Power grid load prediction system and method based on deep learning

Abstract

The invention relates to the technical field of power grid load prediction, in particular to a power grid load prediction system and method based on deep learning, and the system comprises a data collection module which is used for carrying out the real-time collection of power grid load operation data; the data preprocessing module is used for preprocessing the power grid load operation data; the data analysis module is used for performing data reconstruction analysis on the actual power grid load operation data; the deep learning module is used for constructing a power grid load prediction model according to the deep learning model and optimizing the construction process of the power grid load prediction model according to the power grid load characteristic data to obtain an optimized power grid load prediction model; and the load prediction module is used for predicting the power grid load according to the optimized power grid load prediction model to obtain a power grid load prediction value. According to the invention, the power grid load is predicted according to the deep learning model, and the accuracy of power grid load prediction is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power grid load forecasting, and in particular, to a power grid load forecasting system and method based on deep learning. Background Art

[0002] In the operation and management of power systems, power grid load forecasting is a core link to ensure the balance between power supply and demand, optimize dispatching strategies, and ensure the safe and stable operation of the power grid. However, when facing the increasingly complex and changeable power grid load changes, the prediction accuracy of existing power grid load forecasting methods is often low.

[0003] Chinese Patent Publication No. CN114066073A discloses a power grid load forecasting method, including: performing parallel abnormal data detection on historical period load data, and removing abnormal data in the historical period load data; classifying the historical period load data after removing abnormal data into different types through clustering analysis, and determining the typical load data corresponding to each type; training according to the historical period load impact data and the corresponding type to obtain a load classification model; inputting the load impact data of the period to be predicted into the load classification model to determine the type corresponding to the period to be predicted and the typical load data corresponding to the period to be predicted; performing load forecasting on the period to be predicted according to the typical load data corresponding to the period to be predicted and the historical period load data after removing abnormal data, but this solution cannot predict the power grid load based on a deep learning model, resulting in low accuracy of power grid load forecasting. Summary of the Invention

[0004] Therefore, the present invention provides a power grid load forecasting system and method based on deep learning to overcome the problem in the prior art that the accuracy of power grid load forecasting is low because the power grid load cannot be predicted based on a deep learning model.

[0005] To achieve the above object, on the one hand, the present invention provides a power grid load forecasting system based on deep learning, the system includes: A data acquisition module for real-time acquisition of power grid load operation data; A data preprocessing module for preprocessing the power grid load operation data to obtain actual power grid load operation data; A data analysis module for performing data reconstruction analysis on the actual power grid load operation data to obtain power grid load characteristic data; A deep learning module for constructing a power grid load forecasting model according to a deep learning model, and also for optimizing the construction process of the power grid load forecasting model according to the power grid load characteristic data to obtain an optimized power grid load forecasting model; A load forecasting module is used to forecast the grid load according to the optimized grid load forecasting model to obtain the grid load forecasting value, and is also used to output the grid control means according to the grid load forecasting value, and is further used to optimize the grid control means according to the grid load characteristic data and the grid load forecasting value.

[0006] Further, the data preprocessing module detects outliers in the grid load operation data according to the data anomaly detection algorithm to obtain corresponding data outliers, and processes the data outliers according to the data outlier processing scheme corresponding to the grid load operation data to obtain corresponding outlier processing results, and performs data conversion on the outlier processing results according to the data conversion scheme to obtain corresponding actual grid load operation data.

[0007] Further, the data analysis module decomposes the actual grid load operation data according to the data decomposition method to obtain a data decomposition result, and performs a matching analysis on the data decomposition result and the preset data decomposition standard according to the data matching analysis method to obtain an actual matching degree DT, and compares the actual matching degree DT with the preset matching degree DT0, and performs a data reconstruction analysis on the actual grid load operation data according to the comparison result to obtain grid load characteristic data, where: When DT≥DT0, the data analysis module performs a data reconstruction analysis on the actual grid load operation data according to the grid load data reconstruction strategy to obtain grid load characteristic data; When DT<DT0, the data analysis module does not perform a data reconstruction analysis on the actual grid load operation data.

[0008] Further, the deep learning module comprehensively evaluates the grid load characteristic data according to the data comprehensive evaluation method to obtain an actual comprehensive evaluation value PG, and compares the actual comprehensive evaluation value PG with the preset comprehensive evaluation value PG0, and judges the data reconstruction analysis situation of the grid load characteristic data according to the comparison result, and constructs a grid load forecasting model according to the judgment result, where: When PG<PG0, the deep learning module determines that the data reconstruction analysis situation of the grid load characteristic data is a non-compliant situation, and the deep learning module does not construct a grid load forecasting model; When PG≥PG0, the deep learning module determines that the data reconstruction analysis situation of the grid load characteristic data is a compliant situation, and constructs a grid load forecasting model according to the preset grid load forecasting data set.

