A method for short-term load forecasting in power substations based on deep learning model

Abstract

The present application provides a method for predicting load in a substation based on a deep learning model, which relates to data processing technology. The method obtains power input parameter information through a transformer and sends it to a substation management node, and obtains power output parameter information through a load node set and sends it to the substation management node. Then, the substation management node responds to the power grid data call instruction, uses a preset substation load prediction framework, and determines the substation load prediction information according to the power input parameter information and the power output parameter information, and can determine the corresponding model for prediction processing according to different data call requirements, so as to ensure that the output prediction data can match the data call requirements while ensuring the model processing efficiency, that is, reasonably using processing resources while ensuring the prediction accuracy.

Description

Technical Field

[0001] The present application relates to data processing technology, and in particular to a method for short-term load forecasting in a substation based on deep learning technology. Background Art

[0002] In the power system, load forecasting in substations is an important part of power grid operation and management. Accurate load forecasting not only helps power grid companies to formulate reasonable power supply plans, but also effectively guides the allocation of power resources and improves the operation efficiency and stability of the power grid.

[0003] With the development of smart grids, the power grid substation management system has gradually realized informatization and automation, and accumulated a large amount of operation data. These data contain rich grid operation rules and characteristics, which makes it possible to improve the accuracy of load forecasting by using advanced data analysis technology. As a powerful data processing tool, deep learning models can learn complex relationships in data through multi-layer nonlinear mapping, and have achieved remarkable results in many fields. However, in the existing technology, the use of deep learning models to predict substation loads is based on a single model, that is, all the needs for substation load forecasting data in the power grid are predicted using a unified model. Summary of the invention

[0004] The present application provides a substation load forecasting method based on a deep learning model, so that the corresponding model can be determined for forecasting processing according to different data call requirements, thereby ensuring that the output forecast data can match the data call requirements while ensuring the model processing efficiency, that is, while ensuring the forecast accuracy, it can also reasonably utilize the processing resources.

[0005] In a first aspect, the present application provides a substation load prediction method based on a deep learning model, which is applied to a power grid substation management system, wherein the power grid substation management system includes a substation management node, a transformer, a transmission network, and a load node set, wherein the transformer is arranged at the input end of the transmission network, and the load nodes in the load node set are arranged at the output end of the transmission network, and the management node is communicatively connected with the transformer and each load node in the load node; the method includes:

[0006] Acquire power input parameter information through the transformer and send it to the substation management node, wherein the power input parameter information includes the power input parameter of the current time node and the historical power input parameter sequence;

[0007] Acquire power output parameter information through the load node set and send it to the substation management node, wherein the power output parameter information includes the power output parameter of the current time node and the historical power output parameter sequence;

[0008] The substation management node responds to the power grid data call instruction, utilizes the preset substation load forecasting framework, and determines the substation load forecasting information based on the power input parameter information and the power output parameter information. The preset substation load forecasting framework includes multiple load forecasting models trained based on deep learning models, wherein the instruction type of the power grid data call instruction is used to determine the data flow path of the power input parameter information and / or the power output parameter information in the preset substation load forecasting framework.

[0009] In the above scheme, the power input parameter information and the power output parameter information are obtained through the transformer and the load node set, including the parameters of the current time node and the historical parameter sequence, which provides data support for the input and output ends for load forecasting. This data collection method at both ends enables the prediction model to more accurately capture the dynamic characteristics and historical laws of power grid operation. In addition, the communication connection between the substation management node and each device (transformer, load node) can quickly respond to the power grid data call instruction, and use the preset substation load forecasting framework to process the collected data, thereby improving the speed and efficiency of information processing. Then, the deep learning model is used for load forecasting, which can learn the complex characteristics of power grid operation data through training, so as to extract useful information for load forecasting, and through the cooperation of multiple load forecasting models trained based on deep learning models, the corresponding model can be determined for forecasting processing according to different data call requirements, so as to ensure that the output forecast data can match the data call requirements while ensuring the model processing efficiency, that is, while ensuring the prediction accuracy, it can also reasonably use the processing resources. Specifically, the data flow path of the power input parameter information and / or power output parameter information in the preset substation load forecasting framework is determined by the instruction type of the power grid data call instruction, so that the system can select the most appropriate data processing flow according to actual needs, increasing the adaptability of the algorithm. In addition, by integrating the substation management nodes, transformers, transmission networks and load nodes into one system, the systematization and automation of power grid substation management is achieved. By providing load forecasting information, it can provide strong support for the power supply decision-making of the power grid and the output of the forecast operation information of the transmission network, which helps to improve the operation efficiency and stability of the power grid.

[0010] Optionally, the preset area load forecasting framework includes a first load forecasting model and a second load forecasting model;

[0011] The first load forecasting model includes a first input port and a first output port, and the first output port includes a first feedback output port and a first result output port;

[0012] The second load forecasting model includes a second input port and a second output port;

[0013] The first feedback output port is connected to the second input port.

[0014] In the above scheme, by splitting the preset area load forecasting framework into the first load forecasting model and the second load forecasting model, a modular design of the forecasting process is achieved, so that each model can focus on processing different types of data or specific stages of the forecasting task, thereby improving the accuracy and efficiency of the forecast. In addition, the first output port of the first load forecasting model includes a first feedback output port, which is connected to the second input port of the second load forecasting model to form a feedback mechanism. This mechanism enables the forecast results of the first model to be used as the input of the second model, enhances the correlation between the two models, helps to further explore the deep relationship between the data, and improves the accuracy and robustness of the forecast.

[0015] In addition, since the framework contains two independently configurable and trainable load forecasting models, the type and parameters of the model can be flexibly adjusted according to actual needs to adapt to different forecasting scenarios and accuracy requirements. This flexibility makes the technical solution have a wider range of applications and stronger adaptability. Moreover, through the collaborative work of the two models, the massive data from transformers and load nodes can be processed more efficiently, thereby outputting more suitable forecast results for different forecasting scenarios.

[0016] Optionally, the data flow path includes a first data flow path and a second data flow path;

[0017] The first data flow path is a flow path that sequentially passes through the first input port to the first result output port;

[0018] The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port to the second output port in sequence.

[0019] In the above scheme, by clarifying the specific processes of the first data flow path and the second data flow path, the data processing process is made more orderly. This clear path design helps to reduce confusion and errors in the data processing process and improve the accuracy and efficiency of data processing. In addition, by providing two different data flow paths, the system can select the most appropriate processing flow according to the specific needs of the power grid data call instruction. This design enhances the flexibility and adaptability of the system, enabling the system to cope with different prediction tasks and scenario requirements.

