to obtain a multi-scale temporal feature representation; Performing adaptive weight allocation processing on the spatiotemporal dependency according to the multi-scale temporal feature representation to obtain an adaptive spatiotemporal weight coefficient; Performing Chebyshev polynomial expansion processing on the adaptive spatiotemporal weight coefficient and the multi-dimensional dynamic spatiotemporal attention feature matrix to obtain a multi-order spectral graph convolution kernel; Performing nonlinear transformation processing on the spatiotemporal dependency according to the multi-order spectral graph convolution kernel to obtain a spatiotemporal dynamic feature matrix; The spatiotemporal dynamic feature matrix is ​​subjected to multi-head self-attention fusion processing to obtain the spatiotemporal dynamic bearing matrix.

7. The intelligent cable control method for transmitting clean energy according to claim 1, characterized in that: The step of evaluating and predicting the dynamic load of the grid connection point according to the spatiotemporal dynamic load matrix to obtain a short-term load prediction result and a long-term load prediction result includes: Performing time series decomposition processing on the spatiotemporal dynamic load matrix to obtain a trend component, a seasonal component and a residual component; Perform multi-scale feature extraction processing according to the trend component, seasonal component and residual component to obtain a multi-scale spatiotemporal feature representation; Performing a spatiotemporal attention mechanism on the multi-scale spatiotemporal feature representation to obtain an attention-weighted feature; Constructing a deep residual network based on the attention weighted features to obtain a short-term load prediction model; Performing long short-term memory network processing on the attention weighted features to obtain a long-term load prediction model; An integrated learning process is performed based on the short-term load forecasting model and the long-term load forecasting model to obtain the short-term load forecasting result and the long-term load forecasting result.

8. An intelligent cable control system for delivering clean energy, characterized in that: The intelligent cable control system for transmitting clean energy includes: A data processing module, used to obtain historical energy data of the grid connection point of the smart cable, perform dynamic graph structure modeling based on the historical energy data, and obtain a time-space related data set; A weighted fusion module, used for determining a multi-order spatiotemporal weight matrix according to the spatiotemporal related data set and a preset multi-order Chebyshev polynomial; A feature discovery module, used to perform self-attention enhancement construction based on the multi-order spatiotemporal weight matrix to obtain a multi-dimensional dynamic spatiotemporal attention feature matrix; A dynamic tracking module, used to determine a spatiotemporal dynamic bearing matrix according to a plurality of preset gated convolution capture scales and the multi-dimensional dynamic spatiotemporal attention feature matrix; A result prediction module, used to evaluate and predict the dynamic load of the grid connection point according to the spatiotemporal dynamic load matrix, and obtain a short-term load prediction result and a long-term load prediction result; The cable control module is used to generate a load control instruction for the grid connection point of the smart cable based on the short-term load prediction result and the long-term load prediction result.

Images