[0009] Further, when constructing the power grid load prediction model based on the preset power grid load prediction data set, the deep learning module sets the deep learning model as a convolutional neural network model according to the type and characteristics of the preset power grid load prediction data set, and sets the convolutional layer in the convolutional neural network model to process the preset power grid load prediction data set to obtain a convolutional layer data output feature map; The deep learning module sets the pooling layer in the convolutional neural network model to process the preset power grid load prediction data set according to the type and characteristics of the preset power grid load prediction data set to obtain pooling layer data features; The deep learning module connects the convolutional layer data output feature map and the pooling layer data features to the fully connected layer of the convolutional neural network model for data feature extraction to obtain a fully connected layer data feature result, and identifies the fully connected layer data feature result according to the power grid load fault sequence recognition model to obtain power grid load fault sequence feature data, and sets the power grid load fault sequence feature data as the first time series, and sets the actual power grid load data set corresponding to the power grid load fault sequence feature data as the second time series , and analyzes the first time series and the second time series according to the dynamic time warping algorithm to obtain a time series similarity set , where represents the time series similarity at the k-th moment; The deep learning module and the time series similarity set calculate the power grid load prediction value , and set , where represents the i-th time series similarity, represents the i-th actual power grid load data, and k represents the total number of time series set when performing the dynamic time warping algorithm; The deep learning module adds the power grid load prediction value to the preset power grid load prediction data set to obtain a new power grid load prediction data set, and divides the new power grid load prediction data set into a 70% analysis training set, a 20% analysis validation set, and a 10% analysis test set, inputs the analysis training set into the convolutional neural network model to train the convolutional neural network model, inputs the analysis validation set into the trained convolutional neural network model to iteratively optimize the trained convolutional neural network model, inputs the analysis test set into the iteratively optimized convolutional neural network model to analyze and test the iteratively optimized convolutional neural network model, and outputs the convolutional neural network model that meets the preset analysis test accuracy as the power grid load prediction model.

[0010] Further, the deep learning module processes the power grid load characteristic data according to the outlier-optimized detection method to obtain an outlier-optimized detection result, obtains the outlier frequency f1 according to the outlier-optimized detection result, compares the outlier frequency f1 with a preset outlier frequency f0, and optimizes the construction process of the power grid load prediction model according to the comparison result to obtain an optimized power grid load prediction model, where: When f1 < f0, the deep learning model module does not optimize the construction process of the power grid load prediction model; When f1 ≥ f0, the deep learning model module repairs the outlier-optimized detection result according to the data repair method to obtain a data repair result, and adds the data repair result to a new power grid load prediction dataset to optimize the construction process of the power grid load prediction model to obtain an optimized power grid load prediction model.

[0011] Further, the load prediction module obtains the recall rate Recall and the precision Precison according to the optimized power grid load prediction model, calculates the model evaluation index value FS according to the recall rate Recall and the precision Precison, sets and compares the model evaluation index value FS with a preset model evaluation index value FS0, judges the training situation of the optimized power grid load prediction model according to the comparison result, and predicts the power grid load according to the judgment result to obtain a power grid load prediction value, where: When FS ≥ FS0, the load prediction module determines that the training situation of the optimized power grid load prediction model is a qualified situation, and inputs the power grid load characteristic data into the optimized power grid load prediction model determined to be in a qualified situation for prediction to obtain a power grid load prediction value; When FS < FS0, the load prediction module determines that the training situation of the optimized power grid load prediction model is an unqualified situation, and the load prediction module does not predict the power grid load.

[0012] Further, the load prediction module compares the power grid load prediction value with a preset power grid load prediction value When > the load prediction module determines that the power grid load prediction situation is a power grid prediction overload state, and outputs the power grid regulation means according to the power grid prediction overload state; When ≤ When the above conditions are met, the load prediction module determines that the power grid load prediction situation is in a normal state, and the load prediction module does not output power grid control means.

[0013] Further, the load prediction module obtains the actual value of the i-th power grid load according to the power grid load characteristic data , and calculates the mean square error MSE according to the predicted value of the i-th power grid load and the actual value of the i-th power grid load . Set , where n represents the number of samples, and compare the mean square error MSE with the preset mean square error MSE0, judge the calculation situation of the mean square error according to the comparison result, and optimize the power grid control means according to the judgment result, where: When MSE ≤ MSE0, the load prediction module determines that the calculation situation of the mean square error is normal, and the deep learning module does not optimize the power grid control means; When MSE > MSE0, the load prediction module determines that the calculation situation of the mean square error is abnormal. The load prediction module calculates the environmental index HJ according to the temperature W1, wind speed S1, and sunshine duration R1 in the power grid load characteristic data. Set , where 1 / W0 represents the weight coefficient for adjusting the temperature W1, 1 / S0 represents the weight coefficient for adjusting the wind speed S1, 1 / R0 represents the weight coefficient for adjusting the sunshine duration R1. Set 5°C ≤ W0 ≤ 40°C, 25 m / s ≤ S0 ≤ 40 m / s, 1000 < R0 < 4000, and compare the environmental index HJ with the preset environmental index HJ0, judge the power grid environmental factor situation according to the comparison result, and optimize the power grid control means according to the judgment result, where: If HJ ≤ HJ0, the load prediction module determines that the power grid environmental factor situation is normal, and the deep learning module does not optimize the power grid control means; If HJ > HJ0, the load prediction module determines that the power grid environmental factor situation is abnormal. The load prediction module optimizes the mean square error MSE according to the environmental factor optimization parameter b1 to obtain the optimized mean square error , and optimizes the power grid control means to the power grid control optimization means according to the optimized mean square error , where .

[0014] On the other hand, the present invention also provides a power grid load prediction method based on deep learning. The method includes: Step S1, collect the real-time operation data of the power grid load, and preprocess the operation data of the power grid load to obtain the actual operation data of the power grid load; Step S2, perform data reconstruction analysis on the actual operation data of the power grid load to obtain the power grid load characteristic data; Step S3, construct a power grid load prediction model according to the deep learning model, and optimize the construction process of the power grid load prediction model according to the power grid load characteristic data to obtain an optimized power grid load prediction model; Step S4, predict the power grid load according to the optimized power grid load prediction model to obtain the power grid load prediction value, output the power grid regulation means according to the power grid load prediction value, and optimize the power grid regulation means according to the power grid load characteristic data and the power grid load prediction value.