[0020] Then, in the first data flow path, the data is directly processed by the first load forecasting model and the results are output, realizing the preliminary prediction of the input data. In the second data flow path, the data is first processed by the first load forecasting model, and the results are input into the second load forecasting model as feedback, realizing the deep fusion and reuse of data. This processing method helps to fully explore the useful information in the input segment data and the output end data, thereby improving the accuracy and depth of the prediction.

[0021] Moreover, the design of the two data flow paths enables the first load forecasting model and the second load forecasting model to work together to complete the forecasting task. This synergy helps to improve the performance and efficiency of the entire forecasting system, making the forecast results more accurate and reliable. Through the clear design of data flow paths, the uncertainty factors in the system operation process can be reduced and the stability and reliability of the system can be improved. At the same time, the multi-path design also provides redundancy and backup mechanisms for the system, so that when problems occur in some models or paths, the system can continue to run and output forecast results.

[0022] Optionally, in the first data flow path, the first area load output information is determined using Formula 1, where Formula 1 is:

[0023] ,

[0024] in, For the current moment The next moment The first station area load output information, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, The first load forecasting model at the current time The hidden state vector of is a first model parameter set of the first load forecasting model;

[0025] In the second data flow path, the second area load output information is determined using Formula 2, which is:

[0026] ,

[0027] in, For the current moment The next moment The second station area load output information, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load prediction information transmitted to the second input port through the first feedback output port, A second model parameter set of the second load forecasting model.

[0028] In the above scheme, in the first data flow path, the first load prediction model is used to map the power input parameters through formula 1 to obtain preliminary load output information of the substation. This process makes full use of the characteristics of the input data, and realizes the prediction of the load through the mapping function, hidden state vector and model parameter set of the model. This path independently processes the power input parameters and does not need to rely on the output or feedback of other models, so that the first load prediction model is independent and efficient in processing input data. In addition, since it only involves the processing flow of a single model, the complexity and calculation amount of data processing are relatively low, which helps to speed up the processing speed and reduce the use of computing resources.

[0029] In the second data flow path, Formula 2 combines the power output parameter information and the feedback load forecast information of the first load forecast model, and further integrates the forecast through the second load forecast model. This fusion method not only takes into account the actual output of electric energy, but also integrates the results of the preliminary forecast, thereby improving the comprehensiveness and accuracy of the forecast. Through the connection between the first feedback output port and the second input port, data interaction and collaboration between the two models are realized. This collaboration mechanism enables the system to comprehensively utilize the advantages of different models to perform more complex and accurate load forecasts. In addition, by introducing a feedback mechanism, the second load forecast model can be adjusted and optimized based on the results of the preliminary forecast, further improving the forecast accuracy. At the same time, the addition of power output parameter information also provides more dimensional data support for the forecast, enhancing the robustness of the forecast.

[0030] Optionally, in the first data flow path, the first area load output information is determined using Formula 3, and the Formula 3 is:

[0031] ,

[0032] in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term In the second data flow path, the second area load output information is determined using Formula 4, where Formula 4 is:

[0033] ,

[0034] in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector The second load forecasting model at the current time The hidden state vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation, the second model parameter set The second weight matrix includes And the second bias term , To output information to the first area load Perform vector mapping operations in a preset feature vector mapping table.

[0035] In the above scheme, by introducing the first activation function, Formula 3 realizes the nonlinear mapping of the input data, enhances the complexity and expression ability of the model, and can more accurately capture the nonlinear relationship between the power input parameters and the load of the substation. Then, the power parameter feature vectors are vectorized and spliced ​​using a specific function to achieve feature-level fusion. This fusion method not only takes into account the current power input situation, but also combines the historical state information of the model, which helps to improve the accuracy of the prediction. In addition, the first model parameter set includes a first weight matrix and a first bias, which can be optimized through the training process, so that the model can better adapt to the actual data distribution and improve the prediction accuracy.

[0036] In the second data flow path, Formula 4 uses a specific function to fuse the characteristic vector of the power output parameter, the hidden state vector of the second load forecasting model at the current moment, and the load output information of the first substation after the vector mapping operation. This multi-dimensional data fusion method fully considers the actual output of power, the historical state information of the model, and the feedback of the preliminary prediction results, providing more comprehensive and accurate data support for the prediction. In addition, by introducing another specific function, the vector mapping operation is performed on the load output information of the first substation, and it is passed to the second load forecasting model as feedback load forecasting information. This feedback mechanism enables the second model to be adjusted and optimized based on the results of the preliminary prediction, further improving the prediction accuracy. At the same time, the vector mapping operation helps to convert the preliminary prediction results into a form that is more suitable for processing by the second model, enhancing the collaboration between models.

[0037] Optionally, the substation management node responds to the grid data call instruction, uses a preset substation load forecasting framework, and determines the substation load output information according to the power input parameter information and the power output parameter information, including:

[0038] If the grid data calling instruction is used to call the substation load forecast information to output the forecast operation information of the power transmission network to the downstream power consumption monitoring device of the power transmission network, the power input parameter information is converted into a power input parameter feature vector, and the substation load forecast information is determined through the first data flow path;

[0039] If the power grid data call instruction is used to call the substation load forecast information so as to make a power supply decision operation for the upstream power grid of the transmission network to the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, the electric energy output parameter information is converted into an electric energy output parameter feature vector, and the substation load forecast information is determined through the second data flow path.

[0040] In the above scheme, when the goal of the grid data call instruction is to provide predictive operation information to the downstream power monitoring equipment of the transmission network, the substation management node first converts the power input parameter information (such as voltage and current from the transformer) into a power input parameter feature vector. Subsequently, it uses the first data flow path, that is, directly performs load forecasting through the first load forecasting model to quickly generate preliminary substation load forecasting information. This process ensures that the downstream power monitoring equipment can obtain the operation status forecast of the transmission network in a timely manner, which helps it to conduct more accurate power monitoring and scheduling, and improve the operation efficiency and safety of the power grid.

[0041] When the grid data call instruction is intended to assist the upstream grid in making decisions on the amount of power supplied to the transmission network, the processing flow of the substation management node is more complex and sophisticated. It not only converts the power input parameter information into a feature vector, but also considers the power output parameter information (such as the power consumption of each load node, load fluctuation, etc.), and converts this information into the corresponding feature vector. Subsequently, through the second data flow path, that is, combining the preliminary prediction results of the first load forecasting model (as feedback information) and further processing of the second load forecasting model, a more comprehensive and accurate substation load forecasting information is generated. This process fully considers the supply and demand balance and actual operation of the power grid, provides strong data support for the upstream power grid's power supply decision, and helps to achieve efficient allocation of power resources and stable operation of the power grid.