[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: the system can collect the real-time operation data of the power grid load through the data collection module, ensuring the timeliness and accuracy of the data. The system preprocesses the collected operation data of the power grid load through the data preprocessing module to obtain the actual operation data of the power grid load, which can improve the quality and reliability of the data. The system performs data reconstruction analysis on the actual operation data of the power grid load through the data analysis module to extract the power grid load characteristic data, which can more intuitively reflect the change law and trend of the power grid load. The system constructs a power grid load prediction model by using the deep learning model through the deep learning module, and optimizes the model according to the power grid load characteristic data, which can adapt to the complexity of the power grid load and improve the prediction accuracy, contributing to the stable operation of the power grid. The system predicts the power grid load according to the optimized power grid load prediction model through the load prediction module to obtain the power grid load prediction value, which helps to formulate and adjust the power grid regulation strategy in advance and further improve the reliability of the power grid. Description of the Drawings

[0016] Figure 1 It is a schematic structural diagram of the power grid load prediction system based on deep learning in this embodiment; Figure 2 It is a schematic flow diagram of the power grid load prediction method based on deep learning in this embodiment. Detailed Embodiments

[0017] In order to make the purpose and advantages of the present invention clearer, the present invention will be further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present invention and do not limit the protection scope of the present invention.

[0019] It should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be directly connected or indirectly connected through an intermediate medium, and it can be the communication inside two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations.

[0020] Please refer to Figure 1 As shown, it is a schematic structural diagram of the power grid load forecasting system based on deep learning in this embodiment. The system includes: A data acquisition module for real-time acquisition of power grid load operation data; A data preprocessing module for preprocessing the power grid load operation data to obtain actual power grid load operation data. The data preprocessing module is connected to the data acquisition module; A data analysis module for performing data reconstruction analysis on the actual power grid load operation data to obtain power grid load characteristic data. The data analysis module is connected to the data preprocessing module; A deep learning module for constructing a power grid load forecasting model according to a deep learning model, and also for optimizing the construction process of the power grid load forecasting model according to the power grid load characteristic data to obtain an optimized power grid load forecasting model. The deep learning module is connected to the data analysis module; A load forecasting module for forecasting the power grid load according to the optimized power grid load forecasting model to obtain a power grid load forecasting value, and also for outputting power grid regulation means according to the power grid load forecasting value, and further for optimizing the power grid regulation means according to the power grid load characteristic data and the power grid load forecasting value. The load forecasting module is connected to the deep learning module and the data analysis module.

[0021] Specifically, the system is set in the power grid load forecasting terminal. The system can collect the power grid load operation data in real time through the data acquisition module, ensuring the timeliness and accuracy of the data. The system preprocesses the collected power grid load operation data through the data preprocessing module to obtain the actual power grid load operation data, which can improve the quality and reliability of the data. The system conducts data reconstruction analysis on the actual power grid load operation data through the data analysis module to extract the power grid load characteristic data, which can more intuitively reflect the change rules and trends of the power grid load. The system uses a deep learning model to construct a power grid load forecasting model through the deep learning module and optimizes the model according to the power grid load characteristic data, which can adapt to the complexity of the power grid load and improve the forecasting accuracy, contributing to the stable operation of the power grid. The system forecasts the power grid load according to the optimized power grid load forecasting model through the load forecasting module to obtain the power grid load forecasting value, which helps to formulate and adjust the power grid regulation strategy in advance and further improve the reliability of the power grid.

[0022] Specifically, the data acquisition module collects the power grid load operation data in real time through data acquisition devices.

[0023] Specifically, the grid load operation data refers to the data set collected in real time by data acquisition devices, including load data and environmental data. The load data refers to the power data consumed during the operation of the grid, including active power, reactive power, power factor, maximum load, minimum load, and average load. The active power refers to the power actually used for work in the grid. The reactive power refers to the power required to establish and maintain the magnetic field in the grid. The power factor refers to the ratio between the active power and the apparent power. The maximum load refers to the maximum power load borne by the grid within a preset time period. In this embodiment, the specific value of the preset time period is not limited. For example, the preset time period can be set to one month. The minimum load refers to the minimum power load borne by the grid within a preset time period. The average load refers to the average value of the power loads borne by the grid within a preset time period. The environmental data refers to the environmental data corresponding to the operation of the grid, including temperature, wind speed, and sunshine duration. The temperature refers to the actual air temperature of the environment where the grid operates. The wind speed refers to the wind force magnitude in the area where the grid operates. The sunshine duration refers to the time of daily sunlight exposure in the area where the grid operates. The data acquisition device refers to the device used to collect the grid load operation data in real time, including Beidou satellites and the grid equipment management system. The grid equipment management system refers to the system used to interface with the grid load prediction system. The data acquisition module collects the active power in real time through Beidou satellites. The data acquisition module samples the voltage and current signals to obtain analog signals, converts the analog signals into digital signals, calculates the phase difference between the voltage and the current, and calculates the reactive power based on the phase difference between the voltage and the current. The data acquisition module calculates the power factor through the active power and the apparent power. The data acquisition module collects the maximum load, minimum load, and average load in real time through the grid equipment management system.

[0024] Specifically, through the data acquisition module, the grid load operation data can be effectively obtained, and the efficiency and accuracy of collecting the grid load operation data can be improved.

[0025] Specifically, the data preprocessing module detects the outliers of the grid load operation data according to the data anomaly detection algorithm, obtains the corresponding data outliers, performs outlier processing on the data outliers according to the data outlier processing scheme corresponding to the grid load operation data, obtains the corresponding outlier processing results, and performs data conversion on the outlier processing results according to the data conversion scheme to obtain the corresponding actual grid load operation data.