[0042] In summary, the above solution achieves accurate generation and efficient use of load forecasting information in the substation area by flexibly responding to different types of grid data call instructions and selecting the most appropriate data processing flow and load forecasting model for different application scenarios (downstream power consumption monitoring equipment and upstream grid power supply decision-making). This technical solution not only improves the level of intelligent grid management, but also promotes the optimal allocation of power resources and the stable and efficient operation of the grid, with significant technical effects and application value.

[0043] Optionally, the first load forecasting model may be a load forecasting model obtained by training a recurrent neural network model using a first training data set;

[0044] The second load forecasting model may be a load forecasting model obtained by training a fully connected neural network model using a second training data set.

[0045] In the above scheme, the recurrent neural network is selected as the basis of the first load forecasting model because of its strong modeling ability for sequence data. By training the first training data set (which may contain time series power input parameter information), the recurrent neural network can learn the changing trend and dependency of the power input parameters in the time dimension, so as to more accurately predict the load of the substation in the future. This choice enables the first load forecasting model to show high prediction accuracy and stability when quickly responding to grid data call instructions and providing preliminary load forecasting information.

[0046] In the second load forecasting model, the fully connected neural network is trained using the second training data set (which may contain power output parameter information and feedback information transmitted through the first load forecasting model), which can capture the complex relationship between power output parameters and substation load. By combining the preliminary prediction results of the recurrent neural network model, the fully connected neural network can further refine the prediction results and provide more comprehensive and accurate substation load forecasting information.

[0047] In summary, in the above technical solution, by reasonably selecting different types of deep learning models and training them with corresponding training data sets, the optimization of the substation load forecasting method is achieved. The recurrent neural network model provides a fast and relatively accurate preliminary forecast in the first data flow path, while the fully connected neural network model uses richer information and more complex model structures in the second data flow path to further refine and optimize the preliminary forecast results. This data-driven strategy for model selection and training not only improves the overall accuracy and reliability of substation load forecasting, but also enhances the adaptability and flexibility of the forecasting method, enabling it to better cope with different power grid operation scenarios and forecasting needs.

[0048] In a second aspect, the present application provides a power grid substation management system, comprising: a substation management node, a transformer, a power transmission network, and a load node set, wherein the transformer is arranged at the input end of the power transmission network, and the load nodes in the load node set are arranged at the output end of the power transmission network, and the management node is communicatively connected with the transformer and each load node in the load nodes;

[0049] Acquire power input parameter information through the transformer and send it to the substation management node, wherein the power input parameter information includes the power input parameter of the current time node and the historical power input parameter sequence;

[0050] Acquire power output parameter information through the load node set and send it to the substation management node, wherein the power output parameter information includes the power output parameter of the current time node and the historical power output parameter sequence;

[0051] The substation management node responds to the power grid data call instruction, utilizes the preset substation load forecasting framework, and determines the substation load output information based on the power input parameter information and the power output parameter information. The preset substation load forecasting framework includes multiple load forecasting models trained based on deep learning models, wherein the instruction type of the power grid data call instruction is used to determine the data flow path of the power input parameter information and / or the power output parameter information in the preset substation load forecasting framework.

[0052] Optionally, the preset area load forecasting framework includes a first load forecasting model and a second load forecasting model;

[0053] The first load forecasting model includes a first input port and a first output port, and the first output port includes a first feedback output port and a first result output port;

[0054] The second load forecasting model includes a second input port and a second output port;

[0055] The first feedback output port is connected to the second input port.

[0056] Optionally, the data flow path includes a first data flow path and a second data flow path;

[0057] The first data flow path is a flow path that sequentially passes through the first input port to the first result output port;

[0058] The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port to the second output port in sequence.

[0059] Optionally, in the first data flow path, the first area load output information is determined using Formula 1, where Formula 1 is:

[0060] ,

[0061] in, For the current moment The next moment The first station area load output information, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, The first load forecasting model at the current time The hidden state vector of is a first model parameter set of the first load forecasting model;

[0062] In the second data flow path, the second area load output information is determined using Formula 2, which is:

[0063] ,

[0064] in, For the current moment The next moment The second station area load output information, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load prediction information transmitted to the second input port through the first feedback output port, is a second model parameter set of the second load prediction model. Optionally, in the first data flow path, the first area load output information is determined using Formula 3, and the Formula 3 is:

[0065] ,

[0066] in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term ;

[0067] In the second data flow path, the second area load output information is determined using Formula 4, which is:

[0068] ,

[0069] in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector The second load forecasting model at the current time Hidden state

[0070] State vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation, the second model parameter set The second weight matrix includes And the second bias term , To perform a vector mapping operation on the load output information of the first substation in a preset feature vector mapping table.

[0071] Optionally, the substation management node responds to the grid data call instruction, uses a preset substation load forecasting framework, and determines the substation load output information according to the power input parameter information and the power output parameter information, including:

[0072] If the grid data calling instruction is used to call the substation load forecast information to output the forecast operation information of the power transmission network to the downstream power consumption monitoring device of the power transmission network, the power input parameter information is converted into a power input parameter feature vector, and the substation load forecast information is determined through the first data flow path;

[0073] If the power grid data call instruction is used to call the substation load forecast information so as to make a power supply decision operation for the upstream power grid of the transmission network to the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, the electric energy output parameter information is converted into an electric energy output parameter feature vector, and the substation load forecast information is determined through the second data flow path.

[0074] Optionally, the first load forecasting model may be a load forecasting model obtained by training a recurrent neural network model using a first training data set;

[0075] The second load forecasting model may be a load forecasting model obtained by training a fully connected neural network model using a second training data set.

[0076] In a third aspect, the present application provides an electronic device, including:

[0077] A processor; and a memory for storing executable instructions of the processor;

[0078] The processor is configured to perform any possible method described in the first aspect by executing the executable instructions.

[0079] In a fourth aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement any possible method described in the first aspect.

[0080] The substation load prediction method based on deep learning model provided in the present application obtains power input parameter information through a transformer and sends it to a substation management node, and obtains power output parameter information through a load node set and sends it to the substation management node. Then, the substation management node responds to the power grid data call instruction, uses a preset substation load prediction framework, and determines the substation load prediction information according to the power input parameter information and the power output parameter information. It can determine the corresponding model for prediction processing according to different data call requirements, thereby ensuring that the output prediction data can match the data call requirements while also ensuring the model processing efficiency, that is, ensuring the prediction accuracy while rationally utilizing the processing resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0082] Figure 1 It is a flowchart of a method for predicting load in a substation based on a deep learning model according to an example embodiment of the present application;

[0083] Figure 2 The present application is a flowchart of a method for predicting load in a substation based on a deep learning model according to another exemplary embodiment;

[0084] Figure 3 is a structural schematic diagram of a power grid area management system according to an exemplary embodiment of the present application;

[0085] Figure 4 It is a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present application.

[0086] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION

[0087] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0088] In order to solve the above problems, the embodiments provided in this application are intended to improve the accuracy and efficiency of load forecasting in power grid areas to support intelligent management and optimized operation of power grids. This method is particularly suitable for power grid area management systems, which are composed of area management nodes, transformers, transmission networks, and load node sets. The following is a specific concept of this application:

[0089] The system architecture in the embodiments provided in the present application is developed around the substation management node, which maintains a communication connection with the transformer and each load node in the load node set. The transformer is set at the input end of the power transmission network, responsible for obtaining the power input parameter information, including the power input parameters of the current time node and the historical power input parameter sequence, and sending this information to the substation management node. At the same time, each load node in the load node set is set at the output end of the power transmission network, responsible for obtaining the power output parameter information, including the power output parameters of the current time node and the historical power output parameter sequence, and also sending this information to the substation management node.

[0090] In order to achieve accurate load forecasting in substations, the embodiments provided in the present application design a preset substation load forecasting framework, which includes multiple load forecasting models obtained through deep learning model training. These models can perform prediction processing through different data flow paths according to the input power input parameter information and power output parameter information. Among them, the first load prediction model: uses a recurrent neural network model and is trained through a first training data set. The model focuses on processing power input parameter information and quickly provides preliminary substation load forecasting results. The second load prediction model uses a fully connected neural network model and is trained through a second training data set. The model can further refine the preliminary prediction results and generate more comprehensive and accurate substation load forecasting information in combination with the power output parameter information.

[0091] In addition, the preset area load forecasting framework defines two main data flow paths:

[0092] The first data flow path: data is directly processed by the first load forecasting model to quickly generate preliminary load forecast information for the substation. This path is suitable for grid data call instructions that require a quick response, such as providing forecast operation information to downstream power monitoring equipment in the transmission network.

[0093] Second data flow path: The data is first processed by the first load forecasting model, and the results are input into the second load forecasting model as feedback. The second model combines the power output parameter information and the preliminary forecast results to generate more accurate and comprehensive load forecast information for the substation. This path is suitable for grid data call instructions that require high-precision forecasts, such as assisting the upstream grid in making power supply decisions.

[0094] Through the flexible response of the substation management node to the grid data call instruction, the embodiment provided by this application can select the most appropriate data processing flow and load forecasting model according to different application scenarios (such as downstream power consumption monitoring equipment and upstream power grid power supply decision), so as to achieve accurate generation and efficient use of substation load forecasting information. This technical solution not only improves the intelligent level of grid management, but also promotes the optimal allocation of power resources and the stable and efficient operation of the grid.

[0095] Figure 1 FIG. 1 is a flow chart of a method for predicting load in a substation based on a deep learning model according to an exemplary embodiment of the present application. Figure 1 As shown, the substation load prediction method based on the deep learning model provided in this embodiment includes:

[0096] S101. Obtain power input parameter information through a transformer, and send the power input parameter information to a substation management node.

[0097] The method provided in this embodiment can be applied to a power grid area management system. Among them, the power grid area management system includes an area management node, a transformer, a transmission network and a load node set. The transformer is arranged at the input end of the transmission network, and the load nodes in the load node set are arranged at the output end of the transmission network. The management node is connected to the transformer and each load node in the load node. Specifically, the system mainly includes four core components: an area management node, a transformer, a transmission network and a load node set. The transformer is installed at the input end of the transmission network, responsible for receiving the electric energy input from the external power grid and converting it into a form suitable for transmission in the transmission network. The load node set is distributed at the output end of the transmission network, and each load node is connected to different power-consuming equipment in the power grid, responsible for monitoring and collecting electric energy output parameter information. As the control center of the system, the area management node maintains real-time communication connection with the transformer and all load nodes to ensure the instant transmission and processing of information.

[0098] In this step, electric energy input parameter information can be obtained through the transformer and sent to the substation management node. The electric energy input parameter information includes the electric energy input parameters of the current time node and the historical electric energy input parameter sequence.

[0099] Specifically, when the transformer receives power input from the external power grid, it will record the current power input parameters (such as voltage, current, power factor, etc.) in real time and save the historical power input parameter sequence. This information is sent to the substation management node through the built-in communication module. After receiving this information, the substation management node will perform preliminary processing, such as data cleaning and formatting, to facilitate subsequent analysis and prediction.

[0100] S102: Obtain power output parameter information through the load node set, and send the power output parameter information to the substation management node.

[0101] In this step, power output parameter information is obtained through the load node set, and the power output parameter information is sent to the substation management node. The power output parameter information includes the power output parameters of the current time node and the historical power output parameter sequence.

[0102] Specifically, each load node in the load node set will also record the current power output parameters (such as power consumption, load change rate, etc.) in real time and save the historical power output parameter sequence. This information will also be sent to the substation management node. The substation management node summarizes and organizes these output parameter information to form a complete power grid operation data set.

[0103] S103, the substation management node responds to the grid data call instruction, uses the preset substation load forecasting framework, and determines the substation load forecasting information according to the power input parameter information and the power output parameter information.

[0104] In this step, the substation management node responds to the power grid data call instruction, uses the preset substation load forecasting framework, and determines the substation load forecasting information based on the power input parameter information and the power output parameter information. The preset substation load forecasting framework includes multiple load forecasting models trained based on deep learning models, wherein the instruction type of the power grid data call instruction is used to determine the data flow path of the power input parameter information and / or the power output parameter information in the preset substation load forecasting framework.

[0105] Specifically, the substation management node has a built-in preset substation load forecasting framework, which consists of multiple load forecasting models trained based on deep learning models. These models include but are not limited to recurrent neural networks, long short-term memory networks, fully connected neural networks, etc., each of which is optimized for different forecasting tasks. A clear data flow path is defined within the framework to ensure that data can flow through each model efficiently and accurately.

[0106] When the substation management node receives a grid data call instruction, it will determine the corresponding data processing flow and prediction model according to the specific type of instruction. For example, if the instruction requires providing the downstream power consumption monitoring equipment with predicted operation information of the transmission network, the substation management node will give priority to the first data flow path, that is, directly use the first load prediction model to process the power input parameter information, and quickly generate preliminary substation load prediction results. If the instruction requires a more accurate prediction result to assist the upstream power grid in making power supply decisions, the substation management node will use the second data flow path, that is, first make a preliminary prediction through the first load prediction model, and then input the prediction result as feedback into the second load prediction model, and generate the final prediction result in combination with the power output parameter information.