[0026] Specifically, the data anomaly detection algorithm refers to an algorithm for detecting whether there are outliers in data. In this embodiment, the specific implementation manner of the data anomaly detection algorithm is not limited. For example, it can be set to detect outliers in the grid load operation data through the Isolation Forest algorithm. The data outlier refers to the value obtained by detecting outliers in the grid load operation data according to the data anomaly detection algorithm. The data outlier processing scheme refers to a strategy for processing the data outlier. In this embodiment, the specific implementation manner of the data outlier processing scheme is not limited. For example, it can be set to process the data outlier through the 3-sigma method. The outlier processing result refers to the result obtained by processing the data outlier according to the data outlier processing scheme corresponding to the grid load operation data. The data conversion scheme refers to a strategy for converting data. In this embodiment, the specific implementation manner of the data conversion scheme is not limited. For example, it can be set to perform data conversion on the outlier processing result through the Z-score method. The actual grid load operation data refers to the data obtained by performing data conversion on the outlier processing result according to the data conversion scheme.

[0027] Specifically, through the data preprocessing module, outliers in the grid load operation data can be accurately identified, the data quality can be improved, and the usability and accuracy of the data can be enhanced.

[0028] Specifically, the data analysis module decomposes the actual grid load operation data according to the data decomposition method to obtain a data decomposition result, and performs a matching analysis on the data decomposition result and the preset data decomposition standard according to the data matching analysis method to obtain an actual matching degree DT. Then, the actual matching degree DT is compared with the preset matching degree DT0, and according to the comparison result, a data reconstruction analysis is performed on the actual grid load operation data to obtain grid load characteristic data, where: When DT≥DT0, the data analysis module performs a data reconstruction analysis on the actual grid load operation data according to the grid load data reconstruction strategy to obtain grid load characteristic data; When DT<DT0, the data analysis module does not perform a data reconstruction analysis on the actual grid load operation data.

[0029] Specifically, the data decomposition method refers to a method of decomposing the actual power grid load operation data into subsequences with different frequency characteristics. In this embodiment, the specific implementation of the data decomposition method is not limited. For example, it can be set to decompose the actual power grid load operation data through the variational mode decomposition method. The data decomposition result refers to the result obtained by decomposing the actual power grid load operation data according to the data decomposition method. The data matching analysis method refers to a method of performing matching analysis on the data decomposition result and a preset data decomposition standard. In this embodiment, the specific implementation of the data matching analysis method is not limited. For example, it can be set to perform matching analysis on the data decomposition result and the preset data decomposition standard through the cosine similarity method. The preset data decomposition standard refers to a preset standard for comparison with the data decomposition result, such as the average value of historical load data. The actual matching degree DT refers to the value obtained by performing matching analysis on the data decomposition result and the preset data decomposition standard according to the data matching analysis method. The preset matching degree DT0 refers to a preset value for comparison with the actual matching degree DT, such as 0.8. In this embodiment, the specific implementation of data reconstruction analysis is not limited. For example, it can be set to perform data reconstruction analysis through the discrete wavelet transform method. The power grid load characteristic data refers to the result obtained by performing data reconstruction analysis on the actual power grid load operation data according to the comparison result.

[0030] Specifically, by decomposing and performing data reconstruction analysis on the actual power grid load operation data through the data analysis module, the power grid load data can be optimized, and the quality and usability of the data can be improved.

[0031] Specifically, the deep learning module performs a comprehensive evaluation on the power grid load characteristic data according to the data comprehensive evaluation method, obtains the actual comprehensive evaluation value PG, compares the actual comprehensive evaluation value PG with the preset comprehensive evaluation value PG0, judges the data reconstruction analysis situation of the power grid load characteristic data according to the comparison result, and constructs a power grid load prediction model according to the judgment result, where: When PG < PG0, the deep learning module determines that the data reconstruction analysis situation of the power grid load characteristic data is a non-compliant situation, and the deep learning module does not construct a power grid load prediction model; When PG ≥ PG0, the deep learning module determines that the data reconstruction analysis situation of the power grid load characteristic data is a compliant situation, and constructs a power grid load prediction model according to the preset power grid load prediction data set; When constructing the power grid load prediction model according to the preset power grid load prediction data set, the deep learning module sets the deep learning model as a convolutional neural network model according to the type and characteristics of the preset power grid load prediction data set, and sets the convolutional layer in the convolutional neural network model to process the preset power grid load prediction data set to obtain a convolutional layer data output feature map; The deep learning module sets the pooling layer in the convolutional neural network model to process the preset power grid load prediction data set according to the type and characteristics of the preset power grid load prediction data set to obtain pooling layer data features; The deep learning module connects the convolutional layer data output feature map and the pooling layer data features to the fully connected layer of the convolutional neural network model for data feature extraction to obtain a fully connected layer data feature result, and identifies the fully connected layer data feature result according to the power grid load fault sequence recognition model to obtain power grid load fault sequence feature data, and sets the power grid load fault sequence feature data as the first time series, and sets the actual power grid load data set corresponding to the power grid load fault sequence feature data as the second time series and analyzes the first time series and the second time series according to the dynamic time warping algorithm to obtain a time series similarity set , where represents the time series similarity at the k-th moment; The deep learning module according to the second time series and the time series similarity set calculates the power grid load prediction value , set , where represents the i-th time series similarity, represents the i-th actual power grid load data, and k represents the total number of time series set when performing the dynamic time warping algorithm; The deep learning module adds the power grid load prediction value to the preset power grid load prediction data set to obtain a new power grid load prediction data set, and divides the new power grid load prediction data set into a 70% analysis training set, a 20% analysis verification set, and a 10% analysis test set, inputs the analysis training set into the convolutional neural network model to train the convolutional neural network model, inputs the analysis verification set into the trained convolutional neural network model to iteratively optimize the trained convolutional neural network model, and inputs the analysis test set into the iteratively optimized convolutional neural network model to analyze and test the iteratively optimized convolutional neural network model, and outputs the convolutional neural network model that meets the preset analysis test accuracy as the power grid load prediction model.