[0107] After completing the load forecast for the substation, the substation management node will send the forecast information to the designated recipient through the communication network. For downstream power monitoring equipment, this forecast information helps them adjust power consumption plans in advance and optimize load distribution; for upstream power grids, it helps them formulate reasonable power supply plans and ensure the stable operation of the power grid. In addition, the forecast information can also be used in the operation and maintenance management of the power grid, fault warning and other aspects, further improving the intelligence level and operation efficiency of the power grid.

[0108] In this embodiment, electric energy input parameter information is obtained through the transformer and sent to the substation management node, and electric energy output parameter information is obtained through the load node set and sent to the substation management node. Then, the substation management node responds to the power grid data call instruction, uses the preset substation load prediction framework, and determines the substation load prediction information according to the electric energy input parameter information and the electric energy output parameter information, so as to determine the corresponding model for prediction processing according to different data call requirements, thereby ensuring that the output prediction data can match the data call requirements while ensuring the model processing efficiency, that is, while ensuring the prediction accuracy, it can also reasonably utilize the processing resources.

[0109] Figure 2 The present application is a flowchart of a method for predicting load in a substation based on a deep learning model according to another exemplary embodiment. Figure 2 As shown, the substation load prediction method based on the deep learning model provided in this embodiment includes:

[0110] S201. Obtain power input parameter information through a transformer, and send the power input parameter information to a substation management node.

[0111] In this step, electric energy input parameter information can be obtained through the transformer and sent to the substation management node. The electric energy input parameter information includes the electric energy input parameters of the current time node and the historical electric energy input parameter sequence.

[0112] Specifically, when the transformer receives power input from the external power grid, it will record the current power input parameters (such as voltage, current, power factor, etc.) in real time and save the historical power input parameter sequence. This information is sent to the substation management node through the built-in communication module. After receiving this information, the substation management node will perform preliminary processing, such as data cleaning and formatting, to facilitate subsequent analysis and prediction.

[0113] S202: Obtain power output parameter information through the load node set, and send the power output parameter information to the substation management node.

[0114] In this step, power output parameter information is obtained through the load node set, and the power output parameter information is sent to the substation management node. The power output parameter information includes the power output parameters of the current time node and the historical power output parameter sequence.

[0115] Specifically, each load node in the load node set will also record the current power output parameters (such as power consumption, load change rate, etc.) in real time and save the historical power output parameter sequence. This information will also be sent to the substation management node. The substation management node summarizes and organizes these output parameter information to form a complete power grid operation data set.

[0116] S203: The substation management node responds to the power grid data call instruction and determines a data flow path in a preset substation load forecasting framework according to the instruction type of the power grid data call instruction.

[0117] Specifically, the above-mentioned preset area load forecasting framework includes a first load forecasting model and a second load forecasting model;

[0118] The first load forecasting model includes a first input port and a first output port, and the first output port includes a first feedback output port and a first result output port;

[0119] The second load forecasting model includes a second input port and a second output port;

[0120] The first feedback output port is connected to the second input port.

[0121] Specifically, the first load forecasting model includes a first input port and a first output port (further subdivided into a first feedback output port and a first result output port). The model is mainly responsible for processing the power input parameter information, including the power input parameters of the current time node and the historical power input parameter sequence. Through the training of the deep learning algorithm, the model can capture the timing characteristics and change trends of the input data and generate preliminary load forecasting results for the substation. The output of the first load forecasting model is divided into two parts: one part is the preliminary forecast result, which is output through the first result output port; the other part is the feedback of the forecast result, which is passed to the second load forecasting model through the first feedback output port.

[0122] The second load forecasting model includes a second input port and a second output port. The model is responsible for receiving feedback forecast information from the first load forecasting model, and further forecasting processing is performed in combination with the power output parameter information (including the power output parameters of the current time node and the historical power output parameter sequence). Through the deep learning algorithm, the model can integrate the data of the input and output ends to improve the accuracy and robustness of the forecast. In addition to the power output parameter information directly from the load node, the input of the second load forecasting model also includes the feedback forecast information transmitted by the first load forecasting model.

[0123] When the grid data call instruction only requires preliminary prediction results (for example, for rapid response to certain emergency needs), the power input parameter information enters the first load prediction model through the first input port. After the model processes these data, it directly outputs the preliminary prediction results through the first result output port.

[0124] When the grid data call instruction requires a more accurate prediction result (for example, for formulating a long-term power supply plan or conducting a detailed load analysis), the power input parameter information first enters the first load prediction model through the first input port. After the model processes and generates a preliminary prediction result, the result is passed to the second load prediction model through the first feedback output port. At the same time, the power output parameter information also enters the second load prediction model. The second load prediction model combines these two parts of information, performs further prediction processing through a deep learning algorithm, and finally outputs a more accurate prediction result through the second output port.

[0125] Further, the data flow path includes a first data flow path and a second data flow path;

[0126] The first data flow path is a flow path that sequentially passes through the first input port to the first result output port;

[0127] The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port and the second output port in sequence.

[0128] Specifically, the first data flow path is mainly used to quickly generate preliminary load forecast information for the substation. The path starts from the first input port of the first load forecast model and ends at the first result output port. The specific steps are as follows:

[0129] Data input: When the grid data call instruction requires the rapid generation of preliminary prediction results, the substation management node first receives the power input parameter information from the transformer. This information includes the power input parameters of the current time node and the historical power input parameter sequence.

[0130] Data processing: The power input parameter information is fed into the first input port of the first load forecasting model. Inside the model, this information is converted into a power input parameter feature vector and processed by the mapping function of the deep learning algorithm.

[0131] Preliminary prediction: Using the hidden state vector and model parameter set (including weight matrix and bias term) of the first load prediction model, the characteristic vector of the electric energy input parameter is mapped and calculated to obtain the preliminary area load prediction information at the next moment from the current moment.

[0132] Result output: The preliminary prediction results are output through the first result output port of the first load prediction model, and the power supply network management system performs a quick response or preliminary analysis.

[0133] The second data flow path is based on the first data flow path, adding data feedback and further processing to generate more accurate load forecast information for the substation area. The path starts from the first input port of the first load forecast model, passes through the first feedback output port, then to the second input port of the second load forecast model, and finally ends through the second output port. The specific steps are as follows:

[0134] Data input and preliminary processing: Same as the first data flow path, first receive and process the power input parameter information from the transformer, and make a preliminary prediction through the first load prediction model.