[0032] Specifically, the data comprehensive evaluation method refers to a method for comprehensively evaluating the power grid load characteristic data. In this embodiment, the specific implementation manner of the data comprehensive evaluation method is not limited. For example, it can be set to comprehensively evaluate the power grid load characteristic data through the analytic hierarchy process. The actual comprehensive evaluation value PG refers to the value obtained by comprehensively evaluating the power grid load characteristic data according to the data comprehensive evaluation method. The preset comprehensive evaluation value PG0 refers to a preset value for comparison with the actual comprehensive evaluation value PG, such as 0.85, the data reconstruction analysis situation of the grid load characteristic data refers to the result state obtained after performing data reconstruction analysis on the grid load characteristic data. The deep learning model refers to a machine learning algorithm that automatically learns data features through a multi-layer non-linear transformation structure and makes predictions. The preset grid load prediction data set refers to a data set preset for training a convolutional neural network model, in which the storage form is historical grid load prediction feature image data - grid load prediction values. The grid load prediction model refers to a convolutional neural network model that has been trained and whose training results meet the standards. The convolutional neural network model refers to a deep learning model used to process image data with a grid topology structure, and automatically extracts features and classifies them through convolutional operations. The convolutional kernel size refers to the size of the convolutional kernel in the convolutional layer of the convolutional neural network model. The number of convolutional kernels refers to the number of convolutional kernels in the convolutional layer of the convolutional neural network model. The convolutional layer data output feature map refers to the output data obtained after the convolutional layer in the convolutional neural network model performs a convolutional operation on the input data. The pooling layer size refers to the size of the window in the pooling operation of the convolutional neural network model. The number of pooling layers refers to the number of pooling layers in the convolutional neural network model. The pooling layer data feature refers to the output data obtained after the fully connected layer in the convolutional neural network model performs a linear transformation and non-linear activation on the input data. The grid load fault sequence recognition model refers to a deep learning model that takes the historical grid load fault sequence feature map as input and the grid load fault sequence feature data as output. In this embodiment, the construction method of the grid load fault sequence recognition model is not limited, and those skilled in the art can freely set it according to the actual situation, as long as the requirement of recognizing the data feature results of the fully connected layer is met. For example, the grid load fault sequence recognition model can be set as a convolutional neural network model. The first time series refers to the sequence formed by arranging the grid load fault sequence feature data in chronological order. The actual grid load data set refers to the set of actual grid load data corresponding to the grid load fault sequence feature data for recording. The second time series refers to the sequence formed by arranging the actual grid load data set in chronological order. The dynamic time warping algorithm refers to an algorithm used to measure the similarity between two time series. The time series similarity set refers to the set of similarities obtained by analyzing the first time series and the second time series according to the dynamic time warping algorithm. The new grid load prediction data set refers to the set obtained by adding the grid load prediction value. to the preset grid load prediction data set. The preset analysis and test accuracy rate refers to a preset value used to analyze and test the iteratively optimized convolutional neural network model, such as 0.9.

[0033] Specifically, the deep learning module uses a convolutional neural network model to effectively extract the features of power grid load data, and improves the prediction ability of the power grid load prediction model through the dynamic time warping algorithm and time series similarity analysis.

[0034] Specifically, the deep learning module processes the power grid load feature data according to the outlier optimization detection method to obtain the outlier optimization detection result, obtains the outlier frequency f1 according to the outlier optimization detection result, compares the outlier frequency f1 with the preset outlier frequency f0, and optimizes the construction process of the power grid load prediction model according to the comparison result to obtain the optimized power grid load prediction model, where: When f1 < f0, the deep learning model module does not optimize the construction process of the power grid load prediction model; When f1 ≥ f0, the deep learning model module repairs the outlier optimization detection result according to the data repair method to obtain the data repair result, and adds the data repair result to the new power grid load prediction data set to optimize the construction process of the power grid load prediction model to obtain the optimized power grid load prediction model.

[0035] Specifically, the outlier optimization detection method refers to the method used to identify and process outliers in the power grid load feature data. In this embodiment, the specific implementation manner of the outlier optimization detection method is not limited. For example, it can be set to process the power grid load feature data through the isolation forest algorithm. The outlier optimization detection result refers to the result obtained by processing the power grid load feature data according to the outlier optimization detection method. The outlier frequency f1 refers to the frequency of outliers in the power grid load feature data. The preset outlier frequency f0 refers to the preset value compared with the outlier frequency f1, such as 5%. The construction process of the power grid load prediction model refers to the process of training the convolutional neural network model according to the preset power grid load prediction data set. The optimized power grid load prediction model refers to the model obtained by adding the data repair result to the new power grid load prediction data set to optimize the construction process of the power grid load prediction model. The data repair method refers to the method used to repair outliers in the power grid load feature data. In this embodiment, the specific implementation manner of the data repair method is not limited. For example, it can be set to repair the outlier optimization detection result through the linear interpolation method. The new power grid load prediction data set refers to the data set obtained by adding the data repair result to the new power grid load prediction data set.

[0036] Specifically, by optimizing the construction process of the power grid load prediction model according to the comparison result between the outlier frequency f1 and the preset outlier frequency f0 through the deep learning model module, the accuracy of the power grid load prediction model for power grid load prediction can be improved.