[0135] Data feedback: The preliminary prediction results are not only output through the first result output port, but also passed to the second load prediction model through the first feedback output port. At the same time, the substation management node also receives power output parameter information from the load node set, including the power output parameters of the current time node and the historical power output parameter sequence.

[0136] Data fusion and further processing: The preliminary prediction results (as feedback load prediction information) are fused with the power output parameter feature vector within the second load forecasting model. More complex calculations and analysis are performed using the mapping function of the deep learning algorithm, the hidden state vector of the second load forecasting model, and the model parameter set (including the second weight matrix and the second bias term).

[0137] Accurate prediction: By integrating input data and feedback data, the second load forecasting model generates more accurate load forecast information for the area at the next moment after the current moment.

[0138] Result output: The final prediction result is output through the second output port of the second load forecasting model, providing more comprehensive and accurate load forecasting support for the power grid management system.

[0139] S204: Convert the electric energy input parameter information into an electric energy input parameter feature vector, and determine the substation load prediction information through the first data flow path.

[0140] In this step, if the power grid data call instruction is used to call the substation load forecast information to output the predicted operation information of the transmission network to the downstream power consumption monitoring equipment of the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, and the substation load forecast information is determined through the first data flow path.

[0141] Specifically, in the first data flow path, the load output information of the first area is determined using Formula 1, where Formula 1 is:

[0142] ,

[0143] in, For the current moment The next moment The load output information of the first area, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, For the first load forecasting model at the current moment The hidden state vector of It is a first model parameter set of the first load forecasting model.

[0144] In the first data flow path, the first load prediction model is used to map the power input parameters through formula 1 to obtain preliminary load output information of the substation. This process makes full use of the characteristics of the input data and realizes the prediction of the load through the mapping function, hidden state vector and model parameter set of the model. This path independently processes the power input parameters and does not need to rely on the output or feedback of other models, so that the first load prediction model is independent and efficient in processing input data. In addition, since it only involves the processing flow of a single model, the complexity and calculation amount of data processing are relatively low, which helps to speed up the processing speed and reduce the use of computing resources.

[0145] Furthermore, in the first data flow path, the first area load output information is determined using Formula 3, and the Formula 3 is:

[0146] ,

[0147] in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term .

[0148] By introducing the first activation function (such as Sigmoid, ReLU, etc.), Formula 3 realizes the nonlinear mapping of the input data, enhances the complexity and expression ability of the model, and can more accurately capture the nonlinear relationship between the power input parameters and the load of the substation. Then, the power parameter feature vectors are vectorized and spliced ​​using a specific function to achieve feature-level fusion. This fusion method not only takes into account the current power input situation, but also combines the historical state information of the model, which helps to improve the accuracy of the prediction. In addition, the first model parameter set includes a first weight matrix and a first bias, which can be optimized through the training process, so that the model can better adapt to the actual data distribution and improve the prediction accuracy.

[0149] S205: Convert the power input parameter information into a power input parameter feature vector, convert the power output parameter information into a power output parameter feature vector, and determine the load forecast information of the substation through the second data flow path.

[0150] If the power grid data call instruction is used to call the substation load forecast information to make a power supply decision operation for the upstream power grid of the transmission network to the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, the electric energy output parameter information is converted into an electric energy output parameter feature vector, and the substation load forecast information is determined through a second data flow path.

[0151] In the second data flow path, the load output information of the second area is determined using Formula 2, which is:

[0152] ,

[0153] in, For the current moment The next moment The load output information of the second area, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load forecast information transmitted to the second input port through the first feedback output port, A second model parameter set of the second load forecasting model.

[0154] In the second data flow path, Formula 2 combines the power output parameter information and the feedback load forecast information of the first load forecast model, and further integrates the forecast through the second load forecast model. This fusion method not only takes into account the actual output of power, but also integrates the results of the preliminary forecast, thereby improving the comprehensiveness and accuracy of the forecast. Through the connection between the first feedback output port and the second input port, data interaction and collaboration between the two models are realized. This collaboration mechanism enables the system to comprehensively utilize the advantages of different models to perform more complex and accurate load forecasts. In addition, by introducing a feedback mechanism, the second load forecast model can be adjusted and optimized based on the results of the preliminary forecast, further improving the forecast accuracy. At the same time, the addition of power output parameter information also provides more dimensional data support for the forecast, enhancing the robustness of the forecast.

[0155] Furthermore, in the second data flow path, the load output information of the second area is determined using Formula 4, which is:

[0156] ,

[0157] in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector , the second load forecasting model at the current time The hidden state vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation and set of second model parameters Including the second weight matrix and the second bias term , To output information for the first area load Perform vector mapping operations in a preset feature vector mapping table.

[0158] In the second data flow path, Formula 4 uses a specific function to fuse the characteristic vector of the power output parameter, the hidden state vector of the second load forecasting model at the current moment, and the load output information of the first substation after the vector mapping operation. This multi-dimensional data fusion method fully considers the actual output of power, the historical state information of the model, and the feedback of the preliminary prediction results, providing more comprehensive and accurate data support for the prediction. In addition, by introducing another specific function, the vector mapping operation is performed on the load output information of the first substation, and it is passed to the second load forecasting model as feedback load forecasting information. This feedback mechanism enables the second model to be adjusted and optimized based on the results of the preliminary prediction, further improving the prediction accuracy. At the same time, the vector mapping operation helps to convert the preliminary prediction results into a form that is more suitable for processing by the second model, enhancing the collaboration between models.

[0159] On the basis of the above solution, the first load forecasting model may be a load forecasting model obtained by training a recurrent neural network model using the first training data set;

[0160] The second load forecasting model may be a load forecasting model obtained by training a fully connected neural network model using a second training data set.

[0161] Specifically, the first load forecasting model uses a recurrent neural network as the basic architecture. Because the recurrent neural network can process sequence data and capture the temporal dependencies in the data, it is very suitable for processing time series data such as power input parameter information.

[0162] The training process of the first load forecasting model:

[0163] Data preparation: Collect the first training data set, which contains a large amount of power input parameter information, including parameter values ​​such as voltage and current at different time nodes, and corresponding historical sequence data.

[0164] Data preprocessing: Perform preprocessing operations such as cleaning, denoising, and normalization on the collected data to ensure data quality and accelerate the training process.

[0165] Model construction: Construct the network structure of the first load forecasting model, including input layer, hidden layer and output layer. Set the appropriate number of hidden layers and nodes, as well as parameters such as activation function.

[0166] Model training: Use the preprocessed first training data set to train the model. Adjust the model parameters through the back propagation algorithm to minimize the prediction error until the model converges.

[0167] Model evaluation: Evaluate the performance of the trained model on an independent test set, including indicators such as accuracy and recall, to ensure that the model has good generalization ability.