[0037] Specifically, the load prediction module obtains the recall Recall and precision Precison according to the optimized power grid load prediction model, calculates the model evaluation index value FS according to the recall Recall and precision Precison, and sets , compares the model evaluation index value FS with the preset model evaluation index value FS0, judges the training situation of the optimized power grid load prediction model according to the comparison result, and predicts the power grid load according to the judgment result to obtain the power grid load prediction value, where: When FS≥FS0, the load prediction module determines that the training situation of the optimized power grid load prediction model is a qualified situation, and inputs the power grid load characteristic data into the optimized power grid load prediction model determined to be in a qualified situation for prediction to obtain the power grid load prediction value; When FS<FS0, the load prediction module determines that the training situation of the optimized power grid load prediction model is an unqualified situation, and the load prediction module does not predict the power grid load.

[0038] Specifically, the recall Recall refers to the proportion of samples that are actually positive classes and are correctly predicted as positive classes by the model. The precision Precison refers to the proportion of samples that are actually positive classes among all the samples predicted as positive classes by the model. The preset model evaluation index value FS0 refers to the preset value compared with the model evaluation index value FS, such as 90%. The training situation of the optimized power grid load prediction model refers to the performance of the power grid load prediction model on the training set after outlier processing and data repair. The power grid load refers to the power demand borne by the power grid within a preset time period. The specific value of the preset time period is not limited in this embodiment. For example, the preset time period can be set to one month. The power grid load prediction value refers to the value obtained by inputting the power grid load characteristic data into the optimized power grid load prediction model determined to be in a qualified situation for prediction.

[0039] Specifically, by judging the training situation of the optimized power grid load prediction model through the load prediction module and predicting the power grid load according to the judgment result, the accuracy and reliability of the prediction result are improved.

[0040] Specifically, the load prediction module compares the power grid load prediction value Compare them, judge the power grid load prediction situation according to the comparison result, and output the power grid control means according to the judgment result, where: When > The load prediction module determines that the power grid load prediction situation is an overloaded state of power grid prediction, and outputs the power grid control means according to the overloaded state of power grid prediction; When ≤ The load prediction module determines that the power grid load prediction situation is a normal state of power grid prediction, and the load prediction module does not output the power grid control means.

[0041] Specifically, the preset power grid load prediction value refers to the preset value compared with the power grid load prediction value , for example, 500MW. The power grid load prediction situation refers to the prediction situation of the future load state of the power grid after comparing the power grid load prediction value with the preset power grid load prediction value . The power grid control means refers to the power grid control means preset in the overloaded state of power grid prediction. In this embodiment, the specific implementation manner of the power grid control means is not limited. For example, it can be set to determine an abnormal signal according to the overloaded state of power grid prediction, adjust the generator output according to the abnormal signal, and send an alarm signal to the power grid monitoring system to notify the staff to perform maintenance and repair in time.

[0042] Specifically, by comparing the power grid load prediction value with the preset power grid load prediction value through the load prediction module, the abnormal situation of the power grid load can be found in time, providing timely and accurate early warning information for power grid dispatching and operation.

[0043] Specifically, the load prediction module obtains the i-th actual value of the power grid load according to the power grid load characteristic data , and calculates the mean square error MSE according to the i-th power grid load prediction value and the i-th actual value of the power grid load . Set , where n represents the number of samples, and compare the mean square error MSE with the preset mean square error MSE0, judge the mean square error calculation situation according to the comparison result, and optimize the power grid control means according to the judgment result, where: When MSE≤MSE0, the load prediction module determines that the mean square error calculation situation is normal, and the deep learning module does not optimize the power grid control means; When MSE > MSE0, the load forecasting module determines that the mean square error calculation situation is an abnormal situation. The load forecasting module calculates the environmental index HJ based on the temperature W1, wind speed S1, and sunshine duration R1 in the grid load characteristic data, and sets , where 1 / W0 represents the weight coefficient for adjusting the temperature W1, 1 / S0 represents the weight coefficient for adjusting the wind speed S1, 1 / R0 represents the weight coefficient for adjusting the sunshine duration R1. Set 5°C ≤ W0 ≤ 40°C, 25 m / s ≤ S0 ≤ 40 m / s, 1000 < R0 < 4000, and compare the environmental index HJ with the preset environmental index HJ0. Judge the grid environmental factor situation based on the comparison result, and optimize the grid regulation means according to the judgment result, where: If HJ ≤ HJ0, the load forecasting module determines that the grid environmental factor situation is normal, and the deep learning module does not optimize the grid regulation means; If HJ > HJ0, the load forecasting module determines that the grid environmental factor situation is abnormal. The load forecasting module optimizes the mean square error MSE according to the environmental factor optimization parameter b1 to obtain the optimized mean square error , and based on the optimized mean square error optimize the grid regulation means to the grid regulation optimization means, where is set. .

[0044] Specifically, the mean square error calculation situation refers to the situation of calculating the mean square error between the grid load forecast value and the actual grid load value. The preset mean square error MSE0 is the preset value compared with the mean square error MSE, such as 4%. The grid environmental factor situation refers to evaluating the degree to which the grid load is affected by environmental factors by calculating the environmental index. The preset environmental index HJ0 is the preset value compared with the environmental index HJ, such as 0.5. The environmental factor optimization parameter b1 is the parameter used to optimize the mean square error MSE. The optimized mean square error is the value obtained by optimizing the mean square error MSE according to the environmental factor optimization parameter b1. The grid regulation optimization means refers to the means of optimizing the grid regulation work. The specific implementation method of the grid regulation optimization means in this embodiment is not limited. For example, it can be set to perform regulation through the grid regulation means combined with safety measures. The safety measures refer to the measures taken to ensure the safety of grid staff and the stable operation of grid equipment, such as wearing protective equipment.