[0168] The second load forecasting model selects a fully connected neural network as the basic architecture. The fully connected neural network is suitable for combining power output parameter information and preliminary forecast results for further analysis.

[0169] The training process of the second load forecasting model:

[0170] Data preparation and preprocessing: Collect the second training data set, which, in addition to the power output parameter information, also needs to include the preliminary prediction results obtained by the first load prediction model as feedback information. Perform the same preprocessing operation on the data.

[0171] Model construction: Based on the basic principle of fully connected neural network, the network structure of the second load forecasting model is constructed. Multiple fully connected layers are set as needed, and appropriate activation functions are selected.

[0172] Model training: The preprocessed second training data set (including power output parameter information and feedback information) is input into the model for training. The model parameters are adjusted to minimize the prediction error until the model converges.

[0173] Model evaluation and optimization: Evaluate model performance on the test set and optimize the model based on the evaluation results to improve prediction accuracy and stability.

[0174] Through the above steps, we have successfully constructed the first load forecasting model based on recurrent neural network and the second load forecasting model based on fully connected neural network. These two models work together to give full play to their respective advantages in different data flow paths and jointly achieve high-precision and high-efficiency load forecasting in substations.

[0175] It is worth noting that the above-mentioned first load prediction model and the second load prediction model can also be based on any two deep learning prediction models that have been trained in the prior art. However, it needs to be ensured that the first load prediction model is smaller than the second load prediction model, that is, the first load prediction model requires fewer input parameters and / or less prediction process processing volume than the second load prediction model.

[0176] Figure 3 FIG. 1 is a schematic diagram of a power grid area management system according to an exemplary embodiment of the present application. Figure 3 As shown, the power grid area management system 300 provided in this embodiment includes:

[0177] A substation management node 310, a transformer 320, a power transmission network 330, and a load node set 340, wherein the transformer 320 is arranged at the input end of the power transmission network 330, and the load nodes in the load node set 340 are arranged at the output end of the power transmission network 330, and the management node is communicatively connected with the transformer 320 and each of the load nodes;

[0178] Acquire power input parameter information through the transformer 320, and send the power input parameter information to the substation management node 310, wherein the power input parameter information includes the power input parameter of the current time node and the historical power input parameter sequence;

[0179] Acquire power output parameter information through the load node set 340, and send the power output parameter information to the substation management node 310, wherein the power output parameter information includes the power output parameter of the current time node and the historical power output parameter sequence;

[0180] The substation management node 310 responds to the power grid data call instruction, utilizes the preset substation load forecasting framework, and determines the substation load output information based on the power input parameter information and the power output parameter information. The preset substation load forecasting framework includes multiple load forecasting models trained based on deep learning models, wherein the instruction type of the power grid data call instruction is used to determine the data flow path of the power input parameter information and / or the power output parameter information in the preset substation load forecasting framework.

[0181] Optionally, the preset area load forecasting framework includes a first load forecasting model and a second load forecasting model;

[0182] The first load forecasting model includes a first input port and a first output port, and the first output port includes a first feedback output port and a first result output port;

[0183] The second load forecasting model includes a second input port and a second output port;

[0184] The first feedback output port is connected to the second input port.

[0185] Optionally, the data flow path includes a first data flow path and a second data flow path;

[0186] The first data flow path is a flow path that sequentially passes through the first input port to the first result output port;

[0187] The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port to the second output port in sequence.

[0188] Optionally, in the first data flow path, the first area load output information is determined using Formula 1, where Formula 1 is:

[0189] ,

[0190] in, For the current moment The next moment The first station area load output information, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, The first load forecasting model at the current time The hidden state vector of is the first model parameter set of the first load prediction model; in the second data flow path, the load output information of the second area is determined using formula 2, and the formula 2 is:

[0191] ,

[0192] in, For the current moment The next moment The second station area load output information, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load prediction information transmitted to the second input port through the first feedback output port, A second model parameter set of the second load forecasting model.

[0193] Optionally, in the first data flow path, the first area load output information is determined using Formula 3, and the Formula 3 is:

[0194] ,

[0195] in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term ;

[0196] In the second data flow path, the second area load output information is determined using Formula 4, which is:

[0197] ,

[0198] in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector The second load forecasting model at the current time The hidden state vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation, the second model parameter set The second weight matrix includes And the second bias term , To output information to the first area load Perform vector mapping operations in a preset feature vector mapping table.

[0199] Optionally, the substation management node 310 responds to the grid data call instruction, uses a preset substation load forecasting framework, and determines the substation load output information according to the power input parameter information and the power output parameter information, including:

[0200] If the grid data calling instruction is used to call the substation load forecast information to output the forecast operation information of the power transmission network 330 to the downstream power consumption monitoring device of the power transmission network 330, the power input parameter information is converted into a power input parameter feature vector, and the substation load forecast information is determined through the first data flow path;

[0201] If the power grid data call instruction is used to call the substation load forecast information so as to make a power supply decision operation for the upstream power grid of the transmission network 330 to the transmission network 330, the power input parameter information is converted into a power input parameter feature vector, the power output parameter information is converted into a power output parameter feature vector, and the substation load forecast information is determined through the second data flow path.

[0202] Optionally, the first load forecasting model may be a load forecasting model obtained by training a recurrent neural network model using a first training data set;

[0203] The second load forecasting model may be a load forecasting model obtained by training a fully connected neural network model using a second training data set.

[0204] Figure 4 is a schematic diagram of the structure of an electronic device according to an exemplary embodiment of the present application. Figure 4 As shown, an electronic device 400 provided in this embodiment includes: a processor 401 and a memory 402; wherein:

[0205] The memory 402 is used to store computer programs, and the memory may also be a flash memory.

[0206] The processor 401 is used to execute the execution instructions stored in the memory to implement each step in the above method. For details, please refer to the relevant description in the above method embodiment.

[0207] Optionally, the memory 402 may be independent or integrated with the processor 401 .

[0208] When the memory 402 is a device independent of the processor 401, the electronic device 400 may further include:

[0209] The bus 403 is used to connect the memory 402 and the processor 401 .

[0210] This embodiment further provides a readable storage medium, in which a computer program is stored. When at least one processor of an electronic device executes the computer program, the electronic device executes the methods provided in the above-mentioned various implementation modes.

[0211] This embodiment also provides a program product, which includes a computer program stored in a readable storage medium. At least one processor of the electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device implements the methods provided in the above various embodiments.

[0212] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary techniques in the art that are not disclosed in the present application. The specification and examples are intended to be exemplary only, and the true scope and spirit of the present application are indicated by the claims.