[0045] Specifically, the load prediction module optimizes the parameter b1 by adjusting environmental factors to further improve the prediction accuracy of the prediction model. According to the optimized mean square error and the power grid environmental factors, the power grid regulation means are optimized to improve the stability of the power grid.

[0046] Please refer to Figure 2 as shown, which is a schematic flowchart of the power grid load prediction method based on deep learning in this embodiment. The method includes: Step S1: Real-time collect the power grid load operation data and preprocess the power grid load operation data to obtain the actual power grid load operation data; Step S2: Perform data reconstruction analysis on the actual power grid load operation data to obtain the power grid load characteristic data; Step S3: Construct the power grid load prediction model according to the deep learning model, and optimize the construction process of the power grid load prediction model according to the power grid load characteristic data to obtain the optimized power grid load prediction model; Step S4: Predict the power grid load according to the optimized power grid load prediction model to obtain the power grid load prediction value, output the power grid regulation means according to the power grid load prediction value, and optimize the power grid regulation means according to the power grid load characteristic data and the power grid load prediction value.

[0047] So far, the technical solution of the present invention has been described in combination with the preferred embodiments shown in the drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or replacements to the relevant technical features, and the technical solutions after these changes or replacements will all fall within the protection scope of the present invention.

Claims

1. A power grid load forecasting system based on deep learning, characterized in that, The system includes: A data acquisition module for real-time acquisition of power grid load operation data; A data preprocessing module for preprocessing the power grid load operation data to obtain actual power grid load operation data; A data analysis module for performing data reconstruction analysis on the actual power grid load operation data to obtain power grid load characteristic data; A deep learning module for constructing a power grid load prediction model according to a deep learning model, and also for optimizing the construction process of the power grid load prediction model according to the power grid load characteristic data to obtain an optimized power grid load prediction model; A load prediction module for predicting the power grid load according to the optimized power grid load prediction model to obtain a power grid load prediction value, and also for outputting power grid regulation means according to the power grid load prediction value, and further for optimizing the power grid regulation means according to the power grid load characteristic data and the power grid load prediction value.

2. The power grid load prediction system based on deep learning according to claim 1, wherein The data preprocessing module detects outliers in the power grid load operation data according to a data anomaly detection algorithm to obtain corresponding data outliers, and performs outlier processing on the data outliers according to a data outlier processing scheme corresponding to the power grid load operation data to obtain corresponding outlier processing results, and performs data conversion on the outlier processing results according to a data conversion scheme to obtain corresponding actual power grid load operation data.

3. The power grid load prediction system based on deep learning according to claim 1, wherein The data analysis module decomposes the actual power grid load operation data according to a data decomposition method to obtain a data decomposition result, and performs matching analysis on the data decomposition result and a preset data decomposition standard according to a data matching analysis method to obtain an actual matching degree DT, and compares the actual matching degree DT with a preset matching degree DT0, and performs data reconstruction analysis on the actual power grid load operation data according to the comparison result to obtain power grid load characteristic data, where: When DT≥DT0, the data analysis module performs data reconstruction analysis on the actual power grid load operation data according to a power grid load data reconstruction strategy to obtain power grid load characteristic data; When DT<DT0, the data analysis module does not perform data reconstruction analysis on the actual power grid load operation data.

4. The power grid load prediction system based on deep learning according to claim 1, wherein The deep learning module comprehensively evaluates the power grid load characteristic data according to a data comprehensive evaluation method to obtain an actual comprehensive evaluation value PG, and compares the actual comprehensive evaluation value PG with a preset comprehensive evaluation value PG0, and judges the data reconstruction analysis situation of the power grid load characteristic data according to the comparison result, and constructs a power grid load prediction model according to the judgment result, where: When PG<PG0, the deep learning module determines that the data reconstruction analysis situation of the power grid load characteristic data is a non-compliant situation, and the deep learning module does not construct a power grid load prediction model; When PG≥PG0, the deep learning module determines that the data reconstruction analysis situation of the power grid load characteristic data is a compliant situation, and constructs a power grid load prediction model according to a preset power grid load prediction data set.

5. The power grid load prediction system based on deep learning according to claim 4, characterized in that When constructing the power grid load prediction model based on the preset power grid load prediction data set, the deep learning module sets the deep learning model as a convolutional neural network model according to the type and characteristics of the preset power grid load prediction data set, and sets the convolutional layer in the convolutional neural network model to process the preset power grid load prediction data set to obtain a convolutional layer data output feature map; The deep learning module sets the pooling layer in the convolutional neural network model to process the preset power grid load prediction data set according to the type and characteristics of the preset power grid load prediction data set to obtain pooling layer data features; The deep learning module connects the output feature map of the convolutional layer data and the data features of the pooling layer to the fully connected layer of the convolutional neural network model for data feature extraction, obtains the data feature result of the fully connected layer, and identifies the data feature result of the fully connected layer according to the power grid load fault sequence recognition model to obtain the power grid load fault sequence feature data. The power grid load fault sequence feature data is set as the first time series, and the actual power grid load data set corresponding to the power grid load fault sequence feature data is set as the second time series , and analyzes the first time series and the second time series according to the dynamic time warping algorithm to obtain a time series similarity set , where represents the time series similarity at the k-th moment; The deep learning module calculates based on the second time series and the time series similarity set the predicted value of the power grid load and sets , where represents the similarity of the i-th time series, represents the i-th actual power grid load data, and k represents the total number of time series set when performing the dynamic time warping algorithm; The deep learning module adds the power grid load prediction value to a preset power grid load prediction data set to obtain a new power grid load prediction data set, and divides the new power grid load prediction data set into a 70% analysis training set, a 20% analysis validation set, and a 10% analysis test set. The analysis training set is input into a convolutional neural network model to train the convolutional neural network model, the analysis validation set is input into the trained convolutional neural network model to iteratively optimize the trained convolutional neural network model, and the analysis test set is input into the iteratively optimized convolutional neural network model to analyze and test the iteratively optimized convolutional neural network model. The convolutional neural network model that meets the preset analysis test accuracy rate in the analysis test is output as the power grid load prediction model.

6. The power grid load prediction system based on deep learning according to claim 5, characterized in that, The deep learning module processes the power grid load characteristic data according to the outlier optimization detection method to obtain an outlier optimization detection result, obtains an outlier frequency f1 according to the outlier optimization detection result, compares the outlier frequency f1 with a preset outlier frequency f0, and optimizes the construction process of the power grid load prediction model according to the comparison result to obtain an optimized power grid load prediction model, where: When f1 < f0, the deep learning model module does not optimize the construction process of the power grid load prediction model; When f1 ≥ f0, the deep learning model module repairs the outlier optimization detection result according to the data repair method to obtain a data repair result, and adds the data repair result to a new power grid load prediction data set to optimize the construction process of the power grid load prediction model to obtain an optimized power grid load prediction model.

7. The power grid load prediction system based on deep learning according to claim 1, wherein The load forecasting module obtains the recall Recall and precision Precison according to the optimized power grid load forecasting model, calculates the model evaluation index value FS based on the recall Recall and precision Precison, and sets , compares the model evaluation index value FS with the preset model evaluation index value FS0, judges the training situation of the optimized power grid load forecasting model according to the comparison result, and forecasts the power grid load according to the judgment result to obtain the power grid load forecasting value, where: When FS ≥ FS0, the load prediction module determines that the training condition of the optimized power grid load prediction model is a qualified condition, and inputs the power grid load characteristic data into the optimized power grid load prediction model determined to be in a qualified condition for prediction to obtain a power grid load prediction value; When FS < FS0, the load prediction module determines that the training condition of the optimized power grid load prediction model is an unqualified condition, and the load prediction module does not predict the power grid load.

8. The power grid load prediction system based on deep learning according to claim 7, characterized in that, The load forecasting module compares the power grid load forecast value with the preset power grid load forecast value to determine the power grid load forecasting situation based on the comparison result, and output the power grid regulation means according to the judgment result, where: When > the load prediction module determines that the power grid load prediction situation is an overloaded state of power grid prediction, and outputs power grid control means according to the overloaded state of power grid prediction; When ≤ , the load forecasting module determines that the power grid load forecasting situation is in a normal power grid forecasting state, and the load forecasting module does not output power grid control means.

9. The power grid load prediction system based on deep learning according to claim 8, wherein The load forecasting module obtains the actual value of the i-th grid load according to the grid load characteristic data , and according to the predicted value of the i-th grid load and the actual value of the i-th grid load calculate the mean square error MSE, and set , where n represents the number of samples, and compare the mean square error MSE with the preset mean square error MSE0, judge the calculation situation of the mean square error according to the comparison result, and optimize the grid regulation means according to the judgment result, where: When MSE ≤ MSE0, the load prediction module determines that the mean square error calculation condition is a normal condition, and the deep learning module does not optimize the power grid regulation means; When MSE > MSE0, the load forecasting module determines that the mean square error calculation situation is an abnormal situation. The load forecasting module calculates the environmental index HJ based on the temperature W1, wind speed S1, and sunshine duration R1 in the grid load characteristic data, and sets , where 1 / W0 represents the weight coefficient for adjusting the temperature W1, 1 / S0 represents the weight coefficient for adjusting the wind speed S1, 1 / R0 represents the weight coefficient for adjusting the sunshine duration R1. It is set that 5°C ≤ W0 ≤ 40°C, 25 m / s ≤ S0 ≤ 40 m / s, 1000 < R0 < 4000, and the environmental index HJ is compared with the preset environmental index HJ0. The grid environmental factor situation is judged according to the comparison result, and the grid regulation means are optimized according to the judgment result, where: If HJ ≤ HJ0, the load prediction module determines that the power grid environmental factor condition is a normal condition, and the deep learning module does not optimize the power grid regulation means; When HJ > HJ0, the load forecasting module determines that the power grid environmental factor situation is an abnormal situation. The load forecasting module optimizes the mean square error MSE according to the environmental factor optimization parameter b1 to obtain the optimized mean square error , and based on the optimized mean square error optimizes the power grid regulation means to the optimized power grid regulation means, where , .

10. A method applied to the deep learning-based power grid load forecasting system as described in claims 1-9, characterized in that, The method includes: Step S1, collecting the power grid load operation data in real time, and preprocessing the power grid load operation data to obtain actual power grid load operation data; Step S2, performing data reconstruction analysis on the actual power grid load operation data to obtain power grid load characteristic data; Step S3, constructing a power grid load prediction model according to a deep learning model, and optimizing the construction process of the power grid load prediction model according to the power grid load characteristic data to obtain an optimized power grid load prediction model; Step S4, predicting the power grid load according to the optimized power grid load prediction model to obtain a power grid load prediction value, outputting the power grid regulation means according to the power grid load prediction value, and optimizing the power grid regulation means according to the power grid load characteristic data and the power grid load prediction value.

Images