[0213] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for predicting load in a substation based on a deep learning model, applied to a power grid substation management system, characterized in that: The power grid area management system includes an area management node, a transformer, a power transmission network, and a load node set, wherein the transformer is arranged at the input end of the power transmission network, and the load nodes in the load node set are arranged at the output end of the power transmission network, and the management node is communicatively connected with the transformer and each load node in the load nodes; the method includes: Acquire power input parameter information through the transformer and send it to the substation management node, wherein the power input parameter information includes the power input parameter of the current time node and the historical power input parameter sequence; Acquire power output parameter information through the load node set and send it to the substation management node, wherein the power output parameter information includes the power output parameter of the current time node and the historical power output parameter sequence; If the grid data call instruction is used to call the substation load forecast information to output forecast operation information to the downstream power consumption monitoring device of the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, and the substation load forecast information is determined by a first data flow path in a preset substation load forecast framework, wherein a first load forecast model corresponding to the first data flow path includes a first input port and a first output port, the first output port includes a first feedback output port and a first result output port, and the first data flow path is a flow path that passes through the first input port to the first result output port in sequence; If the power grid data call instruction is used to call the substation load forecast information to make a power supply decision operation for the upstream power grid of the transmission network to the transmission network, the power input parameter information is converted into a power input parameter feature vector, and the power output parameter information is converted into a power output parameter feature vector. The substation load forecast information is determined through a second data flow path in the preset substation load forecast framework. The second load forecast model corresponding to the second data flow path includes a second input port and a second output port. The first feedback output port is connected to the second input port. The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port to the second output port in sequence.

2. The method for predicting load in a substation based on a deep learning model according to claim 1, characterized in that: The preset substation load forecasting framework includes the first load forecasting model and the second load forecasting model.

3. The method for predicting load in a substation based on a deep learning model according to claim 1, characterized in that: In the first data flow path, the first area load output information is determined using Formula 1, which is: ， in, For the current moment The next moment The load output information of the first station area, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, The first load forecasting model at the current time The hidden state vector of is a first model parameter set of the first load forecasting model; In the second data flow path, the second area load output information is determined using Formula 2, which is: ， in, For the current moment The next moment The second station area load output information, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load prediction information transmitted to the second input port through the first feedback output port, A second model parameter set of the second load forecasting model.

4. The method for predicting load in a substation based on a deep learning model according to claim 3, characterized in that: In the first data flow path, the first area load output information is determined using Formula 3, which is: ， in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term ; In the second data flow path, the second area load output information is determined using Formula 4, which is: ， in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector The second load forecasting model at the current time The hidden state vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation, the second model parameter set Including the second weight matrix And the second bias term , To output information about the load of the first area Perform vector mapping operations in a preset feature vector mapping table.

5. The method for predicting load in a substation based on a deep learning model according to claim 4, characterized in that: The first load forecasting model is a load forecasting model obtained by training a recurrent neural network model using a first training data set; The second load forecasting model is a load forecasting model obtained by training a fully connected neural network model using a second training data set.

6. A power grid area management system, characterized in that: include: An area management node, a transformer, a power transmission network, and a set of load nodes, wherein the transformer is arranged at the input end of the power transmission network, and the load nodes in the set of load nodes are arranged at the output end of the power transmission network, and the management node is communicatively connected with the transformer and each load node in the load nodes; Acquire power input parameter information through the transformer and send it to the substation management node, wherein the power input parameter information includes the power input parameter of the current time node and the historical power input parameter sequence; Acquire power output parameter information through the load node set and send it to the substation management node, wherein the power output parameter information includes the power output parameter of the current time node and the historical power output parameter sequence; If the grid data call instruction is used to call the substation load forecast information to output the forecast operation information to the downstream power consumption monitoring device of the transmission network, the electric energy input parameter information is converted into an electric energy input parameter feature vector, and the substation load forecast information is determined by a first data flow path in a preset substation load forecast framework, the first load forecast model corresponding to the first data flow path includes a first input port and a first output port, the first output port includes a first feedback output port and a first result output port, and the first data flow path is a flow path that passes through the first input port to the first result output port in sequence; If the power grid data call instruction is used to call the substation load forecast information to make a power supply decision operation for the upstream power grid of the transmission network to the transmission network, the power input parameter information is converted into a power input parameter feature vector, and the power output parameter information is converted into a power output parameter feature vector. The substation load forecast information is determined through a second data flow path in the preset substation load forecast framework. The second load forecast model corresponding to the second data flow path includes a second input port and a second output port. The first feedback output port is connected to the second input port. The second data flow path is a flow path that passes through the first input port, the first feedback output port, the second input port to the second output port in sequence.

7. The power grid area management system according to claim 6, characterized in that: The preset substation load prediction framework includes a first load prediction model and a second load prediction model.

8. The power grid area management system according to claim 7, characterized in that: In the first data flow path, the first area load output information is determined using Formula 1, which is: ， in, For the current moment The next moment The load output information of the first station area, is the mapping function of the first load forecasting model, is the electric energy input parameter feature vector corresponding to the electric energy input parameter information, The first load forecasting model at the current time The hidden state vector of is a first model parameter set of the first load forecasting model; In the second data flow path, the second area load output information is determined using Formula 2, which is: ， in, For the current moment The next moment The second station area load output information, is the mapping function of the second load forecasting model, is the electric energy output parameter feature vector corresponding to the electric energy output parameter information, The second load forecasting model is at the current moment The hidden state vector of is the feedback load prediction information transmitted to the second input port through the first feedback output port, A second model parameter set of the second load forecasting model.

9. The power grid area management system according to claim 8, characterized in that: In the first data flow path, the first area load output information is determined using Formula 3, which is: ， in, is the first activation function, is the first weight matrix of the first load forecasting model, is the first bias term of the first load forecasting model, Used to convert the electric energy parameter characteristic vector With the first load forecasting model at the current moment The hidden state vector Perform vector concatenation, the first model parameter set Including the first weight matrix And the first bias term ; In the second data flow path, the second area load output information is determined using Formula 4, which is: ， in, is the second activation function, is the second weight matrix of the second load forecasting model, is the second bias term of the second load forecasting model, Used to convert the electric energy output parameter characteristic vector The second load forecasting model at the current time The hidden state vector and feedback load forecast information transmitted to the second input port through the first feedback output port Perform vector concatenation, the second model parameter set The second weight matrix includes And the second bias term , To output information to the first area load Perform vector mapping operations in a preset feature vector mapping table.

10. An electronic device, characterized in that: include: processor; and, a memory for storing executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 5 by executing the executable instructions.

11. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 5 when executed by a processor.

Images